fix(rpc): common:getNodes returns uuid-keyed map instead of array

Each value still contains its own uuid field. Single-uuid query keeps returning a single object.

.github/actions/build-frontend/action.yml

@@ -0,0 +1,55 @@
+name: Build frontend theme
+description: Clone komari-web, build the default theme, and optionally upload it as an artifact.
+
+inputs:
+  node-version:
+    description: Node.js version used to build komari-web.
+    default: "23"
+  repository:
+    description: komari-web repository URL.
+    default: https://github.com/komari-monitor/komari-web
+  artifact-name:
+    description: Optional artifact name for the packed frontend theme.
+    default: ""
+
+runs:
+  using: composite
+  steps:
+    - name: Setup Node.js
+      uses: actions/setup-node@v4
+      with:
+        node-version: ${{ inputs.node-version }}
+
+    - name: Clone and build frontend
+      shell: bash
+      run: |
+        set -euxo pipefail
+        rm -rf komari-web
+        git clone --depth=1 "${{ inputs.repository }}" komari-web
+        cd komari-web
+        npm install
+        npm run build
+        cd ..
+
+        mkdir -p web/public/defaultTheme/dist
+        rm -rf web/public/defaultTheme/dist/*
+        cp -r komari-web/dist/* web/public/defaultTheme/dist/
+        cp -f komari-web/komari-theme.json web/public/defaultTheme/
+        if [ -f komari-web/preview.png ]; then cp -f komari-web/preview.png web/public/defaultTheme/; fi
+        if [ -f komari-web/perview.png ]; then cp -f komari-web/perview.png web/public/defaultTheme/; fi
+        if [ -f web/public/defaultTheme/preview.png ] && [ ! -f web/public/defaultTheme/perview.png ]; then cp -f web/public/defaultTheme/preview.png web/public/defaultTheme/perview.png; fi
+        if [ -f web/public/defaultTheme/perview.png ] && [ ! -f web/public/defaultTheme/preview.png ]; then cp -f web/public/defaultTheme/perview.png web/public/defaultTheme/preview.png; fi
+
+    - name: Pack frontend artifact
+      if: ${{ inputs.artifact-name != '' }}
+      shell: bash
+      run: |
+        set -euxo pipefail
+        tar -czf "$RUNNER_TEMP/${{ inputs.artifact-name }}.tar.gz" -C web/public/defaultTheme .
+
+    - name: Upload frontend artifact
+      if: ${{ inputs.artifact-name != '' }}
+      uses: actions/upload-artifact@v4
+      with:
+        name: ${{ inputs.artifact-name }}
+        path: ${{ runner.temp }}/${{ inputs.artifact-name }}.tar.gz

.github/actions/restore-frontend/action.yml

@@ -0,0 +1,24 @@
+name: Restore frontend theme
+description: Download and unpack a previously built frontend theme artifact.
+
+inputs:
+  artifact-name:
+    description: Frontend theme artifact name.
+    default: komari-frontend
+
+runs:
+  using: composite
+  steps:
+    - name: Download frontend artifact
+      uses: actions/download-artifact@v4
+      with:
+        name: ${{ inputs.artifact-name }}
+        path: ${{ runner.temp }}/frontend-artifact
+
+    - name: Extract frontend artifact
+      shell: bash
+      run: |
+        set -euxo pipefail
+        mkdir -p web/public/defaultTheme
+        rm -rf web/public/defaultTheme/*
+        tar -xzf "$RUNNER_TEMP/frontend-artifact/${{ inputs.artifact-name }}.tar.gz" -C web/public/defaultTheme

.github/actions/setup-zig/action.yml

@@ -0,0 +1,31 @@
+name: Setup Zig
+description: Restore or install the Zig toolchain used for CGO cross-compilation.
+
+inputs:
+  zig-version:
+    description: Zig version to install.
+    default: 0.14.1
+
+runs:
+  using: composite
+  steps:
+    - name: Restore Zig cache
+      id: zig-cache
+      uses: actions/cache@v4
+      with:
+        path: ~/.cache/komari-zig/zig-x86_64-linux-${{ inputs.zig-version }}
+        key: zig-${{ runner.os }}-x86_64-${{ inputs.zig-version }}
+
+    - name: Download Zig
+      if: ${{ steps.zig-cache.outputs.cache-hit != 'true' }}
+      shell: bash
+      run: |
+        set -euxo pipefail
+        mkdir -p "$HOME/.cache/komari-zig"
+        curl -fsSL "https://ziglang.org/download/${{ inputs.zig-version }}/zig-x86_64-linux-${{ inputs.zig-version }}.tar.xz" -o "$RUNNER_TEMP/zig.tar.xz"
+        tar -C "$HOME/.cache/komari-zig" -Jxf "$RUNNER_TEMP/zig.tar.xz"
+
+    - name: Add Zig to PATH
+      shell: bash
+      run: |
+        echo "$HOME/.cache/komari-zig/zig-x86_64-linux-${{ inputs.zig-version }}" >> "$GITHUB_PATH"

.github/workflows/build.yml

@@ -1,16 +1,26 @@
-name: Build Binaries on Main Push and PR
+name: Build Binaries on PR
 
 on:
-  push:
-    branches:
-      - main
   pull_request:
     branches:
       - main
   workflow_dispatch:
 
 jobs:
+  build-frontend:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Build frontend artifact
+        uses: ./.github/actions/build-frontend
+        with:
+          artifact-name: komari-frontend
+
   build-binaries:
+    needs: build-frontend
     runs-on: ubuntu-latest
 
     strategy:
@@ -58,27 +68,8 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
-      - name: Setup Node.js
-        uses: actions/setup-node@v4
-        with:
-          node-version: "23"
-
-      - name: Clone and build frontend
-        run: |
-          git clone https://github.com/komari-monitor/komari-web komari-web
-          cd komari-web
-          npm install
-          npm run build
-          cd ..
-          mkdir -p web/public/defaultTheme/dist
-          rm -rf web/public/defaultTheme/dist/*
-          cp -r komari-web/dist/* web/public/defaultTheme/dist/
-          cp -f komari-web/komari-theme.json web/public/defaultTheme/
-          if [ -f komari-web/preview.png ]; then cp -f komari-web/preview.png web/public/defaultTheme/; fi
-          if [ -f komari-web/perview.png ]; then cp -f komari-web/perview.png web/public/defaultTheme/; fi
-          # Compatibility: ensure both filenames exist if only one is present
-          if [ -f web/public/defaultTheme/preview.png ] && [ ! -f web/public/defaultTheme/perview.png ]; then cp -f web/public/defaultTheme/preview.png web/public/defaultTheme/perview.png; fi
-          if [ -f web/public/defaultTheme/perview.png ] && [ ! -f web/public/defaultTheme/preview.png ]; then cp -f web/public/defaultTheme/perview.png web/public/defaultTheme/preview.png; fi
+      - name: Restore frontend artifact
+        uses: ./.github/actions/restore-frontend
 
       - name: Set up Go
         uses: actions/setup-go@v5
@@ -86,12 +77,7 @@ jobs:
           go-version: "1.23"
 
       - name: Install Zig cross-compiler toolchain
-        run: |
-          sudo apt-get update # Update apt package lists
-          sudo apt-get install -y build-essential curl
-          wget https://ziglang.org/download/0.14.1/zig-x86_64-linux-0.14.1.tar.xz
-          sudo tar -C /usr/local -Jxf zig-x86_64-linux-0.14.1.tar.xz
-          echo "PATH=$PATH:/usr/local/zig-x86_64-linux-0.14.1" >> $GITHUB_ENV
+        uses: ./.github/actions/setup-zig
 
       - name: Build binary
         env:

.github/workflows/development.yml

@@ -20,21 +20,7 @@ jobs:
       #    echo "<html><body><h1>This is komari development server.</h1></body></html>" > web/public/defaultTheme/dist/index.html
 
       - name: Clone and build frontend
-        run: |
-          git clone https://github.com/komari-monitor/komari-web komari-web
-          cd komari-web
-          npm install
-          npm run build
-          cd ..
-          mkdir -p web/public/defaultTheme/dist
-          rm -rf web/public/defaultTheme/dist/*
-          cp -r komari-web/dist/* web/public/defaultTheme/dist/
-          cp -f komari-web/komari-theme.json web/public/defaultTheme/
-          if [ -f komari-web/preview.png ]; then cp -f komari-web/preview.png web/public/defaultTheme/; fi
-          if [ -f komari-web/perview.png ]; then cp -f komari-web/perview.png web/public/defaultTheme/; fi
-          # Compatibility: ensure both filenames exist if only one is present
-          if [ -f web/public/defaultTheme/preview.png ] && [ ! -f web/public/defaultTheme/perview.png ]; then cp -f web/public/defaultTheme/preview.png web/public/defaultTheme/perview.png; fi
-          if [ -f web/public/defaultTheme/perview.png ] && [ ! -f web/public/defaultTheme/preview.png ]; then cp -f web/public/defaultTheme/perview.png web/public/defaultTheme/preview.png; fi
+        uses: ./.github/actions/build-frontend
 
       - name: Set up Go
         uses: actions/setup-go@v5

.github/workflows/docker-publish.yml

@@ -11,37 +11,28 @@ env:
   IMAGE_NAME: ${{ github.repository }}
 
 jobs:
-  build-and-push-multi-arch:
+  build-frontend:
     runs-on: ubuntu-latest
-    permissions:
-      contents: read
-      packages: write
 
     steps:
-      - name: Checkout repository
+      - name: Checkout code
         uses: actions/checkout@v4
 
-      - name: Setup Node.js
-        uses: actions/setup-node@v4
+      - name: Build frontend artifact
+        uses: ./.github/actions/build-frontend
         with:
-          node-version: "23"
+          artifact-name: komari-frontend
 
-      - name: Clone and build frontend
-        run: |
-          git clone https://github.com/komari-monitor/komari-web komari-web
-          cd komari-web
-          npm install
-          npm run build
-          cd ..
-          mkdir -p web/public/defaultTheme/dist
-          rm -rf web/public/defaultTheme/dist/*
-          cp -r komari-web/dist/* web/public/defaultTheme/dist/
-          cp -f komari-web/komari-theme.json web/public/defaultTheme/
-          if [ -f komari-web/preview.png ]; then cp -f komari-web/preview.png web/public/defaultTheme/; fi
-          if [ -f komari-web/perview.png ]; then cp -f komari-web/perview.png web/public/defaultTheme/; fi
-          # Compatibility: ensure both filenames exist if only one is present
-          if [ -f web/public/defaultTheme/preview.png ] && [ ! -f web/public/defaultTheme/perview.png ]; then cp -f web/public/defaultTheme/preview.png web/public/defaultTheme/perview.png; fi
-          if [ -f web/public/defaultTheme/perview.png ] && [ ! -f web/public/defaultTheme/preview.png ]; then cp -f web/public/defaultTheme/perview.png web/public/defaultTheme/preview.png; fi
+  build-docker-binaries:
+    needs: build-frontend
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Restore frontend artifact
+        uses: ./.github/actions/restore-frontend
 
       - name: Set up Go
         uses: actions/setup-go@v5
@@ -49,10 +40,7 @@ jobs:
           go-version: "1.23"
 
       - name: Install Zig
-        run: |
-          wget https://ziglang.org/download/0.14.1/zig-x86_64-linux-0.14.1.tar.xz
-          sudo tar -C /usr/local -Jxf zig-x86_64-linux-0.14.1.tar.xz
-          echo "PATH=$PATH:/usr/local/zig-x86_64-linux-0.14.1" >> $GITHUB_ENV
+        uses: ./.github/actions/setup-zig
 
       - name: Build static binaries for multi-arch
         env:
@@ -70,6 +58,29 @@ jobs:
           echo "Building for linux/arm64..."
           GOARCH=arm64 CC="zig cc -target aarch64-linux-musl" go build -trimpath -ldflags="$LDFLAGS" -o komari-linux-arm64
 
+      - name: Upload Docker binaries
+        uses: actions/upload-artifact@v4
+        with:
+          name: komari-docker-binaries
+          path: komari-linux-*
+
+  build-and-push-multi-arch:
+    needs: build-docker-binaries
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Download Docker binaries
+        uses: actions/download-artifact@v4
+        with:
+          name: komari-docker-binaries
+          path: .
+
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v3
 

.github/workflows/generate-release-notes.yml

@@ -0,0 +1,608 @@
+name: Generate Release Notes
+
+on:
+  release:
+    types: [published]
+  workflow_dispatch:
+    inputs:
+      tag:
+        description: "Release tag to regenerate notes for. Leave empty to use the latest release."
+        required: false
+        type: string
+      base_ref:
+        description: "Base ref or SHA for the release comparison. Snapshot builds pass the previous main SHA."
+        required: false
+        type: string
+
+permissions:
+  contents: write
+
+jobs:
+  update-release-notes:
+    runs-on: ubuntu-latest
+    if: ${{ github.event_name != 'release' || !startsWith(github.event.release.tag_name, 'Snapshot-') }}
+
+    steps:
+      - name: Generate and apply release notes
+        uses: actions/github-script@v9
+        env:
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+          OPENAI_BASE_URL: ${{ secrets.OPENAI_BASE_URL }}
+          OPENAI_MODEL: ${{ vars.OPENAI_MODEL }}
+        with:
+          script: |
+            const { owner, repo } = context.repo;
+            const webRepo = { owner: "komari-monitor", repo: "komari-web" };
+
+            const AI_CONTEXT_CHAR_LIMIT = 120000;
+            const AI_MAX_COMMITS = 160;
+            const AI_MAX_CHANGED_FILES = 120;
+            const AI_MAX_PULL_REQUESTS = 80;
+            const AI_MAX_PATCH_CHARS_PER_FILE = 2200;
+            const OPENAI_TIMEOUT_MS = 120000;
+
+            const clean = (value) => (typeof value === "string" ? value.trim() : "");
+            const isSnapshotTag = (tag) => /^Snapshot-/i.test(clean(tag));
+            const releaseTimestamp = (release) => Date.parse(release.published_at || release.created_at || "");
+            const isStableRelease = (release) => !release.draft && !release.prerelease && !isSnapshotTag(release.tag_name);
+
+            function limitText(value, maxLength) {
+              const text = clean(value);
+              if (text.length <= maxLength) {
+                return text;
+              }
+
+              return `${text.slice(0, maxLength)}\n...[truncated ${text.length - maxLength} chars]`;
+            }
+
+            function appendWithinLimit(parts, value, maxLength) {
+              const text = clean(value);
+              if (!text) {
+                return false;
+              }
+
+              const currentLength = parts.reduce((total, part) => total + part.length + 2, 0);
+              if (currentLength + text.length + 2 <= maxLength) {
+                parts.push(text);
+                return true;
+              }
+
+              const remaining = maxLength - currentLength - 64;
+              if (remaining > 500) {
+                parts.push(`${text.slice(0, remaining)}\n...[context truncated]`);
+              }
+
+              return false;
+            }
+
+            async function resolveRelease() {
+              const requestedTag = clean(context.payload.release?.tag_name)
+                || clean(context.payload.client_payload?.tag)
+                || clean(context.payload.inputs?.tag);
+
+              if (requestedTag) {
+                const release = context.payload.release?.tag_name === requestedTag
+                  ? context.payload.release
+                  : (await github.rest.repos.getReleaseByTag({ owner, repo, tag: requestedTag })).data;
+                return { tag: requestedTag, release };
+              }
+
+              const release = (await github.rest.repos.getLatestRelease({ owner, repo })).data;
+              return { tag: release.tag_name, release };
+            }
+
+            async function getPreviousReleaseTag(tag, release) {
+              const releases = await github.paginate(github.rest.repos.listReleases, {
+                owner,
+                repo,
+                per_page: 100,
+              });
+
+              const currentTime = releaseTimestamp(release);
+              const currentIsSnapshot = isSnapshotTag(tag);
+              const currentIsStable = isStableRelease(release);
+              const candidates = releases
+                .filter((item) => !item.draft && item.tag_name !== tag)
+                .filter((item) => {
+                  if (!Number.isFinite(currentTime)) {
+                    return true;
+                  }
+
+                  const itemTime = releaseTimestamp(item);
+                  return Number.isFinite(itemTime) && itemTime < currentTime;
+                })
+                .filter((item) => {
+                  if (currentIsSnapshot || currentIsStable) {
+                    return isStableRelease(item);
+                  }
+
+                  return item.prerelease && !isSnapshotTag(item.tag_name);
+                })
+                .sort((left, right) => releaseTimestamp(right) - releaseTimestamp(left));
+
+              return candidates[0]?.tag_name || "";
+            }
+
+            async function resolveComparisonBase(tag, release) {
+              const requestedBaseRef = clean(context.payload.client_payload?.base_ref)
+                || clean(context.payload.inputs?.base_ref);
+
+              if (requestedBaseRef) {
+                return {
+                  ref: requestedBaseRef,
+                  previousTag: "",
+                  source: "requested base ref",
+                };
+              }
+
+              const previousTag = await getPreviousReleaseTag(tag, release);
+              if (previousTag) {
+                return {
+                  ref: previousTag,
+                  previousTag,
+                  source: isSnapshotTag(tag) ? "previous stable release" : "previous release in the same channel",
+                };
+              }
+
+              if (isSnapshotTag(tag)) {
+                try {
+                  const commit = (await github.rest.repos.getCommit({
+                    owner,
+                    repo,
+                    ref: release.target_commitish || tag,
+                  })).data;
+                  const parentSha = commit.parents?.[0]?.sha || "";
+                  if (parentSha) {
+                    return {
+                      ref: parentSha,
+                      previousTag: "",
+                      source: "snapshot target parent commit",
+                    };
+                  }
+                } catch (error) {
+                  core.warning(`Failed to resolve parent commit for ${tag}: ${error.message}`);
+                }
+              }
+
+              return {
+                ref: "",
+                previousTag: "",
+                source: "none",
+              };
+            }
+
+            async function getGeneratedNotes(tag, release, previousTag) {
+              const params = {
+                owner,
+                repo,
+                tag_name: tag,
+                target_commitish: release.target_commitish || undefined,
+              };
+
+              if (previousTag) {
+                params.previous_tag_name = previousTag;
+              }
+
+              const generated = await github.request(
+                "POST /repos/{owner}/{repo}/releases/generate-notes",
+                params,
+              );
+
+              return clean(generated.data.body);
+            }
+
+            async function getComparison(baseRef, tag) {
+              if (!baseRef) {
+                return { commits: [], files: [], htmlUrl: "" };
+              }
+
+              const commits = [];
+              let files = [];
+              let htmlUrl = "";
+
+              for (let page = 1; page <= 20; page += 1) {
+                const response = await github.rest.repos.compareCommitsWithBasehead({
+                  owner,
+                  repo,
+                  basehead: `${baseRef}...${tag}`,
+                  per_page: 100,
+                  page,
+                });
+
+                if (page === 1) {
+                  files = response.data.files || [];
+                  htmlUrl = response.data.html_url || "";
+                }
+
+                commits.push(...response.data.commits);
+
+                if (!response.headers.link?.includes('rel="next"')) {
+                  break;
+                }
+              }
+
+              return { commits, files, htmlUrl };
+            }
+
+            function renderCommitLine(commit) {
+              const subject = clean(commit.commit.message.split("\n")[0]) || "(no subject)";
+              const shortSha = commit.sha.slice(0, 7);
+              const author = commit.author?.login
+                ? ` @${commit.author.login}`
+                : (commit.commit.author?.name ? ` ${commit.commit.author.name}` : "");
+
+              return `- ${subject} ([${shortSha}](${commit.html_url}))${author}`;
+            }
+
+            function buildWhatsChangedSection(baseRef, tag, comparison) {
+              if (!baseRef) {
+                return "## What's Changed\n\nNo comparison base was found for this release.";
+              }
+
+              if (comparison.commits.length === 0) {
+                return `## What's Changed\n\nNo commits found between \`${baseRef}\` and \`${tag}\`.`;
+              }
+
+              const lines = ["## What's Changed", "", ...comparison.commits.map(renderCommitLine)];
+              if (comparison.htmlUrl) {
+                lines.push("", `**Full Changelog**: ${comparison.htmlUrl}`);
+              }
+
+              return lines.join("\n");
+            }
+
+            function appendCommitsToGeneratedNotes(generatedNotes, baseRef, tag, comparison) {
+              const notes = clean(generatedNotes);
+              if (!notes || comparison.commits.length === 0) {
+                return notes || buildWhatsChangedSection(baseRef, tag, comparison);
+              }
+
+              const lines = notes.split("\n");
+              const fullChangelogIndex = lines.findIndex((line) => /^\*\*Full Changelog\*\*:/i.test(line.trim()));
+              const nextTopLevelSectionIndex = lines.findIndex((line, index) => (
+                index > 0
+                && /^##\s+/.test(line)
+                && !/^##\s+What's Changed\b/i.test(line)
+              ));
+              const insertionIndex = nextTopLevelSectionIndex >= 0
+                ? nextTopLevelSectionIndex
+                : (fullChangelogIndex >= 0 ? fullChangelogIndex : lines.length);
+              const commitSection = [
+                "### Commits",
+                "",
+                ...comparison.commits.map(renderCommitLine),
+                "",
+              ];
+
+              lines.splice(insertionIndex, 0, ...commitSection);
+              return clean(lines.join("\n"));
+            }
+
+            async function getPullRequestContext(commits) {
+              const pullRequests = new Map();
+
+              for (const commit of commits.slice(0, AI_MAX_COMMITS)) {
+                try {
+                  const response = await github.rest.repos.listPullRequestsAssociatedWithCommit({
+                    owner,
+                    repo,
+                    commit_sha: commit.sha,
+                  });
+
+                  for (const pullRequest of response.data || []) {
+                    pullRequests.set(pullRequest.html_url, pullRequest);
+                  }
+                } catch (error) {
+                  core.warning(`Failed to load pull requests for ${commit.sha.slice(0, 7)}: ${error.message}`);
+                }
+
+                if (pullRequests.size >= AI_MAX_PULL_REQUESTS) {
+                  break;
+                }
+              }
+
+              return [...pullRequests.values()].slice(0, AI_MAX_PULL_REQUESTS);
+            }
+
+            function buildChangedFilesContext(files) {
+              if (!files.length) {
+                return "No changed file metadata was returned by GitHub.";
+              }
+
+              return files.slice(0, AI_MAX_CHANGED_FILES).map((file) => {
+                const lines = [
+                  `File: ${file.filename}`,
+                  `Status: ${file.status}; additions: ${file.additions}; deletions: ${file.deletions}; changes: ${file.changes}`,
+                ];
+
+                if (file.patch) {
+                  lines.push("Patch excerpt:");
+                  lines.push(limitText(file.patch, AI_MAX_PATCH_CHARS_PER_FILE));
+                }
+
+                return lines.join("\n");
+              }).join("\n\n");
+            }
+
+            function buildCommitContext(commits) {
+              if (!commits.length) {
+                return "No commits were returned by GitHub.";
+              }
+
+              return commits.slice(0, AI_MAX_COMMITS).map((commit) => {
+                const shortSha = commit.sha.slice(0, 7);
+                const author = commit.author?.login || commit.commit.author?.name || "unknown";
+                const message = limitText(commit.commit.message, 1800);
+                return [
+                  `Commit: ${shortSha}`,
+                  `Author: ${author}`,
+                  `URL: ${commit.html_url}`,
+                  "Message:",
+                  message,
+                ].join("\n");
+              }).join("\n\n");
+            }
+
+            function buildPullRequestContext(pullRequests) {
+              if (!pullRequests.length) {
+                return "No associated pull requests were found for the compared commits.";
+              }
+
+              return pullRequests.map((pullRequest) => {
+                const labels = (pullRequest.labels || []).map((label) => label.name).filter(Boolean).join(", ") || "none";
+                return [
+                  `PR: #${pullRequest.number} ${pullRequest.title}`,
+                  `URL: ${pullRequest.html_url}`,
+                  `Author: ${pullRequest.user?.login || "unknown"}`,
+                  `Labels: ${labels}`,
+                  "Body:",
+                  limitText(pullRequest.body, 1800) || "(empty)",
+                ].join("\n");
+              }).join("\n\n");
+            }
+
+            function buildAiContext({ tag, release, comparisonBase, whatsChangedSection, comparison, pullRequests, webReleaseSection }) {
+              const sections = [];
+
+              appendWithinLimit(sections, [
+                `Repository: ${owner}/${repo}`,
+                `Release tag: ${tag}`,
+                `Release title: ${release.name || tag}`,
+                `Prerelease: ${release.prerelease ? "yes" : "no"}`,
+                `Snapshot tag: ${isSnapshotTag(tag) ? "yes" : "no"}`,
+                `Target commitish: ${release.target_commitish || "(unknown)"}`,
+                `Comparison base: ${comparisonBase.ref || "(none)"}`,
+                `Comparison base source: ${comparisonBase.source}`,
+                `Compare URL: ${comparison.htmlUrl || "(unavailable)"}`,
+              ].join("\n"), AI_CONTEXT_CHAR_LIMIT);
+
+              appendWithinLimit(sections, `What's Changed section:\n${whatsChangedSection}`, AI_CONTEXT_CHAR_LIMIT);
+              appendWithinLimit(sections, `Changed files and patch excerpts:\n${buildChangedFilesContext(comparison.files)}`, AI_CONTEXT_CHAR_LIMIT);
+              appendWithinLimit(sections, `Associated pull requests:\n${buildPullRequestContext(pullRequests)}`, AI_CONTEXT_CHAR_LIMIT);
+              appendWithinLimit(sections, `Commit details:\n${buildCommitContext(comparison.commits)}`, AI_CONTEXT_CHAR_LIMIT);
+
+              if (webReleaseSection) {
+                appendWithinLimit(sections, `Komari Web release notes for the same tag:\n${webReleaseSection}`, AI_CONTEXT_CHAR_LIMIT);
+              }
+
+              return sections.join("\n\n---\n\n");
+            }
+
+            function buildChatCompletionsUrl(baseUrl) {
+              const trimmed = clean(baseUrl).replace(/\/+$/, "");
+              if (!trimmed) {
+                return "";
+              }
+
+              if (trimmed.endsWith("/chat/completions")) {
+                return trimmed;
+              }
+
+              if (trimmed.endsWith("/v1")) {
+                return `${trimmed}/chat/completions`;
+              }
+
+              return `${trimmed}/v1/chat/completions`;
+            }
+
+            function normalizeReleaseSummary(markdown) {
+              const summary = clean(markdown);
+              if (!summary) {
+                return "";
+              }
+
+              const withoutAiHeading = clean(summary
+                .replace(/^#{1,6}\s+AI Summary\b\s*/im, "")
+                .replace(/^\*\*AI Summary\*\*\s*/im, ""));
+              const whatsChangedIndex = withoutAiHeading.search(/^##\s+What's Changed\b/im);
+              if (whatsChangedIndex >= 0) {
+                return clean(withoutAiHeading.slice(0, whatsChangedIndex));
+              }
+
+              return withoutAiHeading;
+            }
+
+            async function callOpenAiChatCompletion(url, apiKey, model, messages) {
+              const controller = new AbortController();
+              const timeout = setTimeout(() => controller.abort(), OPENAI_TIMEOUT_MS);
+
+              try {
+                const response = await fetch(url, {
+                  method: "POST",
+                  headers: {
+                    Authorization: `Bearer ${apiKey}`,
+                    "Content-Type": "application/json",
+                  },
+                  body: JSON.stringify({ model, messages, temperature: 0 }),
+                  signal: controller.signal,
+                });
+
+                const responseText = await response.text();
+                let responseJson = {};
+                try {
+                  responseJson = responseText ? JSON.parse(responseText) : {};
+                } catch {
+                  responseJson = {};
+                }
+
+                if (!response.ok) {
+                  const message = clean(responseJson.error?.message) || limitText(responseText, 500);
+                  throw new Error(`OpenAI request failed with ${response.status}: ${message}`);
+                }
+
+                return clean(responseJson.choices?.[0]?.message?.content);
+              } finally {
+                clearTimeout(timeout);
+              }
+            }
+
+            async function buildReleaseSummarySection({ tag, release, comparisonBase, whatsChangedSection, comparison, webReleaseSection }) {
+              const apiKey = clean(process.env.OPENAI_API_KEY);
+              const baseUrl = clean(process.env.OPENAI_BASE_URL);
+              const model = clean(process.env.OPENAI_MODEL);
+
+              if (!apiKey || !baseUrl || !model) {
+                core.warning("OPENAI_API_KEY, OPENAI_BASE_URL, or OPENAI_MODEL is missing. Skipping generated release summary.");
+                return "";
+              }
+
+              const pullRequests = await getPullRequestContext(comparison.commits);
+              const aiContext = buildAiContext({
+                tag,
+                release,
+                comparisonBase,
+                whatsChangedSection,
+                comparison,
+                pullRequests,
+                webReleaseSection,
+              });
+
+              const messages = [
+                {
+                  role: "system",
+                  content: [
+                    "You write accurate release notes for the Komari server project.",
+                    "Use only the supplied repository context.",
+                    "Include both Komari server changes and Komari Web frontend changes when frontend release notes are supplied.",
+                    "Prioritize user-visible behavior, compatibility or migration notes, operations/deployment changes, security fixes, and important bug fixes.",
+                    "Do not invent features or implementation details that are not supported by the context.",
+                    "Write concise Markdown in Simplified Chinese.",
+                  ].join(" "),
+                },
+                {
+                  role: "user",
+                  content: [
+                    `Create the release summary section for release ${tag}.`,
+                    comparisonBase.ref
+                      ? `Summarize only changes in the comparison ${comparisonBase.ref}...${tag}.`
+                      : "No comparison base was found, so summarize only the supplied release context.",
+                    "Return only Markdown for the summary section that appears before '## What's Changed'.",
+                    "Do not use the heading 'AI Summary'. Do not add a 'What's Changed' section.",
+                    "Use these top-level headings, in this order: '## 主要更新内容', '## Bug修复', and '## 其他'.",
+                    "Omit '## Bug修复' when there are no concrete bug fixes. Keep bullets short and specific.",
+                    "",
+                    aiContext,
+                  ].join("\n"),
+                },
+              ];
+
+              try {
+                const url = buildChatCompletionsUrl(baseUrl);
+                const summary = await callOpenAiChatCompletion(url, apiKey, model, messages);
+                return normalizeReleaseSummary(summary);
+              } catch (error) {
+                core.warning(`Generated release summary failed: ${error.message}`);
+                return "";
+              }
+            }
+
+            async function buildWebReleaseSection(tag) {
+              const demoteMarkdownHeadings = (markdown) => markdown.replace(/^(#{1,5})(\s+)/gm, "#$1$2");
+
+              try {
+                const webRelease = (await github.rest.repos.getReleaseByTag({
+                  owner: webRepo.owner,
+                  repo: webRepo.repo,
+                  tag,
+                })).data;
+
+                const webBody = clean(webRelease.body);
+                if (!webBody) {
+                  return `## Komari Web Release Notes\n\nKomari Web release [${tag}](${webRelease.html_url}) exists, but its release notes are empty.`;
+                }
+
+                return [
+                  "## Komari Web Release Notes",
+                  "",
+                  `Source: [komari-web ${tag}](${webRelease.html_url})`,
+                  "",
+                  demoteMarkdownHeadings(webBody),
+                ].join("\n");
+              } catch (error) {
+                if (error.status === 404) {
+                  core.info(`Komari Web release ${tag} does not exist yet. Skipping web release notes for now.`);
+                  return "";
+                }
+
+                throw error;
+              }
+            }
+
+            const { tag, release } = await resolveRelease();
+            if (!tag) {
+              core.setFailed("No release tag was provided or found.");
+              return;
+            }
+
+            const comparisonBase = await resolveComparisonBase(tag, release);
+            core.info(comparisonBase.ref
+              ? `Comparison base for ${tag}: ${comparisonBase.ref} (${comparisonBase.source}).`
+              : `No comparison base found for ${tag}.`);
+
+            const comparison = await getComparison(comparisonBase.ref, tag);
+            const whatsChangedSection = comparisonBase.previousTag
+              ? appendCommitsToGeneratedNotes(
+                await getGeneratedNotes(tag, release, comparisonBase.previousTag),
+                comparisonBase.ref,
+                tag,
+                comparison,
+              )
+              : (!comparisonBase.ref && !isSnapshotTag(tag)
+                ? await getGeneratedNotes(tag, release, "")
+                : buildWhatsChangedSection(comparisonBase.ref, tag, comparison));
+
+            if (!whatsChangedSection) {
+              core.setFailed(`Failed to build What's Changed section for ${tag}.`);
+              return;
+            }
+
+            const webReleaseSection = await buildWebReleaseSection(tag);
+            const releaseSummarySection = await buildReleaseSummarySection({
+              tag,
+              release,
+              comparisonBase,
+              whatsChangedSection,
+              comparison,
+              webReleaseSection,
+            });
+
+            const body = [
+              releaseSummarySection,
+              whatsChangedSection,
+            ].filter(Boolean).join("\n\n");
+
+            if (clean(release.body) === clean(body)) {
+              core.info(`Release notes for ${tag} are already up to date.`);
+            } else {
+              await github.rest.repos.updateRelease({
+                owner,
+                repo,
+                release_id: release.id,
+                body,
+              });
+            }
+
+            await core.summary
+              .addHeading("Release Notes")
+              .addRaw(`Updated release notes for ${tag}.`)
+              .write();

.github/workflows/release-docker.yml

@@ -1,9 +1,12 @@
-name: Publish Docker Image
+name: Publish Docker Image Manually
 
 on:
-  release:
-    types: [published]
   workflow_dispatch:
+    inputs:
+      version:
+        description: Docker image tag and embedded version.
+        required: true
+        type: string
 
 permissions:
   contents: read
@@ -14,7 +17,61 @@ env:
   IMAGE_NAME: ${{ github.repository }}
 
 jobs:
+  build-frontend:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Build frontend artifact
+        uses: ./.github/actions/build-frontend
+        with:
+          artifact-name: komari-frontend
+
+  build-docker-binaries:
+    needs: build-frontend
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Restore frontend artifact
+        uses: ./.github/actions/restore-frontend
+
+      - name: Set up Go
+        uses: actions/setup-go@v5
+        with:
+          go-version: "1.23"
+
+      - name: Install Zig
+        uses: ./.github/actions/setup-zig
+
+      - name: Build static binaries for multi-arch
+        env:
+          GOOS: linux
+          CGO_ENABLED: 1
+          GIN_MODE: release
+          VERSION: ${{ github.event.inputs.version }}
+        run: |
+          VERSION_HASH="${{ github.sha }}"
+          LDFLAGS="-s -w -X github.com/komari-monitor/komari/utils.CurrentVersion=${VERSION} -X github.com/komari-monitor/komari/utils.VersionHash=${VERSION_HASH}"
+
+          echo "Building for linux/amd64..."
+          GOARCH=amd64 CC="zig cc -target x86_64-linux-musl" go build -trimpath -ldflags="$LDFLAGS" -o komari-linux-amd64
+
+          echo "Building for linux/arm64..."
+          GOARCH=arm64 CC="zig cc -target aarch64-linux-musl" go build -trimpath -ldflags="$LDFLAGS" -o komari-linux-arm64
+
+      - name: Upload Docker binaries
+        uses: actions/upload-artifact@v4
+        with:
+          name: komari-docker-binaries
+          path: komari-linux-*
+
   build-and-push-multi-arch:
+    needs: build-docker-binaries
     runs-on: ubuntu-latest
     permissions:
       contents: read
@@ -24,54 +81,11 @@ jobs:
       - name: Checkout repository
         uses: actions/checkout@v4
 
-      - name: Setup Node.js
-        uses: actions/setup-node@v4
+      - name: Download Docker binaries
+        uses: actions/download-artifact@v4
         with:
-          node-version: "23"
-
-      - name: Clone and build frontend
-        run: |
-          git clone https://github.com/komari-monitor/komari-web komari-web
-          cd komari-web
-          npm install
-          npm run build
-          cd ..
-          mkdir -p web/public/defaultTheme/dist
-          rm -rf web/public/defaultTheme/dist/*
-          cp -r komari-web/dist/* web/public/defaultTheme/dist/
-          cp -f komari-web/komari-theme.json web/public/defaultTheme/
-          if [ -f komari-web/preview.png ]; then cp -f komari-web/preview.png web/public/defaultTheme/; fi
-          if [ -f komari-web/perview.png ]; then cp -f komari-web/perview.png web/public/defaultTheme/; fi
-          # Compatibility: ensure both filenames exist if only one is present
-          if [ -f web/public/defaultTheme/preview.png ] && [ ! -f web/public/defaultTheme/perview.png ]; then cp -f web/public/defaultTheme/preview.png web/public/defaultTheme/perview.png; fi
-          if [ -f web/public/defaultTheme/perview.png ] && [ ! -f web/public/defaultTheme/preview.png ]; then cp -f web/public/defaultTheme/perview.png web/public/defaultTheme/preview.png; fi
-
-      - name: Set up Go
-        uses: actions/setup-go@v5
-        with:
-          go-version: "1.23"
-
-      - name: Install Zig
-        run: |
-          wget https://ziglang.org/download/0.14.1/zig-x86_64-linux-0.14.1.tar.xz
-          sudo tar -C /usr/local -Jxf zig-x86_64-linux-0.14.1.tar.xz
-          echo "PATH=$PATH:/usr/local/zig-x86_64-linux-0.14.1" >> $GITHUB_ENV
-
-      - name: Build static binaries for multi-arch
-        env:
-          GOOS: linux
-          CGO_ENABLED: 1
-          GIN_MODE: release
-        run: |
-          VERSION="${{ github.event.release.tag_name }}"
-          VERSION_HASH="${{ github.sha }}"
-          LDFLAGS="-s -w -X github.com/komari-monitor/komari/utils.CurrentVersion=${VERSION} -X github.com/komari-monitor/komari/utils.VersionHash=${VERSION_HASH}"
-
-          echo "Building for linux/amd64..."
-          GOARCH=amd64 CC="zig cc -target x86_64-linux-musl" go build -trimpath -ldflags="$LDFLAGS" -o komari-linux-amd64
-
-          echo "Building for linux/arm64..."
-          GOARCH=arm64 CC="zig cc -target aarch64-linux-musl" go build -trimpath -ldflags="$LDFLAGS" -o komari-linux-arm64
+          name: komari-docker-binaries
+          path: .
 
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v3
@@ -92,7 +106,7 @@ jobs:
         with:
           images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
           tags: |
-            type=raw,value=${{ github.event.release.tag_name }}
+            type=raw,value=${{ github.event.inputs.version }}
             type=raw,value=latest
 
       - name: Build and push Docker image

.github/workflows/release.yml

@@ -13,7 +13,20 @@ env:
   IMAGE_NAME: ${{ github.repository }}
 
 jobs:
+  build-frontend:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Build frontend artifact
+        uses: ./.github/actions/build-frontend
+        with:
+          artifact-name: komari-frontend
+
   build-and-attach-binaries:
+    needs: build-frontend
     runs-on: ubuntu-latest
 
     strategy:
@@ -61,27 +74,8 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
-      - name: Setup Node.js
-        uses: actions/setup-node@v4
-        with:
-          node-version: "23"
-
-      - name: Clone and build frontend
-        run: |
-          git clone https://github.com/komari-monitor/komari-web komari-web
-          cd komari-web
-          npm install
-          npm run build
-          cd ..
-          mkdir -p web/public/defaultTheme/dist
-          rm -rf web/public/defaultTheme/dist/*
-          cp -r komari-web/dist/* web/public/defaultTheme/dist/
-          cp -f komari-web/komari-theme.json web/public/defaultTheme/
-          if [ -f komari-web/preview.png ]; then cp -f komari-web/preview.png web/public/defaultTheme/; fi
-          if [ -f komari-web/perview.png ]; then cp -f komari-web/perview.png web/public/defaultTheme/; fi
-          # Compatibility: ensure both filenames exist if only one is present
-          if [ -f web/public/defaultTheme/preview.png ] && [ ! -f web/public/defaultTheme/perview.png ]; then cp -f web/public/defaultTheme/preview.png web/public/defaultTheme/perview.png; fi
-          if [ -f web/public/defaultTheme/perview.png ] && [ ! -f web/public/defaultTheme/preview.png ]; then cp -f web/public/defaultTheme/perview.png web/public/defaultTheme/preview.png; fi
+      - name: Restore frontend artifact
+        uses: ./.github/actions/restore-frontend
 
       - name: Set up Go
         uses: actions/setup-go@v5
@@ -89,12 +83,7 @@ jobs:
           go-version: "1.23"
 
       - name: Install Zig cross-compiler toolchain
-        run: |
-          sudo apt-get update # Update apt package lists
-          sudo apt-get install -y build-essential curl
-          wget https://ziglang.org/download/0.14.1/zig-x86_64-linux-0.14.1.tar.xz
-          sudo tar -C /usr/local -Jxf zig-x86_64-linux-0.14.1.tar.xz
-          echo "PATH=$PATH:/usr/local/zig-x86_64-linux-0.14.1" >> $GITHUB_ENV
+        uses: ./.github/actions/setup-zig
 
       - name: Build binary
         env:
@@ -112,6 +101,12 @@ jobs:
           VERSION_HASH="${{ github.sha }}" # Use commit SHA for hash
           go build -trimpath -ldflags="-s -w -X github.com/komari-monitor/komari/utils.CurrentVersion=${VERSION} -X github.com/komari-monitor/komari/utils.VersionHash=${VERSION_HASH}" -o $BINARY_NAME
 
+      - name: Upload binary as artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: komari-${{ matrix.goos }}-${{ matrix.goarch }}
+          path: komari-${{ matrix.goos }}-${{ matrix.goarch }}*
+
       - name: Upload binary to release
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
@@ -124,3 +119,54 @@ jobs:
           # Upload the binary as a release asset using GitHub CLI
           gh release upload ${{ github.event.release.tag_name }} $BINARY_NAME --repo ${{ github.repository }}
         shell: bash
+
+  build-and-push-docker:
+    needs: build-and-attach-binaries
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Download Docker binaries
+        uses: actions/download-artifact@v4
+        with:
+          pattern: komari-linux-*
+          path: .
+          merge-multiple: true
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to the Container registry
+        uses: docker/login-action@v3
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Extract metadata
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
+          tags: |
+            type=raw,value=${{ github.event.release.tag_name }}
+            type=raw,value=latest
+
+      - name: Build and push Docker image
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          push: true
+          platforms: linux/amd64,linux/arm64
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          no-cache: true
+          provenance: false

.github/workflows/snapshot.yml

@@ -0,0 +1,289 @@
+name: Publish Snapshot
+
+on:
+  workflow_dispatch:
+  push:
+    branches:
+      - main
+
+permissions:
+  contents: write
+  packages: write
+
+env:
+  REGISTRY: ghcr.io
+  IMAGE_NAME: ${{ github.repository }}
+
+concurrency:
+  group: snapshot-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  validate-ref:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 1
+
+      - name: Ensure snapshot target is main
+        run: |
+          if [ "${GITHUB_REF}" != "refs/heads/main" ]; then
+            echo "Snapshot builds must run from refs/heads/main. Current ref: ${GITHUB_REF}"
+            exit 1
+          fi
+
+          git fetch --no-tags --depth=1 origin main
+          MAIN_SHA="$(git rev-parse FETCH_HEAD)"
+
+          if [ "${GITHUB_SHA}" != "${MAIN_SHA}" ]; then
+            echo "Snapshot builds must use the current main commit."
+            echo "Workflow SHA: ${GITHUB_SHA}"
+            echo "Current main: ${MAIN_SHA}"
+            exit 1
+          fi
+
+  prepare:
+    needs: validate-ref
+    runs-on: ubuntu-latest
+    outputs:
+      version: ${{ steps.version.outputs.version }}
+    steps:
+      - name: Generate snapshot version
+        id: version
+        run: |
+          # Snapshot 版本号格式: Snapshot-yymmddhhMM (UTC+8)
+          VERSION="Snapshot-$(date -u -d '+8 hours' +%y%m%d%H%M)"
+          echo "version=${VERSION}" >> "$GITHUB_OUTPUT"
+          echo "Snapshot version: ${VERSION}"
+
+  build-frontend:
+    needs: validate-ref
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Build frontend artifact
+        uses: ./.github/actions/build-frontend
+        with:
+          artifact-name: komari-frontend
+
+  build-and-attach-binaries:
+    needs: [prepare, build-frontend]
+    runs-on: ubuntu-latest
+
+    strategy:
+      fail-fast: false
+      matrix:
+        goos: [windows, linux]
+        goarch: [amd64, arm64, 386, riscv64]
+        include:
+          # Windows
+          - goos: windows
+            goarch: amd64
+            zig_target_triple: x86_64-windows-gnu
+            binary_extension: .exe
+          - goos: windows
+            goarch: arm64
+            zig_target_triple: aarch64-windows-gnu
+            binary_extension: .exe
+          - goos: windows
+            goarch: 386
+            zig_target_triple: x86-windows-gnu
+            binary_extension: .exe
+
+          # Linux
+          - goos: linux
+            goarch: amd64
+            zig_target_triple: x86_64-linux-musl
+            binary_extension: ""
+          - goos: linux
+            goarch: arm64
+            zig_target_triple: aarch64-linux-musl
+            binary_extension: ""
+          - goos: linux
+            goarch: 386
+            zig_target_triple: x86-linux-musl
+            binary_extension: ""
+          - goos: linux
+            goarch: riscv64
+            zig_target_triple: riscv64-linux-musl
+            binary_extension: ""
+
+        exclude:
+          - goos: windows
+            goarch: riscv64
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Restore frontend artifact
+        uses: ./.github/actions/restore-frontend
+
+      - name: Set up Go
+        uses: actions/setup-go@v5
+        with:
+          go-version: "1.23"
+
+      - name: Install Zig cross-compiler toolchain
+        uses: ./.github/actions/setup-zig
+
+      - name: Build binary
+        env:
+          GOOS: ${{ matrix.goos }}
+          GOARCH: ${{ matrix.goarch }}
+          CGO_ENABLED: 1
+          CC: zig cc -target ${{ matrix.zig_target_triple }}
+          GIN_MODE: release
+        run: |
+          BINARY_NAME=komari-${{ matrix.goos }}-${{ matrix.goarch }}
+          if [ "${{ matrix.binary_extension }}" != "" ]; then
+            BINARY_NAME=${BINARY_NAME}${{ matrix.binary_extension }}
+          fi
+          VERSION="${{ needs.prepare.outputs.version }}"
+          VERSION_HASH="${{ github.sha }}" # Use commit SHA for hash
+          go build -trimpath -ldflags="-s -w -X github.com/komari-monitor/komari/utils.CurrentVersion=${VERSION} -X github.com/komari-monitor/komari/utils.VersionHash=${VERSION_HASH}" -o $BINARY_NAME
+
+      - name: Upload binary as artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: komari-${{ matrix.goos }}-${{ matrix.goarch }}
+          path: komari-${{ matrix.goos }}-${{ matrix.goarch }}*
+
+  publish-snapshot-release:
+    needs: [prepare, build-and-attach-binaries]
+    runs-on: ubuntu-latest
+    permissions:
+      actions: write
+      contents: write
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Ensure commit is still current main
+        run: |
+          git fetch --no-tags --depth=1 origin main
+          MAIN_SHA="$(git rev-parse FETCH_HEAD)"
+
+          if [ "${GITHUB_SHA}" != "${MAIN_SHA}" ]; then
+            echo "Snapshot publish skipped because a newer main commit exists."
+            echo "Workflow SHA: ${GITHUB_SHA}"
+            echo "Current main: ${MAIN_SHA}"
+            exit 1
+          fi
+
+      - name: Download all binaries
+        uses: actions/download-artifact@v4
+        with:
+          path: artifacts
+          merge-multiple: true
+
+      - name: Create snapshot prerelease and upload binaries
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          VERSION: ${{ needs.prepare.outputs.version }}
+        run: |
+          set -euxo pipefail
+          echo "Snapshot version: ${VERSION}"
+          ls -alR artifacts
+
+          # 如果同名 release/tag 已存在（例如重跑），先删除避免创建冲突
+          if gh release view "${VERSION}" --repo "${{ github.repository }}" >/dev/null 2>&1; then
+            echo "Release ${VERSION} 已存在，正在删除以重新创建..."
+            gh release delete "${VERSION}" --repo "${{ github.repository }}" --cleanup-tag --yes
+          fi
+
+          # 创建预发布并上传所有二进制文件
+          gh release create "${VERSION}" \
+            --repo "${{ github.repository }}" \
+            --target "${{ github.sha }}" \
+            --title "${VERSION}" \
+            --notes "Automated snapshot build (${VERSION}) from commit ${{ github.sha }}." \
+            --prerelease \
+            artifacts/komari-*
+
+          gh workflow run generate-release-notes.yml \
+            --repo "${{ github.repository }}" \
+            --ref main \
+            -f tag="${VERSION}"
+
+          gh release list \
+            --repo "${{ github.repository }}" \
+            --limit 1000 \
+            --json tagName,isPrerelease \
+            --jq '.[] | select(.isPrerelease and (.tagName | startswith("Snapshot-"))) | .tagName' |
+          while IFS= read -r tag; do
+            if [ -z "${tag}" ] || [ "${tag}" = "${VERSION}" ]; then
+              continue
+            fi
+
+            echo "Deleting old snapshot release ${tag}..."
+            gh release delete "${tag}" --repo "${{ github.repository }}" --cleanup-tag --yes
+          done
+        shell: bash
+
+  build-and-push-docker:
+    needs: [prepare, build-and-attach-binaries, publish-snapshot-release]
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Ensure commit is still current main
+        run: |
+          git fetch --no-tags --depth=1 origin main
+          MAIN_SHA="$(git rev-parse FETCH_HEAD)"
+
+          if [ "${GITHUB_SHA}" != "${MAIN_SHA}" ]; then
+            echo "Snapshot Docker publish skipped because a newer main commit exists."
+            echo "Workflow SHA: ${GITHUB_SHA}"
+            echo "Current main: ${MAIN_SHA}"
+            exit 1
+          fi
+
+      - name: Download Docker binaries
+        uses: actions/download-artifact@v4
+        with:
+          pattern: komari-linux-*
+          path: .
+          merge-multiple: true
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to the Container registry
+        uses: docker/login-action@v3
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Extract metadata
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
+          tags: |
+            type=raw,value=snapshot
+
+      - name: Build and push Docker image
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          push: true
+          platforms: linux/amd64,linux/arm64
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          no-cache: true
+          provenance: false

README.md

@@ -9,6 +9,11 @@
 
 Komari is a lightweight, self-hosted server monitoring tool designed to provide a simple and efficient solution for monitoring server performance. It supports viewing server status through a web interface and collects data through a lightweight agent.
 
+> [!WARNING]
+> Komari is a self-hosted monitoring and control program intended only for systems you own or are authorized to administer. Do not weaponize Komari, deploy it without consent, or use it for unauthorized access, persistence, command execution, or other abusive activity. For context on real-world abuse risks, see Huntress' analysis: [Komari C2 agent abuse](https://www.huntress.com/blog/komari-c2-agent-abuse).
+> Users are solely responsible for how they deploy and operate Komari. The developers do not accept responsibility for unauthorized or abusive use, or for any resulting consequences.
+> On Windows, when remote control is enabled, the client displays a Windows notification at each user login to remind the user that Komari is remote control software.
+
 [Documentation](https://komari-document.pages.dev/) | [文档(镜像站 By Geekertao)](https://www.komari.wiki) | [Telegram Group](https://t.me/komari_monitor)
 
 ## Features

cmd/server.go

@@ -134,7 +134,7 @@ func RunServer() {
 		}
 
 	})
-	r.Use(security.CorsMiddleware(conf.AllowCors, conf.CorsAllowedOrigins))
+	r.Use(security.CorsMiddleware(conf.CorsOriginCheckEnabled, conf.CorsAllowedOrigins))
 
 	r.Use(api.IdentityMiddleware())
 	r.Use(api.PrivateSiteMiddleware())

database/clients/report.go

@@ -11,6 +11,7 @@ import (
 	"github.com/komari-monitor/komari/database/dbcore"
 	"github.com/komari-monitor/komari/database/models"
 	v1 "github.com/komari-monitor/komari/protocol/v1"
+	"github.com/komari-monitor/komari/utils"
 
 	"gorm.io/gorm"
 )
@@ -22,17 +23,6 @@ func getReportSaveLock(clientUUID string) *sync.Mutex {
 	return lock.(*sync.Mutex)
 }
 
-func computeTrafficDelta(current, previous int64) int64 {
-	if previous < 0 {
-		return 0
-	}
-	if current >= previous {
-		return current - previous
-	}
-	// Counter reset or reboot: treat the new counter value as traffic since reset.
-	return current
-}
-
 func getLatestTrafficRecord(tx *gorm.DB, clientUUID string) (*models.Record, error) {
 	var record models.Record
 	err := tx.Where("client = ?", clientUUID).Order("time DESC").First(&record).Error
@@ -246,8 +236,8 @@ func SaveClientReport(clientUUID string, report v1.Report) (err error) {
 			return fmt.Errorf("failed to load previous Record: %w", err)
 		}
 		if previous != nil {
-			Record.TrafficUp = computeTrafficDelta(report.Network.TotalUp, previous.NetTotalUp)
-			Record.TrafficDown = computeTrafficDelta(report.Network.TotalDown, previous.NetTotalDown)
+			Record.TrafficUp = utils.ComputeTrafficDelta(report.Network.TotalUp, previous.NetTotalUp)
+			Record.TrafficDown = utils.ComputeTrafficDelta(report.Network.TotalDown, previous.NetTotalDown)
 		}
 
 		// 保存 Record

database/clients/report_test.go

@@ -5,6 +5,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/komari-monitor/komari/utils"
 	"github.com/stretchr/testify/assert"
 )
 
@@ -23,7 +24,7 @@ func TestComputeTrafficDeltaHandlesZeroAndReset(t *testing.T) {
 
 	for _, test := range tests {
 		t.Run(test.name, func(t *testing.T) {
-			assert.Equal(t, test.want, computeTrafficDelta(test.current, test.previous))
+			assert.Equal(t, test.want, utils.ComputeTrafficDelta(test.current, test.previous))
 		})
 	}
 }

database/dbcore/dbcore.go

@@ -170,6 +170,28 @@ var (
 	once     sync.Once
 )
 
+func buildSQLiteDSN(databaseFile string) string {
+	if databaseFile == "" {
+		databaseFile = "./data/komari.db"
+	}
+
+	params := "_busy_timeout=5000&_txlock=immediate&_journal_mode=WAL&_synchronous=NORMAL"
+	separator := "?"
+	if strings.Contains(databaseFile, "?") {
+		separator = "&"
+	}
+
+	if strings.HasPrefix(databaseFile, "file:") {
+		return databaseFile + separator + params
+	}
+
+	if databaseFile == ":memory:" {
+		return "file::memory:?cache=shared&" + params
+	}
+
+	return "file:" + filepath.ToSlash(databaseFile) + separator + params
+}
+
 func GetDBInstance() *gorm.DB {
 	once.Do(func() {
 
@@ -227,16 +249,37 @@ func GetDBInstance() *gorm.DB {
 		switch flags.ApplyDatabaseTypeNormalization() {
 		case flags.DatabaseTypeSQLite:
 			// SQLite 连接
-			instance, err = gorm.Open(sqlite.Open(flags.DatabaseFile), logConfig)
+			// 通过 DSN 传入 _busy_timeout / _txlock 等参数，确保连接池中的每一条连接
+			// 都生效：
+			//   - _busy_timeout=5000：遇到写锁时最多等待 5s 再返回，避免瞬时
+			//     "database is locked" 直接失败（仅靠后续 PRAGMA Exec 只作用于
+			//     当时执行该语句的单条连接，池内其它连接不生效）。
+			//   - _txlock=immediate：事务一开始即获取写锁，避免「先 SELECT 后写」
+			//     的锁升级在并发写入下产生死锁式的立即 SQLITE_BUSY。
+			//   - _journal_mode=WAL / _synchronous=NORMAL：与下方 PRAGMA 保持一致，
+			//     在 DSN 层为所有连接预设。
+			dsn := buildSQLiteDSN(flags.DatabaseFile)
+			instance, err = gorm.Open(sqlite.Open(dsn), logConfig)
 			if err != nil {
 				log.Fatalf("Failed to connect to SQLite3 database: %v", err)
 			}
 			log.Printf("Using SQLite database file: %s", flags.DatabaseFile)
+			if sqlDB, dbErr := instance.DB(); dbErr == nil {
+				// SQLite 同一时刻只允许一个写者；限制连接数可避免连接池层面的写写竞争。
+				// 负载历史每分钟会执行包含读和写的事务，若连接池允许多个连接，容易与
+				// ping 结果等短写入撞锁并导致整批负载记录回滚。
+				sqlDB.SetMaxOpenConns(1)
+				sqlDB.SetMaxIdleConns(1)
+				sqlDB.SetConnMaxLifetime(0)
+			} else {
+				log.Printf("Failed to access underlying sql.DB for SQLite tuning: %v", dbErr)
+			}
 			instance.Exec("PRAGMA wal = ON;")
 			if err := instance.Exec("PRAGMA journal_mode = WAL;").Error; err != nil {
 				log.Printf("Failed to enable WAL mode for SQLite: %v", err)
 			}
 			instance.Exec("PRAGMA synchronous = NORMAL;")
+			instance.Exec("PRAGMA busy_timeout = 5000;")
 			instance.Exec("PRAGMA cache_size = -65536;")
 			instance.Exec("PRAGMA temp_store = MEMORY;")
 			instance.Exec("PRAGMA wal_checkpoint(TRUNCATE);")

database/records/records.go

@@ -1,6 +1,7 @@
 package records
 
 import (
+	"errors"
 	"log"
 	"sort"
 	"strings"
@@ -11,6 +12,7 @@ import (
 	"github.com/komari-monitor/komari/cmd/flags"
 	"github.com/komari-monitor/komari/database/dbcore"
 	"github.com/komari-monitor/komari/database/models"
+	"github.com/komari-monitor/komari/utils"
 )
 
 func RecordOne(rec models.Record) error {
@@ -199,6 +201,11 @@ func migrateOldRecordsAt(db *gorm.DB, now time.Time) error {
 	if len(records) == 0 {
 		return nil
 	}
+	previousByClient, err := getPreviousTrafficRecordsBefore(db, records)
+	if err != nil {
+		return err
+	}
+	repairZeroTrafficDeltas(records, previousByClient)
 
 	// 按 Client 和 15 分钟时间段分组，并存储所有记录以计算分位数
 	type groupData struct {
@@ -379,6 +386,84 @@ func migrateOldRecordsAt(db *gorm.DB, now time.Time) error {
 	})
 }
 
+func getPreviousTrafficRecordsBefore(db *gorm.DB, records []models.Record) (map[string]*models.Record, error) {
+	firstTimeByClient := make(map[string]time.Time)
+	for _, record := range records {
+		recordTime := record.Time.ToTime()
+		firstTime, ok := firstTimeByClient[record.Client]
+		if !ok || recordTime.Before(firstTime) {
+			firstTimeByClient[record.Client] = recordTime
+		}
+	}
+
+	previousByClient := make(map[string]*models.Record, len(firstTimeByClient))
+	for clientUUID, firstTime := range firstTimeByClient {
+		previous, err := getPreviousTrafficRecordBefore(db, clientUUID, firstTime)
+		if err != nil {
+			return nil, err
+		}
+		if previous != nil {
+			previousByClient[clientUUID] = previous
+		}
+	}
+	return previousByClient, nil
+}
+
+func getPreviousTrafficRecordBefore(db *gorm.DB, clientUUID string, before time.Time) (*models.Record, error) {
+	var latest *models.Record
+	for _, table := range []string{"records", "records_long_term"} {
+		var record models.Record
+		queryBefore := before
+		if table == "records_long_term" {
+			queryBefore = before.Truncate(15 * time.Minute)
+		}
+		err := db.Table(table).
+			Where("client = ? AND time < ?", clientUUID, models.FromTime(queryBefore)).
+			Order("time DESC").
+			First(&record).Error
+		if err == nil {
+			if latest == nil || record.Time.ToTime().After(latest.Time.ToTime()) {
+				latest = &record
+			}
+			continue
+		}
+		if !errors.Is(err, gorm.ErrRecordNotFound) {
+			return nil, err
+		}
+	}
+	return latest, nil
+}
+
+func repairZeroTrafficDeltas(records []models.Record, previousByClient map[string]*models.Record) {
+	recordsByClient := make(map[string][]*models.Record)
+	for i := range records {
+		recordsByClient[records[i].Client] = append(recordsByClient[records[i].Client], &records[i])
+	}
+
+	for _, clientRecords := range recordsByClient {
+		sort.Slice(clientRecords, func(i, j int) bool {
+			return clientRecords[i].Time.ToTime().Before(clientRecords[j].Time.ToTime())
+		})
+		var previous *models.Record
+		if previousByClient != nil {
+			previous = previousByClient[clientRecords[0].Client]
+		}
+		for _, current := range clientRecords {
+			if previous == nil {
+				previous = current
+				continue
+			}
+			if current.TrafficUp == 0 {
+				current.TrafficUp = utils.ComputeTrafficDelta(current.NetTotalUp, previous.NetTotalUp)
+			}
+			if current.TrafficDown == 0 {
+				current.TrafficDown = utils.ComputeTrafficDelta(current.NetTotalDown, previous.NetTotalDown)
+			}
+			previous = current
+		}
+	}
+}
+
 // migrateGPURecords 压缩GPU记录数据
 func migrateGPURecords(db *gorm.DB) error {
 	cutoff := compactRecordCutoff(time.Now())

database/records/records_test.go

@@ -8,6 +8,7 @@ import (
 	"time"
 
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 	"gorm.io/driver/sqlite"
 	"gorm.io/gorm"
 
@@ -25,7 +26,8 @@ var _ = func() bool {
 // then running migrateOldRecords and verifying the aggregation and cleanup.
 func TestCompactRecord(t *testing.T) {
 	const totalMinutes = 12*60 + 30
-	now := time.Now().UTC().Truncate(time.Minute)
+	loc := models.GetAppLocation()
+	now := time.Date(2026, 6, 15, 14, 30, 0, 0, loc)
 	threshold := compactRecordCutoff(now)
 	overlapCutoff := threshold.Add(-1 * time.Hour)
 
@@ -132,7 +134,8 @@ func TestCompactRecord(t *testing.T) {
 }
 
 func TestCompactRecordPreservesExactTrafficDelta(t *testing.T) {
-	currentTime := time.Now().UTC().Truncate(time.Minute)
+	loc := models.GetAppLocation()
+	currentTime := time.Date(2026, 6, 15, 14, 30, 0, 0, loc)
 	now := currentTime.Truncate(15 * time.Minute).Add(-5*time.Hour + time.Minute)
 	db, err := gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
 	assert.NoError(t, err)
@@ -174,7 +177,7 @@ func TestCompactRecordPreservesExactTrafficDelta(t *testing.T) {
 
 	var compacted []models.Record
 	assert.NoError(t, db.Table("records_long_term").Find(&compacted).Error)
-	assert.Len(t, compacted, 1)
+	require.Len(t, compacted, 1)
 	assert.Equal(t, int64(60), compacted[0].TrafficUp)
 	assert.Equal(t, int64(90), compacted[0].TrafficDown)
 	assert.Equal(t, int64(10), compacted[0].NetTotalUp)
@@ -182,8 +185,91 @@ func TestCompactRecordPreservesExactTrafficDelta(t *testing.T) {
 	assert.True(t, compacted[0].Time.ToTime().Equal(records[2].Time.ToTime().Truncate(15*time.Minute)))
 }
 
+func TestRepairZeroTrafficDeltasPreservesRawResetDetailBeforeCompaction(t *testing.T) {
+	loc := models.GetAppLocation()
+	start := time.Date(2026, 6, 6, 0, 0, 0, 0, loc)
+	records := []models.Record{
+		{Client: uuid, Time: models.FromTime(start), NetTotalUp: 100, NetTotalDown: 200},
+		{Client: uuid, Time: models.FromTime(start.Add(5 * time.Minute)), NetTotalUp: 140, NetTotalDown: 260},
+		{Client: uuid, Time: models.FromTime(start.Add(10 * time.Minute)), NetTotalUp: 10, NetTotalDown: 20},
+		{Client: uuid, Time: models.FromTime(start.Add(15 * time.Minute)), NetTotalUp: 25, NetTotalDown: 35},
+	}
+
+	repairZeroTrafficDeltas(records, nil)
+
+	assert.Equal(t, int64(0), records[0].TrafficUp)
+	assert.Equal(t, int64(0), records[0].TrafficDown)
+	assert.Equal(t, int64(40), records[1].TrafficUp)
+	assert.Equal(t, int64(60), records[1].TrafficDown)
+	assert.Equal(t, int64(10), records[2].TrafficUp)
+	assert.Equal(t, int64(20), records[2].TrafficDown)
+	assert.Equal(t, int64(15), records[3].TrafficUp)
+	assert.Equal(t, int64(15), records[3].TrafficDown)
+}
+
+func TestRepairZeroTrafficDeltasUsesPreviousPersistedBaseline(t *testing.T) {
+	loc := models.GetAppLocation()
+	now := time.Date(2026, 6, 7, 12, 0, 0, 0, loc)
+	db, err := gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
+	assert.NoError(t, err)
+	assert.NoError(t, db.AutoMigrate(&models.Record{}))
+	assert.NoError(t, db.Table("records_long_term").AutoMigrate(&models.Record{}))
+
+	assert.NoError(t, db.Table("records_long_term").Create(&models.Record{
+		Client:       uuid,
+		Time:         models.FromTime(now.Add(-10 * time.Minute)),
+		NetTotalUp:   100,
+		NetTotalDown: 200,
+	}).Error)
+
+	rawRecords := []models.Record{
+		{Client: uuid, Time: models.FromTime(now), NetTotalUp: 150, NetTotalDown: 260},
+		{Client: uuid, Time: models.FromTime(now.Add(5 * time.Minute)), NetTotalUp: 175, NetTotalDown: 300},
+	}
+
+	previousByClient, err := getPreviousTrafficRecordsBefore(db, rawRecords)
+	assert.NoError(t, err)
+	repairZeroTrafficDeltas(rawRecords, previousByClient)
+
+	assert.Equal(t, int64(50), rawRecords[0].TrafficUp)
+	assert.Equal(t, int64(60), rawRecords[0].TrafficDown)
+	assert.Equal(t, int64(25), rawRecords[1].TrafficUp)
+	assert.Equal(t, int64(40), rawRecords[1].TrafficDown)
+}
+
+func TestRepairZeroTrafficDeltasIgnoresSameSlotLongTermBaseline(t *testing.T) {
+	loc := models.GetAppLocation()
+	now := time.Date(2026, 6, 7, 12, 0, 30, 0, loc)
+	db, err := gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
+	assert.NoError(t, err)
+	assert.NoError(t, db.AutoMigrate(&models.Record{}))
+	assert.NoError(t, db.Table("records_long_term").AutoMigrate(&models.Record{}))
+
+	assert.NoError(t, db.Table("records_long_term").Create(&models.Record{
+		Client:       uuid,
+		Time:         models.FromTime(now.Truncate(15 * time.Minute)),
+		NetTotalUp:   900,
+		NetTotalDown: 900,
+	}).Error)
+
+	rawRecords := []models.Record{
+		{Client: uuid, Time: models.FromTime(now), NetTotalUp: 150, NetTotalDown: 260},
+		{Client: uuid, Time: models.FromTime(now.Add(5 * time.Minute)), NetTotalUp: 175, NetTotalDown: 300},
+	}
+
+	previousByClient, err := getPreviousTrafficRecordsBefore(db, rawRecords)
+	assert.NoError(t, err)
+	repairZeroTrafficDeltas(rawRecords, previousByClient)
+
+	assert.Equal(t, int64(0), rawRecords[0].TrafficUp)
+	assert.Equal(t, int64(0), rawRecords[0].TrafficDown)
+	assert.Equal(t, int64(25), rawRecords[1].TrafficUp)
+	assert.Equal(t, int64(40), rawRecords[1].TrafficDown)
+}
+
 func TestCompactRecordRetainsOneHourOverlapWindow(t *testing.T) {
-	now := time.Date(2026, 6, 7, 12, 7, 0, 0, time.UTC)
+	loc := models.GetAppLocation()
+	now := time.Date(2026, 6, 7, 12, 7, 0, 0, loc)
 	cutoff := compactRecordCutoff(now)
 	db, err := gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
 	assert.NoError(t, err)
@@ -203,13 +289,14 @@ func TestCompactRecordRetainsOneHourOverlapWindow(t *testing.T) {
 
 	var remainTimes []models.Record
 	assert.NoError(t, db.Table("records").Order("time ASC").Find(&remainTimes).Error)
-	assert.Len(t, remainTimes, 2)
+	require.Len(t, remainTimes, 2)
 	assert.True(t, remainTimes[0].Time.ToTime().Equal(records[1].Time.ToTime()))
 	assert.True(t, remainTimes[1].Time.ToTime().Equal(records[2].Time.ToTime()))
 }
 
 func TestCompactRecordOnlyMigratesCompleteFifteenMinuteBuckets(t *testing.T) {
-	now := time.Date(2026, 6, 7, 12, 7, 0, 0, time.UTC)
+	loc := models.GetAppLocation()
+	now := time.Date(2026, 6, 7, 12, 7, 0, 0, loc)
 	cutoff := compactRecordCutoff(now)
 	db, err := gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
 	assert.NoError(t, err)
@@ -225,7 +312,7 @@ func TestCompactRecordOnlyMigratesCompleteFifteenMinuteBuckets(t *testing.T) {
 
 	var compacted []models.Record
 	assert.NoError(t, db.Table("records_long_term").Order("time ASC").Find(&compacted).Error)
-	assert.Len(t, compacted, 1)
+	require.Len(t, compacted, 1)
 	assert.True(t, compacted[0].Time.ToTime().Equal(compactable.Time.ToTime().Truncate(15*time.Minute)))
 
 	var rawCount int64

database/utils.go

@@ -134,7 +134,7 @@ func GetPublicInfo() (map[string]interface{}, error) {
 		"oauth_enable":              cst.OAuthEnabled,
 		"oauth_provider":            cst.OAuthProvider,
 		"disable_password_login":    cst.DisablePasswordLogin,
-		"allow_cors":                cst.AllowCors,
+		"cors_origin_check_enabled": cst.CorsOriginCheckEnabled,
 		"record_enabled":            cst.RecordEnabled,
 		"record_preserve_time":      cst.RecordPreserveTime,
 		"ping_record_preserve_time": cst.PingRecordPreserveTime,

docs/README_ja.md

@@ -6,6 +6,11 @@
 
 Komariは、サーバーのパフォーマンスを監視するためのシンプルで効率的なソリューションを提供することを目的とした、軽量の自己ホスト型サーバー監視ツールです。Webインターフェースを介してサーバーのステータスを表示し、軽量エージェントを介してデータを収集します。
 
+> [!WARNING]
+> Komariは、自己ホスト型の監視/制御プログラムであり、所有しているシステム、または管理権限を得ているシステムでのみ使用してください。Komariを武器化したり、許可なく展開、アクセス、永続化、コマンド実行、その他の不正利用に使用したりしないでください。実際の悪用リスクについては、Huntressの分析をご参照ください：[Komari C2 agent abuse](https://www.huntress.com/blog/komari-c2-agent-abuse)。
+> Komariの展開および運用方法については、利用者自身が責任を負います。開発者は、無許可または不正な利用、およびその結果について責任を負いません。
+> Windows端末でリモートコントロールを有効にした場合、クライアントはユーザーがログインするたびに、KomariがリモートコントロールソフトウェアであることをWindows通知で知らせます。
+
 [ドキュメント](https://komari-document.pages.dev/) | [Telegramグループ](https://t.me/komari_monitor)
 
 ## 特徴

docs/README_zh-TW.md

@@ -6,6 +6,11 @@
 
 Komari 是一款輕量級的自託管伺服器監控工具，旨在提供簡單、高效的伺服器性能監控解決方案。它支援透過 Web 介面查看伺服器狀態，並透過輕量級 Agent 收集數據。
 
+> [!WARNING]
+> Komari 是一款自託管的監控/控制程式，僅應部署在你擁有或已獲授權管理的系統上。請勿將 Komari 武器化，或在未獲授權的情況下部署、存取、持久化、執行命令及從事其他濫用行為。關於現實中的濫用風險，可參考 Huntress 的分析：[Komari C2 agent abuse](https://www.huntress.com/blog/komari-c2-agent-abuse)。
+> 使用者需要自行承擔部署和使用 Komari 的責任。開發者不對未經授權或濫用行為及其後果承擔責任。
+> 在 Windows 端啟用遠端控制後，客戶端會在每次使用者登入時透過 Windows 通知提醒使用者 Komari 是一款遠端控制軟體。
+
 [文檔](https://komari-document.pages.dev/) | [Telegram 群組](https://t.me/komari_monitor)
 
 ## 特性

docs/README_zh.md

@@ -6,6 +6,11 @@
 
 Komari 是一款轻量级的自托管服务器监控工具，旨在提供简单、高效的服务器性能监控解决方案。它支持通过 Web 界面查看服务器状态，并通过轻量级 Agent 收集数据。
 
+> [!WARNING]
+> Komari 是一款自托管的监控/控制程序，仅应部署在你拥有或已获得授权管理的系统上。请勿将 Komari 武器化，或在未获授权的情况下部署、访问、持久化、执行命令及从事其他滥用行为。关于现实中的滥用风险，可参考 Huntress 的分析：[Komari C2 agent abuse](https://www.huntress.com/blog/komari-c2-agent-abuse)。
+> 用户需要自行承担部署和使用 Komari 的责任。开发者不对未经授权或滥用行为及其后果承担责任。
+> 在 Windows 端开启远程控制后，客户端会在每次用户登录时通过 Windows 通知提醒用户 Komari 是一款远程控制软件。
+
 [文档](https://komari-document.pages.dev/) | [文档(镜像站 By Geekertao)](https://www.komari.wiki) | [Telegram 群组](https://t.me/komari_monitor)
 
 ## 特性

go.mod

@@ -1,21 +1,22 @@
 module github.com/komari-monitor/komari
 
-go 1.24.0
-
-toolchain go1.24.11
+go 1.25.0
 
 require (
 	github.com/coreos/go-oidc/v3 v3.15.0
 	github.com/dop251/goja v0.0.0-20251008123653-cf18d89f3cf6
 	github.com/dop251/goja_nodejs v0.0.0-20251015164255-5e94316bedaf
 	github.com/gin-gonic/gin v1.10.0
+	github.com/go-sql-driver/mysql v1.10.0
 	github.com/google/uuid v1.6.0
 	github.com/gorilla/websocket v1.5.3
+	github.com/jackc/pgx/v5 v5.10.0
+	github.com/mattn/go-sqlite3 v1.14.27
 	github.com/oschwald/maxminddb-golang v1.13.1
 	github.com/patrickmn/go-cache v2.1.0+incompatible
 	github.com/pquerna/otp v1.5.0
 	github.com/spf13/cobra v1.9.1
-	github.com/stretchr/testify v1.10.0
+	github.com/stretchr/testify v1.11.1
 	golang.org/x/image v0.35.0
 	google.golang.org/grpc v1.75.0
 	google.golang.org/protobuf v1.36.6
@@ -32,6 +33,7 @@ require golang.org/x/crypto v0.39.0 // indirect
 require golang.org/x/net v0.41.0 // indirect
 
 require (
+	filippo.io/edwards25519 v1.2.0 // indirect
 	github.com/boombuler/barcode v1.0.1-0.20190219062509-6c824513bacc // indirect
 	github.com/bytedance/sonic v1.13.2 // indirect
 	github.com/bytedance/sonic/loader v0.2.4 // indirect
@@ -48,6 +50,9 @@ require (
 	github.com/goccy/go-json v0.10.5 // indirect
 	github.com/google/pprof v0.0.0-20240727154555-813a5fbdbec8 // indirect
 	github.com/inconshreveable/mousetrap v1.1.0 // indirect
+	github.com/jackc/pgpassfile v1.0.0 // indirect
+	github.com/jackc/pgservicefile v0.0.0-20240606120523-5a60cdf6a761 // indirect
+	github.com/jackc/puddle/v2 v2.2.2 // indirect
 	github.com/jinzhu/inflection v1.0.0 // indirect
 	github.com/jinzhu/now v1.1.5 // indirect
 	github.com/json-iterator/go v1.1.12 // indirect
@@ -55,7 +60,6 @@ require (
 	github.com/kr/text v0.2.0 // indirect
 	github.com/leodido/go-urn v1.4.0 // indirect
 	github.com/mattn/go-isatty v0.0.20 // indirect
-	github.com/mattn/go-sqlite3 v1.14.27 // indirect
 	github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd // indirect
 	github.com/modern-go/reflect2 v1.0.2 // indirect
 	github.com/pelletier/go-toml/v2 v2.2.3 // indirect
@@ -66,9 +70,9 @@ require (
 	github.com/ugorji/go/codec v1.2.12 // indirect
 	golang.org/x/arch v0.15.0 // indirect
 	golang.org/x/oauth2 v0.30.0 // indirect
+	golang.org/x/sync v0.19.0 // indirect
 	golang.org/x/sys v0.33.0 // indirect
 	golang.org/x/text v0.33.0 // indirect
 	google.golang.org/genproto/googleapis/rpc v0.0.0-20250707201910-8d1bb00bc6a7 // indirect
-	gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c // indirect
 	gopkg.in/yaml.v3 v3.0.1 // indirect
 )

go.sum

@@ -1,3 +1,5 @@
+filippo.io/edwards25519 v1.2.0 h1:crnVqOiS4jqYleHd9vaKZ+HKtHfllngJIiOpNpoJsjo=
+filippo.io/edwards25519 v1.2.0/go.mod h1:xzAOLCNug/yB62zG1bQ8uziwrIqIuxhctzJT18Q77mc=
 github.com/Masterminds/semver/v3 v3.2.1 h1:RN9w6+7QoMeJVGyfmbcgs28Br8cvmnucEXnY0rYXWg0=
 github.com/Masterminds/semver/v3 v3.2.1/go.mod h1:qvl/7zhW3nngYb5+80sSMF+FG2BjYrf8m9wsX0PNOMQ=
 github.com/boombuler/barcode v1.0.1-0.20190219062509-6c824513bacc h1:biVzkmvwrH8WK8raXaxBx6fRVTlJILwEwQGL1I/ByEI=
@@ -45,6 +47,8 @@ github.com/go-playground/validator/v10 v10.26.0 h1:SP05Nqhjcvz81uJaRfEV0YBSSSGMc
 github.com/go-playground/validator/v10 v10.26.0/go.mod h1:I5QpIEbmr8On7W0TktmJAumgzX4CA1XNl4ZmDuVHKKo=
 github.com/go-sourcemap/sourcemap v2.1.4+incompatible h1:a+iTbH5auLKxaNwQFg0B+TCYl6lbukKPc7b5x0n1s6Q=
 github.com/go-sourcemap/sourcemap v2.1.4+incompatible/go.mod h1:F8jJfvm2KbVjc5NqelyYJmf/v5J0dwNLS2mL4sNA1Jg=
+github.com/go-sql-driver/mysql v1.10.0 h1:Q+1LV8DkHJvSYAdR83XzuhDaTykuDx0l6fkXxoWCWfw=
+github.com/go-sql-driver/mysql v1.10.0/go.mod h1:M+cqaI7+xxXGG9swrdeUIoPG3Y3KCkF0pZej+SK+nWk=
 github.com/goccy/go-json v0.10.5 h1:Fq85nIqj+gXn/S5ahsiTlK3TmC85qgirsdTP/+DeaC4=
 github.com/goccy/go-json v0.10.5/go.mod h1:oq7eo15ShAhp70Anwd5lgX2pLfOS3QCiwU/PULtXL6M=
 github.com/golang/protobuf v1.5.4 h1:i7eJL8qZTpSEXOPTxNKhASYpMn+8e5Q6AdndVa1dWek=
@@ -60,6 +64,14 @@ github.com/gorilla/websocket v1.5.3 h1:saDtZ6Pbx/0u+bgYQ3q96pZgCzfhKXGPqt7kZ72aN
 github.com/gorilla/websocket v1.5.3/go.mod h1:YR8l580nyteQvAITg2hZ9XVh4b55+EU/adAjf1fMHhE=
 github.com/inconshreveable/mousetrap v1.1.0 h1:wN+x4NVGpMsO7ErUn/mUI3vEoE6Jt13X2s0bqwp9tc8=
 github.com/inconshreveable/mousetrap v1.1.0/go.mod h1:vpF70FUmC8bwa3OWnCshd2FqLfsEA9PFc4w1p2J65bw=
+github.com/jackc/pgpassfile v1.0.0 h1:/6Hmqy13Ss2zCq62VdNG8tM1wchn8zjSGOBJ6icpsIM=
+github.com/jackc/pgpassfile v1.0.0/go.mod h1:CEx0iS5ambNFdcRtxPj5JhEz+xB6uRky5eyVu/W2HEg=
+github.com/jackc/pgservicefile v0.0.0-20240606120523-5a60cdf6a761 h1:iCEnooe7UlwOQYpKFhBabPMi4aNAfoODPEFNiAnClxo=
+github.com/jackc/pgservicefile v0.0.0-20240606120523-5a60cdf6a761/go.mod h1:5TJZWKEWniPve33vlWYSoGYefn3gLQRzjfDlhSJ9ZKM=
+github.com/jackc/pgx/v5 v5.10.0 h1:VhSvgU2jSli8o3AqIEOTJr7rZwAEUVo4E4XhR94Zfr0=
+github.com/jackc/pgx/v5 v5.10.0/go.mod h1:mal1tBGAFfLHvZzaYh77YS/eC6IX9OWbRV1QIIM0Jn4=
+github.com/jackc/puddle/v2 v2.2.2 h1:PR8nw+E/1w0GLuRFSmiioY6UooMp6KJv0/61nB7icHo=
+github.com/jackc/puddle/v2 v2.2.2/go.mod h1:vriiEXHvEE654aYKXXjOvZM39qJ0q+azkZFrfEOc3H4=
 github.com/jinzhu/inflection v1.0.0 h1:K317FqzuhWc8YvSVlFMCCUb36O/S9MCKRDI7QkRKD/E=
 github.com/jinzhu/inflection v1.0.0/go.mod h1:h+uFLlag+Qp1Va5pdKtLDYj+kHp5pxUVkryuEj+Srlc=
 github.com/jinzhu/now v1.1.5 h1:/o9tlHleP7gOFmsnYNz3RGnqzefHA47wQpKrrdTIwXQ=
@@ -71,7 +83,6 @@ github.com/klauspost/cpuid/v2 v2.2.10 h1:tBs3QSyvjDyFTq3uoc/9xFpCuOsJQFNPiAhYdw2
 github.com/klauspost/cpuid/v2 v2.2.10/go.mod h1:hqwkgyIinND0mEev00jJYCxPNVRVXFQeu1XKlok6oO0=
 github.com/knz/go-libedit v1.10.1/go.mod h1:MZTVkCWyz0oBc7JOWP3wNAzd002ZbM/5hgShxwh4x8M=
 github.com/kr/pretty v0.1.0/go.mod h1:dAy3ld7l9f0ibDNOQOHHMYYIIbhfbHSm3C4ZsoJORNo=
-github.com/kr/pretty v0.2.1/go.mod h1:ipq/a2n7PKx3OHsz4KJII5eveXtPO4qwEXGdVfWzfnI=
 github.com/kr/pretty v0.3.0 h1:WgNl7dwNpEZ6jJ9k1snq4pZsg7DOEN8hP9Xw0Tsjwk0=
 github.com/kr/pretty v0.3.0/go.mod h1:640gp4NfQd8pI5XOwp5fnNeVWj67G7CFk/SaSQn7NBk=
 github.com/kr/pty v1.1.1/go.mod h1:pFQYn66WHrOpPYNljwOMqo10TkYh1fy3cYio2l3bCsQ=
@@ -117,8 +128,9 @@ github.com/stretchr/testify v1.7.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/
 github.com/stretchr/testify v1.8.0/go.mod h1:yNjHg4UonilssWZ8iaSj1OCr/vHnekPRkoO+kdMU+MU=
 github.com/stretchr/testify v1.8.1/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o6fzry7u4=
 github.com/stretchr/testify v1.8.4/go.mod h1:sz/lmYIOXD/1dqDmKjjqLyZ2RngseejIcXlSw2iwfAo=
-github.com/stretchr/testify v1.10.0 h1:Xv5erBjTwe/5IxqUQTdXv5kgmIvbHo3QQyRwhJsOfJA=
 github.com/stretchr/testify v1.10.0/go.mod h1:r2ic/lqez/lEtzL7wO/rwa5dbSLXVDPFyf8C91i36aY=
+github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu7U=
+github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=
 github.com/twitchyliquid64/golang-asm v0.15.1 h1:SU5vSMR7hnwNxj24w34ZyCi/FmDZTkS4MhqMhdFk5YI=
 github.com/twitchyliquid64/golang-asm v0.15.1/go.mod h1:a1lVb/DtPvCB8fslRZhAngC2+aY1QWCk3Cedj/Gdt08=
 github.com/ugorji/go/codec v1.2.12 h1:9LC83zGrHhuUA9l16C9AHXAqEV/2wBQ4nkvumAE65EE=
@@ -145,6 +157,8 @@ golang.org/x/net v0.41.0 h1:vBTly1HeNPEn3wtREYfy4GZ/NECgw2Cnl+nK6Nz3uvw=
 golang.org/x/net v0.41.0/go.mod h1:B/K4NNqkfmg07DQYrbwvSluqCJOOXwUjeb/5lOisjbA=
 golang.org/x/oauth2 v0.30.0 h1:dnDm7JmhM45NNpd8FDDeLhK6FwqbOf4MLCM9zb1BOHI=
 golang.org/x/oauth2 v0.30.0/go.mod h1:B++QgG3ZKulg6sRPGD/mqlHQs5rB3Ml9erfeDY7xKlU=
+golang.org/x/sync v0.19.0 h1:vV+1eWNmZ5geRlYjzm2adRgW2/mcpevXNg50YZtPCE4=
+golang.org/x/sync v0.19.0/go.mod h1:9KTHXmSnoGruLpwFjVSX0lNNA75CykiMECbovNTZqGI=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.33.0 h1:q3i8TbbEz+JRD9ywIRlyRAQbM0qF7hu24q3teo2hbuw=
 golang.org/x/sys v0.33.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=

install-komari.sh

@@ -31,9 +31,135 @@ SERVICE_NAME="komari"
 BINARY_PATH="$INSTALL_DIR/komari"
 DEFAULT_PORT="25774"
 LISTEN_PORT=""
+REPO="komari-monitor/komari"
+# 发布通道: stable（稳定版）或 snapshot（快照版）
+CHANNEL="stable"
+# TUI 工具: whiptail / dialog / 空（回退纯文本）
+TUI_TOOL=""
 
-# Show banner
+# ==========================================================
+# TUI / 交互层
+# ==========================================================
+
+# 检测可用的 TUI 工具
+detect_tui() {
+    if command -v whiptail >/dev/null 2>&1; then
+        TUI_TOOL="whiptail"
+    elif command -v dialog >/dev/null 2>&1; then
+        TUI_TOOL="dialog"
+    else
+        TUI_TOOL=""
+    fi
+}
+
+# 是否启用 TUI
+tui_enabled() {
+    [ -n "$TUI_TOOL" ]
+}
+
+# 菜单选择
+# 用法: ui_menu "标题" "提示" tag1 "item1" tag2 "item2" ...
+# 返回: 选中的 tag（输出到 stdout），取消返回非零
+ui_menu() {
+    local title="$1"; shift
+    local prompt="$1"; shift
+
+    if tui_enabled; then
+        $TUI_TOOL --title "$title" --menu "$prompt" 20 70 10 "$@" 3>&1 1>&2 2>&3
+        return $?
+    fi
+
+    # 纯文本回退
+    {
+        echo
+        echo "=============================================================="
+        echo "  $title"
+        echo "=============================================================="
+        echo "$prompt"
+        echo
+        local tag item
+        local args=("$@")
+        local i=0
+        while [ $i -lt ${#args[@]} ]; do
+            tag="${args[$i]}"
+            item="${args[$((i + 1))]}"
+            echo "  $tag) $item"
+            i=$((i + 2))
+        done
+        echo
+    } >&2
+    local choice
+    read -r -p "输入选项: " choice >&2
+    echo "$choice"
+}
+
+# 输入框
+# 用法: ui_input "标题" "提示" "默认值"
+# 返回: 输入内容（输出到 stdout），取消返回非零
+ui_input() {
+    local title="$1"
+    local prompt="$2"
+    local default="$3"
+
+    if tui_enabled; then
+        $TUI_TOOL --title "$title" --inputbox "$prompt" 12 70 "$default" 3>&1 1>&2 2>&3
+        return $?
+    fi
+
+    local input
+    read -r -p "$prompt [默认: $default]: " input >&2
+    if [ -z "$input" ]; then
+        echo "$default"
+    else
+        echo "$input"
+    fi
+}
+
+# 是/否确认
+# 用法: ui_yesno "标题" "提示"
+# 返回: 0 表示 是，1 表示 否
+ui_yesno() {
+    local title="$1"
+    local prompt="$2"
+
+    if tui_enabled; then
+        $TUI_TOOL --title "$title" --yesno "$prompt" 12 70
+        return $?
+    fi
+
+    local confirm
+    read -r -p "$prompt (Y/n): " confirm >&2
+    if [[ $confirm =~ ^[Nn]$ ]]; then
+        return 1
+    fi
+    return 0
+}
+
+# 信息提示框
+# 用法: ui_msgbox "标题" "内容"
+ui_msgbox() {
+    local title="$1"
+    local content="$2"
+
+    if tui_enabled; then
+        $TUI_TOOL --title "$title" --msgbox "$content" 20 72
+        return
+    fi
+
+    echo
+    echo "=============================================================="
+    echo "  $title"
+    echo "=============================================================="
+    echo -e "$content"
+    echo "=============================================================="
+    read -r -p "按回车键继续..." _
+}
+
+# 显示横幅（仅纯文本模式）
 show_banner() {
+    if tui_enabled; then
+        return
+    fi
     clear
     echo "=============================================================="
     echo "            Komari Monitoring System Installer"
@@ -42,6 +168,31 @@ show_banner() {
     echo
 }
 
+# 选择发布通道，结果写入全局变量 CHANNEL
+select_channel() {
+    local choice
+    choice=$(ui_menu "选择发布通道" "请选择要使用的发布通道：" \
+        "stable" "稳定版 (推荐)" \
+        "snapshot" "快照版 (最新功能)")
+
+    case "$choice" in
+        snapshot|2)
+            CHANNEL="snapshot"
+            ;;
+        stable|1|"")
+            CHANNEL="stable"
+            ;;
+        *)
+            CHANNEL="stable"
+            ;;
+    esac
+    log_info "已选择通道: $CHANNEL"
+}
+
+# ==========================================================
+# 基础检查
+# ==========================================================
+
 # Check if running as root
 check_root() {
     if [ "$EUID" -ne 0 ]; then
@@ -113,19 +264,54 @@ install_dependencies() {
     fi
 }
 
+# Get download URL based on channel
+get_download_url() {
+    local arch=$1
+    local file_name="komari-linux-${arch}"
+
+    if [ "$CHANNEL" = "snapshot" ]; then
+        # 获取最新的 snapshot 预发布版本
+        log_info "获取最新 snapshot 版本..." >&2
+        local latest_snapshot=$(curl -s "https://api.github.com/repos/${REPO}/releases" | grep '"tag_name"' | grep 'Snapshot-' | head -1 | sed -e 's/.*"tag_name": *"//' -e 's/".*//')
+
+        if [ -z "$latest_snapshot" ]; then
+            log_error "未找到 snapshot 版本" >&2
+            return 1
+        fi
+
+        log_info "最新 snapshot 版本: $latest_snapshot" >&2
+        echo "https://github.com/${REPO}/releases/download/${latest_snapshot}/${file_name}"
+    else
+        # 稳定版：使用 latest
+        echo "https://github.com/${REPO}/releases/latest/download/${file_name}"
+    fi
+}
+
+# ==========================================================
+# 业务操作
+# ==========================================================
+
 # Binary installation
 install_binary() {
     log_step "开始二进制安装..."
 
     if is_installed; then
-        log_info "Komari 已安装。要升级，请使用升级选项。"
+        ui_msgbox "提示" "Komari 已安装。\n如需升级，请使用主菜单中的升级选项。"
         return
     fi
 
+    # 选择发布通道
+    select_channel
 
     # 监听端口输入，校验范围 1-65535
     while true; do
-        read -p "请输入监听端口 [默认: $DEFAULT_PORT]: " input_port
+        local input_port
+        input_port=$(ui_input "监听端口" "请输入 Komari 的监听端口 (1-65535)：" "$DEFAULT_PORT")
+        # 取消输入
+        if [ $? -ne 0 ]; then
+            log_info "安装已取消"
+            return
+        fi
         if [[ -z "$input_port" ]]; then
             LISTEN_PORT="$DEFAULT_PORT"
             break
@@ -133,7 +319,7 @@ install_binary() {
             LISTEN_PORT="$input_port"
             break
         else
-            log_error "端口号无效，请输入 1-65535 之间的数字。"
+            ui_msgbox "错误" "端口号无效，请输入 1-65535 之间的数字。"
         fi
     done
 
@@ -148,14 +334,17 @@ install_binary() {
     log_step "创建数据目录: $DATA_DIR"
     mkdir -p "$DATA_DIR"
 
-    local file_name="komari-linux-${arch}"
-    local download_url="https://github.com/komari-monitor/komari/releases/latest/download/${file_name}"
+    local download_url=$(get_download_url "$arch")
+    if [ $? -ne 0 ]; then
+        ui_msgbox "错误" "获取下载链接失败，请检查网络连接或稍后重试。"
+        return 1
+    fi
 
     log_step "下载 Komari 二进制文件..."
     log_info "URL: $download_url"
 
     if ! curl -L -o "$BINARY_PATH" "$download_url"; then
-        log_error "下载失败"
+        ui_msgbox "错误" "下载失败，请检查网络连接。"
         return 1
     fi
 
@@ -163,11 +352,7 @@ install_binary() {
     log_success "Komari 二进制文件安装完成: $BINARY_PATH"
 
     if ! check_systemd; then
-        log_step "警告：未检测到 systemd，跳过服务创建。"
-        log_step "您可以从命令行手动运行 Komari："
-        log_step "    $BINARY_PATH server -l 0.0.0.0:$LISTEN_PORT"
-        echo
-        log_success "安装完成！"
+        ui_msgbox "安装完成" "警告：未检测到 systemd，已跳过服务创建。\n\n您可以手动运行 Komari：\n    $BINARY_PATH server -l 0.0.0.0:$LISTEN_PORT"
         return
     fi
 
@@ -179,17 +364,16 @@ install_binary() {
 
     if systemctl is-active --quiet ${SERVICE_NAME}.service; then
         log_success "Komari 服务启动成功"
-        
+
         log_step "正在获取初始密码..."
-        sleep 5 
+        sleep 5
         local password=$(journalctl -u ${SERVICE_NAME} --since "1 minute ago" | grep "admin account created." | tail -n 1 | sed -e 's/.*admin account created.//')
         if [ -z "$password" ]; then
             log_error "未能获取初始密码，请检查日志"
         fi
         show_access_info "$password" "$LISTEN_PORT"
     else
-        log_error "Komari 服务启动失败"
-        log_info "查看日志: journalctl -u ${SERVICE_NAME} -f"
+        ui_msgbox "错误" "Komari 服务启动失败。\n\n查看日志: journalctl -u ${SERVICE_NAME} -f"
         return 1
     fi
 }
@@ -223,21 +407,22 @@ EOF
 show_access_info() {
     local password=$1
     local port=${2:-$DEFAULT_PORT}
-    echo
-    log_success "安装完成！"
-    echo
-    log_info "访问信息："
-    log_info "  URL: http://$(hostname -I | awk '{print $1}'):${port}"
+    local ip=$(hostname -I | awk '{print $1}')
+
+    local content="安装完成！\n\n"
+    content+="访问信息：\n"
+    content+="  URL: http://${ip}:${port}\n"
     if [ -n "$password" ]; then
-        log_info "初始登录信息（仅显示一次）: $password"
+        content+="  初始登录信息（仅显示一次）: ${password}\n"
     fi
-    echo
-    log_info "服务管理命令："
-    log_info "  状态:  systemctl status $SERVICE_NAME"
-    log_info "  启动:   systemctl start $SERVICE_NAME"
-    log_info "  停止:    systemctl stop $SERVICE_NAME"
-    log_info "  重启: systemctl restart $SERVICE_NAME"
-    log_info "  日志:    journalctl -u $SERVICE_NAME -f"
+    content+="\n服务管理命令：\n"
+    content+="  状态: systemctl status $SERVICE_NAME\n"
+    content+="  启动: systemctl start $SERVICE_NAME\n"
+    content+="  停止: systemctl stop $SERVICE_NAME\n"
+    content+="  重启: systemctl restart $SERVICE_NAME\n"
+    content+="  日志: journalctl -u $SERVICE_NAME -f"
+
+    ui_msgbox "安装完成" "$content"
 }
 
 # Upgrade function
@@ -245,15 +430,18 @@ upgrade_komari() {
     log_step "升级 Komari..."
 
     if ! is_installed; then
-        log_error "Komari 未安装。请先安装它。"
+        ui_msgbox "错误" "Komari 未安装。请先安装它。"
         return 1
     fi
 
     if ! check_systemd; then
-        log_error "未检测到 systemd。无法管理服务。"
+        ui_msgbox "错误" "未检测到 systemd。无法管理服务。"
         return 1
     fi
 
+    # 选择发布通道
+    select_channel
+
     log_step "停止 Komari 服务..."
     systemctl stop ${SERVICE_NAME}.service
 
@@ -261,14 +449,21 @@ upgrade_komari() {
     cp "$BINARY_PATH" "${BINARY_PATH}.backup.$(date +%Y%m%d_%H%M%S)"
 
     local arch=$(detect_arch)
-    local file_name="komari-linux-${arch}"
-    local download_url="https://github.com/komari-monitor/komari/releases/latest/download/${file_name}"
+    local download_url=$(get_download_url "$arch")
+    if [ $? -ne 0 ]; then
+        log_error "获取下载链接失败，正在从备份恢复"
+        mv "${BINARY_PATH}.backup."* "$BINARY_PATH"
+        systemctl start ${SERVICE_NAME}.service
+        ui_msgbox "错误" "获取下载链接失败，已从备份恢复。"
+        return 1
+    fi
 
     log_step "下载最新版本..."
     if ! curl -L -o "$BINARY_PATH" "$download_url"; then
         log_error "下载失败，正在从备份恢复"
         mv "${BINARY_PATH}.backup."* "$BINARY_PATH"
         systemctl start ${SERVICE_NAME}.service
+        ui_msgbox "错误" "下载失败，已从备份恢复。"
         return 1
     fi
 
@@ -278,9 +473,9 @@ upgrade_komari() {
     systemctl start ${SERVICE_NAME}.service
 
     if systemctl is-active --quiet ${SERVICE_NAME}.service; then
-        log_success "Komari 升级成功"
+        ui_msgbox "升级完成" "Komari 升级成功 (通道: $CHANNEL)。"
     else
-        log_error "服务在升级后未能启动"
+        ui_msgbox "错误" "服务在升级后未能启动，请检查日志。"
     fi
 }
 
@@ -289,12 +484,11 @@ uninstall_komari() {
     log_step "卸载 Komari..."
 
     if ! is_installed; then
-        log_info "Komari 未安装"
+        ui_msgbox "提示" "Komari 未安装。"
         return 0
     fi
 
-    read -p "这将删除 Komari。您确定吗？(Y/n): " confirm
-    if [[ $confirm =~ ^[Nn]$ ]]; then
+    if ! ui_yesno "确认卸载" "这将删除 Komari 二进制文件和服务。\n\n您确定要继续吗？"; then
         log_info "卸载已取消"
         return 0
     fi
@@ -314,102 +508,128 @@ uninstall_komari() {
     rmdir "$INSTALL_DIR" 2>/dev/null || log_info "数据目录 $INSTALL_DIR 不为空，未删除"
     log_success "Komari 二进制文件已删除"
 
-    log_success "Komari 卸载完成"
-    log_info "数据文件保留在 $DATA_DIR"
+    ui_msgbox "卸载完成" "Komari 卸载完成。\n\n数据文件保留在 $DATA_DIR"
 }
 
 # Show service status
 show_status() {
     if ! is_installed; then
-        log_error "Komari 未安装"
+        ui_msgbox "错误" "Komari 未安装。"
         return
     fi
     if ! check_systemd; then
-        log_error "未检测到 systemd。无法获取服务状态。"
+        ui_msgbox "错误" "未检测到 systemd。无法获取服务状态。"
         return
     fi
-    log_step "Komari 服务状态:"
-    systemctl status ${SERVICE_NAME}.service --no-pager -l
+    if tui_enabled; then
+        local status_output
+        status_output=$(systemctl status ${SERVICE_NAME}.service --no-pager -l 2>&1)
+        ui_msgbox "服务状态" "$status_output"
+    else
+        log_step "Komari 服务状态:"
+        systemctl status ${SERVICE_NAME}.service --no-pager -l
+        read -r -p "按回车键继续..." _
+    fi
 }
 
 # Show service logs
 show_logs() {
     if ! is_installed; then
-        log_error "Komari 未安装"
+        ui_msgbox "错误" "Komari 未安装。"
         return
     fi
     if ! check_systemd; then
-        log_error "未检测到 systemd。无法获取服务日志。"
+        ui_msgbox "错误" "未检测到 systemd。无法获取服务日志。"
         return
     fi
-    log_step "查看 Komari 服务日志..."
+    # 日志为实时流，直接在终端显示
+    if tui_enabled; then
+        clear
+    fi
+    log_step "查看 Komari 服务日志 (按 Ctrl+C 退出)..."
     journalctl -u ${SERVICE_NAME} -f --no-pager
 }
 
 # Restart service
 restart_service() {
     if ! is_installed; then
-        log_error "Komari 未安装"
+        ui_msgbox "错误" "Komari 未安装。"
         return
     fi
     if ! check_systemd; then
-        log_error "未检测到 systemd。无法重启服务。"
+        ui_msgbox "错误" "未检测到 systemd。无法重启服务。"
         return
     fi
     log_step "重启 Komari 服务..."
     systemctl restart ${SERVICE_NAME}.service
     if systemctl is-active --quiet ${SERVICE_NAME}.service; then
-        log_success "服务重启成功"
+        ui_msgbox "成功" "服务重启成功。"
     else
-        log_error "服务重启失败"
+        ui_msgbox "错误" "服务重启失败，请检查日志。"
     fi
 }
 
 # Stop service
 stop_service() {
     if ! is_installed; then
-        log_error "Komari 未安装"
+        ui_msgbox "错误" "Komari 未安装。"
         return
     fi
     if ! check_systemd; then
-        log_error "未检测到 systemd。无法停止服务。"
+        ui_msgbox "错误" "未检测到 systemd。无法停止服务。"
         return
     fi
     log_step "停止 Komari 服务..."
     systemctl stop ${SERVICE_NAME}.service
-    log_success "服务已停止"
+    ui_msgbox "成功" "服务已停止。"
 }
 
 
 # Main menu
 main_menu() {
-    show_banner
-    echo "请选择操作："
-    echo "  1) 安装 Komari"
-    echo "  2) 升级 Komari"
-    echo "  3) 卸载 Komari"
-    echo "  4) 查看状态"
-    echo "  5) 查看日志"
-    echo "  6) 重启服务"
-    echo "  7) 停止服务"
-    echo "  8) 退出"
-    echo
+    while true; do
+        show_banner
 
-    read -p "输入选项 [1-8]: " choice
+        local choice
+        choice=$(ui_menu "Komari 监控系统安装器" "请选择操作：" \
+            "1" "安装 Komari" \
+            "2" "升级 Komari" \
+            "3" "卸载 Komari" \
+            "4" "查看状态" \
+            "5" "查看日志" \
+            "6" "重启服务" \
+            "7" "停止服务" \
+            "8" "退出")
 
-    case $choice in
-        1) install_binary ;;
-        2) upgrade_komari ;;
-        3) uninstall_komari ;;
-        4) show_status ;;
-        5) show_logs ;;
-        6) restart_service ;;
-        7) stop_service ;;
-        8) exit 0 ;;
-        *) log_error "无效选项" ;;
-    esac
+        # 用户在 TUI 中取消（ESC/Cancel）则退出
+        if [ $? -ne 0 ] && tui_enabled; then
+            clear
+            exit 0
+        fi
+
+        case $choice in
+            1) install_binary ;;
+            2) upgrade_komari ;;
+            3) uninstall_komari ;;
+            4) show_status ;;
+            5) show_logs ;;
+            6) restart_service ;;
+            7) stop_service ;;
+            8) 
+                tui_enabled && clear
+                exit 0 
+                ;;
+            *) ui_msgbox "错误" "无效选项" ;;
+        esac
+
+        # 纯文本模式下单次执行后退出循环（保持原有行为，避免输出被覆盖）
+        if ! tui_enabled; then
+            break
+        fi
+    done
 }
 
 # Main execution
 check_root
+detect_tui
 main_menu

pkg/config/settings.go

@@ -3,21 +3,21 @@ package config
 import "time"
 
 type Settings struct {
-	ID                   uint   `json:"id,omitempty"`                                        // 1
-	Sitename             string `json:"sitename" default:"Komari"`                           // 站点名称，默认 "Komari"
-	Description          string `json:"description" default:"A simple server monitor tool."` // 站点描述
-	AllowCors            bool   `json:"allow_cors" default:"false"`                          // 是否允许跨域，默认 false
-	CorsAllowedOrigins   string `json:"cors_allowed_origins" default:""`                     // API 跨域允许列表
-	WsOriginCheckEnabled bool   `json:"ws_origin_check_enabled" default:"true"`              // 是否校验 WebSocket Origin
-	WsAllowedOrigins     string `json:"ws_allowed_origins" default:""`                       // WebSocket Origin 允许列表
-	Theme                string `json:"theme" default:"default"`                             // 主题名称，默认 'default'
-	PrivateSite          bool   `json:"private_site" default:"false"`                        // 是否为私有站点，默认 false
-	ApiKey               string `json:"api_key" default:""`                                  // API 密钥，默认空字符串
-	AutoDiscoveryKey     string `json:"auto_discovery_key" default:""`                       // 自动发现密钥
-	ScriptDomain         string `json:"script_domain" default:""`                            // 自定义脚本域名
-	SendIpAddrToGuest    bool   `json:"send_ip_addr_to_guest" default:"false"`               // 是否向访客页面发送 IP 地址，默认 false
-	EulaAccepted         bool   `json:"eula_accepted" default:"false"`
-	BaseScriptsURLKey    string `json:"base_scripts_url" default:""`
+	ID                     uint   `json:"id,omitempty"`                                        // 1
+	Sitename               string `json:"sitename" default:"Komari"`                           // 站点名称，默认 "Komari"
+	Description            string `json:"description" default:"A simple server monitor tool."` // 站点描述
+	CorsOriginCheckEnabled bool   `json:"cors_origin_check_enabled" default:"true"`            // 是否启用 API CORS 跨域请求校验，默认 true
+	CorsAllowedOrigins     string `json:"cors_allowed_origins" default:""`                     // API 跨域允许列表
+	WsOriginCheckEnabled   bool   `json:"ws_origin_check_enabled" default:"true"`              // 是否校验 WebSocket Origin
+	WsAllowedOrigins       string `json:"ws_allowed_origins" default:""`                       // WebSocket Origin 允许列表
+	Theme                  string `json:"theme" default:"default"`                             // 主题名称，默认 'default'
+	PrivateSite            bool   `json:"private_site" default:"false"`                        // 是否为私有站点，默认 false
+	ApiKey                 string `json:"api_key" default:""`                                  // API 密钥，默认空字符串
+	AutoDiscoveryKey       string `json:"auto_discovery_key" default:""`                       // 自动发现密钥
+	ScriptDomain           string `json:"script_domain" default:""`                            // 自定义脚本域名
+	SendIpAddrToGuest      bool   `json:"send_ip_addr_to_guest" default:"false"`               // 是否向访客页面发送 IP 地址，默认 false
+	EulaAccepted           bool   `json:"eula_accepted" default:"false"`
+	BaseScriptsURLKey      string `json:"base_scripts_url" default:""`
 	// GeoIP 配置
 	GeoIpEnabled  bool   `json:"geo_ip_enabled" default:"true"`
 	GeoIpProvider string `json:"geo_ip_provider" default:"ipinfo"` // empty, mmdb, ip-api, geojs
@@ -50,7 +50,7 @@ type Settings struct {
 const (
 	SitenameKey                   = "sitename"
 	DescriptionKey                = "description"
-	AllowCorsKey                  = "allow_cors"
+	CorsOriginCheckEnabledKey     = "cors_origin_check_enabled"
 	CorsAllowedOriginsKey         = "cors_allowed_origins"
 	WsOriginCheckEnabledKey       = "ws_origin_check_enabled"
 	WsAllowedOriginsKey           = "ws_allowed_origins"

pkg/metric/aggregate.go

@@ -0,0 +1,270 @@
+package metric
+
+import (
+	"fmt"
+	"math"
+	"sort"
+	"time"
+)
+
+// AggregatePoints groups raw points into time buckets and computes aggregate values.
+//
+// AggregatePoints 在内存中按查询配置将原始点分桶并聚合。
+func AggregatePoints(points []Point, query AggregateQuery) ([]AggregatePoint, error) {
+	if err := query.Validate(); err != nil {
+		return nil, err
+	}
+	if len(points) == 0 {
+		if query.FillEmpty {
+			return emptyBuckets(query), nil
+		}
+		return []AggregatePoint{}, nil
+	}
+
+	points = sortedPoints(points)
+	groups := make(map[int64][]Point)
+	for _, point := range points {
+		bucket := alignTime(point.Timestamp, query.Interval).UnixNano()
+		groups[bucket] = append(groups[bucket], point)
+	}
+
+	var buckets []int64
+	if query.FillEmpty {
+		for t := alignTime(query.Start, query.Interval); !t.After(query.End); t = t.Add(query.Interval) {
+			buckets = append(buckets, t.UnixNano())
+		}
+	} else {
+		for bucket := range groups {
+			buckets = append(buckets, bucket)
+		}
+		sort.Slice(buckets, func(i, j int) bool { return buckets[i] < buckets[j] })
+	}
+
+	out := make([]AggregatePoint, 0, len(buckets))
+	for _, bucket := range buckets {
+		group := groups[bucket]
+		value, err := aggregateValue(group, query.Aggregation)
+		if err != nil {
+			return nil, err
+		}
+		out = append(out, AggregatePoint{
+			MetricName: query.MetricName,
+			EntityID:   query.EntityID,
+			Bucket:     time.Unix(0, bucket).UTC(),
+			Value:      value,
+			Count:      len(group),
+		})
+	}
+	return out, nil
+}
+
+// CalculateStats computes summary statistics for a point series.
+//
+// CalculateStats 基于一组点计算统计摘要，包括均值、百分位、首尾值和标准差。
+func CalculateStats(points []Point) (Stats, error) {
+	if len(points) == 0 {
+		// Distinguish "the metric/range yielded no samples" from "the metric
+		// definition does not exist" (which surfaces as ErrNotFound elsewhere).
+		return Stats{}, ErrNoData
+	}
+	points = sortedPoints(points)
+	avg, _ := aggregateValue(points, AggAvg)
+	min, _ := aggregateValue(points, AggMin)
+	max, _ := aggregateValue(points, AggMax)
+	sum, _ := aggregateValue(points, AggSum)
+	p50, _ := aggregateValue(points, AggP50)
+	p95, _ := aggregateValue(points, AggP95)
+	p99, _ := aggregateValue(points, AggP99)
+	first, _ := aggregateValue(points, AggFirst)
+	last, _ := aggregateValue(points, AggLast)
+	rate, _ := aggregateValue(points, AggRate)
+
+	var variance float64
+	for _, p := range points {
+		diff := p.Value - avg
+		variance += diff * diff
+	}
+
+	return Stats{
+		Count:  len(points),
+		Min:    min,
+		Max:    max,
+		Avg:    avg,
+		Sum:    sum,
+		P50:    p50,
+		P95:    p95,
+		P99:    p99,
+		First:  first,
+		Last:   last,
+		Rate:   rate,
+		Start:  points[0].Timestamp,
+		End:    points[len(points)-1].Timestamp,
+		StdDev: math.Sqrt(variance / float64(len(points))),
+	}, nil
+}
+
+// aggregateValue computes one aggregation over a point group.
+//
+// aggregateValue 对一组点执行单个聚合类型的计算。
+func aggregateValue(points []Point, agg Aggregation) (float64, error) {
+	if len(points) == 0 {
+		return 0, nil
+	}
+	switch agg {
+	case AggAvg:
+		sum, _ := aggregateValue(points, AggSum)
+		return sum / float64(len(points)), nil
+	case AggMin:
+		v := points[0].Value
+		for _, p := range points[1:] {
+			v = math.Min(v, p.Value)
+		}
+		return v, nil
+	case AggMax:
+		v := points[0].Value
+		for _, p := range points[1:] {
+			v = math.Max(v, p.Value)
+		}
+		return v, nil
+	case AggSum:
+		var sum float64
+		for _, p := range points {
+			sum += p.Value
+		}
+		return sum, nil
+	case AggCount:
+		return float64(len(points)), nil
+	case AggFirst:
+		// Callers (AggregatePoints, CalculateStats) pass time-ordered slices.
+		return points[0].Value, nil
+	case AggLast:
+		return points[len(points)-1].Value, nil
+	case AggRate:
+		return counterRate(points), nil
+	case AggStdDev:
+		return stdDevPop(points), nil
+	default:
+		// Any percentile (p50, p95, p99, and arbitrary pXX / pXX.X) is computed
+		// here via linear interpolation over the sorted values. The fixed
+		// AggP50/AggP95/AggP99 constants are just common cases of this.
+		if frac, ok := parsePercentile(agg); ok {
+			return percentile(points, frac), nil
+		}
+		return 0, fmt.Errorf("%w: unsupported aggregation %q", ErrInvalidArgument, agg)
+	}
+}
+
+// counterRate computes a per-second rate of change that is resilient to counter
+// resets. It walks the time-ordered series and sums only positive deltas; a
+// decrease is treated as a reset (the counter restarted) and contributes zero
+// rather than a negative spike. For a strictly increasing counter this equals
+// (last-first)/seconds. For a gauge it yields the total upward movement per
+// second, which is a stable definition for an otherwise ill-defined quantity.
+//
+// counterRate 计算能抵抗计数器重置的每秒变化率。它会遍历按时间排序的序列，
+// 并且只累加正向增量；当值下降时会被视为重置（计数器重新开始），该段贡献为零，
+// 而不是产生负向尖峰。对于严格递增的计数器，这等于 (last-first)/seconds。
+// 对于 gauge，它表示每秒总上升量，为这种本来定义不明确的量提供稳定定义。
+func counterRate(points []Point) float64 {
+	if len(points) < 2 {
+		return 0
+	}
+	// Input is time-ordered by the callers; avoid an extra sort here.
+	seconds := points[len(points)-1].Timestamp.Sub(points[0].Timestamp).Seconds()
+	if seconds <= 0 {
+		return 0
+	}
+	var increase float64
+	for i := 1; i < len(points); i++ {
+		delta := points[i].Value - points[i-1].Value
+		if delta > 0 {
+			increase += delta
+		}
+	}
+	return increase / seconds
+}
+
+// stdDevPop computes the population standard deviation (dividing by N), matching
+// SQL STDDEV_POP and the StdDev field produced by CalculateStats.
+//
+// stdDevPop 计算总体标准差（除以 N），与 SQL 的 STDDEV_POP 以及
+// CalculateStats 返回的 StdDev 语义保持一致。
+func stdDevPop(points []Point) float64 {
+	if len(points) == 0 {
+		return 0
+	}
+	var sum float64
+	for _, p := range points {
+		sum += p.Value
+	}
+	mean := sum / float64(len(points))
+	var variance float64
+	for _, p := range points {
+		d := p.Value - mean
+		variance += d * d
+	}
+	return math.Sqrt(variance / float64(len(points)))
+}
+
+// percentile computes a quantile from point values.
+//
+// percentile 提取点值、排序，并计算指定小数形式的百分位。
+func percentile(points []Point, p float64) float64 {
+	values := make([]float64, len(points))
+	for i, point := range points {
+		values[i] = point.Value
+	}
+	sort.Float64s(values)
+	return percentileSorted(values, p)
+}
+
+// percentileSorted returns the linear-interpolation percentile of an
+// already-sorted slice. Shared by percentile() and the raw-value paths so the
+// interpolation method stays identical everywhere.
+//
+// percentileSorted 基于已排序切片用线性插值计算百分位，供原始值路径和
+// percentile 共用，确保所有路径的插值方法一致。
+func percentileSorted(values []float64, p float64) float64 {
+	if len(values) == 0 {
+		return 0
+	}
+	if len(values) == 1 {
+		return values[0]
+	}
+	pos := p * float64(len(values)-1)
+	lower := int(math.Floor(pos))
+	upper := int(math.Ceil(pos))
+	if lower == upper {
+		return values[lower]
+	}
+	weight := pos - float64(lower)
+	return values[lower]*(1-weight) + values[upper]*weight
+}
+
+// alignTime floors a timestamp to the start of its interval bucket.
+//
+// alignTime 将时间向下对齐到指定间隔的桶起点。
+func alignTime(t time.Time, interval time.Duration) time.Time {
+	nano := t.UTC().UnixNano()
+	size := interval.Nanoseconds()
+	// Floor division toward negative infinity so timestamps before the Unix
+	// epoch (negative nanos) align to the bucket start rather than rounding up.
+	// Go's % returns a remainder with the dividend's sign, so normalize it.
+	rem := ((nano % size) + size) % size
+	return time.Unix(0, nano-rem).UTC()
+}
+
+// emptyBuckets builds zero-count aggregate buckets for an empty range.
+//
+// emptyBuckets 根据查询范围生成空聚合桶，用于 FillEmpty 场景。
+func emptyBuckets(query AggregateQuery) []AggregatePoint {
+	var out []AggregatePoint
+	for t := alignTime(query.Start, query.Interval); !t.After(query.End); t = t.Add(query.Interval) {
+		out = append(out, AggregatePoint{
+			MetricName: query.MetricName,
+			EntityID:   query.EntityID,
+			Bucket:     t,
+		})
+	}
+	return out
+}

pkg/metric/config.go

@@ -0,0 +1,485 @@
+package metric
+
+import (
+	"database/sql"
+	"fmt"
+	"net/url"
+	"path/filepath"
+	"strings"
+	"time"
+)
+
+// Config describes database, pooling, migration, retention, and rollup settings.
+//
+// Config 描述 Store 的数据库后端、连接池、迁移、保留和 rollup 配置。
+type Config struct {
+	// Driver selects the database backend.
+	//
+	// Driver 选择数据库后端。
+	Driver Driver
+	// DSN is the database connection string when DB is not supplied.
+	//
+	// DSN 是未传入 DB 时使用的数据库连接字符串。
+	DSN string
+
+	// DB can be supplied by the host app when it owns the connection pool.
+	// When DB is set, DSN is ignored and Close leaves the supplied DB open.
+	//
+	// 当宿主程序自己管理连接池时可以传入 DB；设置 DB 后会忽略 DSN，
+	// Close 也不会关闭这个外部传入的连接池。
+	DB *sql.DB
+
+	// TablePrefix prefixes every table managed by the store.
+	//
+	// TablePrefix 是 Store 管理的所有表名前缀。
+	TablePrefix string
+	// DefaultRetentionDays is used when a metric definition omits retention.
+	//
+	// DefaultRetentionDays 在指标定义未设置保留时间时使用。
+	DefaultRetentionDays int
+	// AutoMigrate controls whether Open creates or updates the schema.
+	//
+	// AutoMigrate 控制 Open 是否创建或更新表结构。
+	AutoMigrate bool
+	// MaxOpenConns sets the primary pool's maximum open connections.
+	//
+	// MaxOpenConns 设置主连接池的最大打开连接数。
+	MaxOpenConns int
+	// MaxIdleConns sets the primary pool's maximum idle connections.
+	//
+	// MaxIdleConns 设置主连接池的最大空闲连接数。
+	MaxIdleConns int
+	// ConnMaxLifetime sets the maximum lifetime for pooled connections.
+	//
+	// ConnMaxLifetime 设置池化连接的最大生命周期。
+	ConnMaxLifetime time.Duration
+	// ConnectTimeout bounds the initial database ping in Open.
+	//
+	// ConnectTimeout 限制 Open 中首次 ping 数据库的时间。
+	ConnectTimeout time.Duration
+	// SQLite holds SQLite-specific options.
+	//
+	// SQLite 保存 SQLite 专用选项。
+	SQLite SQLiteOptions
+
+	// RollupPolicy configures downsampling tiers and tiered retention. When it
+	// defines tiers, Store.Compact materializes them and enforces every
+	// retention window. The zero value disables rollups (raw points only).
+	//
+	// RollupPolicy 配置降采样层级和分层保留时间；定义层级后，
+	// Store.Compact 会生成 rollup 并执行保留策略。零值表示禁用 rollup。
+	RollupPolicy RollupPolicy
+}
+
+// SQLiteOptions contains SQLite-specific performance and read-concurrency settings.
+//
+// SQLiteOptions 保存 SQLite 专用的性能和并发读取选项。
+type SQLiteOptions struct {
+	// PerformanceProfile selects the SQLite synchronous durability preset.
+	//
+	// PerformanceProfile 选择 SQLite synchronous 持久化预设。
+	PerformanceProfile SQLitePerformanceProfile
+	// BusyTimeout is applied to SQLite busy_timeout.
+	//
+	// BusyTimeout 会应用到 SQLite busy_timeout。
+	BusyTimeout time.Duration
+	// CacheSizeKB sets the SQLite page cache size in KB.
+	//
+	// CacheSizeKB 设置 SQLite 页缓存大小，单位为 KB。
+	CacheSizeKB int
+	// PageSize sets SQLite page_size when positive.
+	//
+	// PageSize 为正数时设置 SQLite page_size。
+	PageSize int
+	// TempStoreMemory enables memory-backed temporary storage.
+	//
+	// TempStoreMemory 启用基于内存的临时存储。
+	TempStoreMemory bool
+	// MMapSizeBytes sets SQLite mmap_size in bytes.
+	//
+	// MMapSizeBytes 设置 SQLite mmap_size，单位为字节。
+	MMapSizeBytes int64
+	// WALAutoCheckpoint sets the SQLite WAL auto-checkpoint page count.
+	//
+	// WALAutoCheckpoint 设置 SQLite WAL 自动 checkpoint 页数。
+	WALAutoCheckpoint int
+	// ReadPoolSize, when > 0, opens a second read-only connection pool with
+	// this many connections for SELECT-style calls. SQLite serializes writes,
+	// but WAL lets readers run concurrently, so a read pool lifts read
+	// throughput while keeping writes on the single primary connection.
+	// 0 (default) means all calls share the single primary pool, preserving
+	// the previous single-connection behavior.
+	//
+	// ReadPoolSize 大于 0 时会为 SELECT 类调用打开独立只读连接池。
+	// SQLite 写入会被串行化，但 WAL 允许读取并发执行，因此读池可以在写入仍走
+	// 单主连接的同时提升读取吞吐。0（默认值）表示所有调用共享单个主连接池，
+	// 保留之前的单连接行为。
+	ReadPoolSize int
+}
+
+// SQLitePerformanceProfile names SQLite durability and performance presets.
+//
+// SQLitePerformanceProfile 表示 SQLite synchronous 等持久化策略预设。
+type SQLitePerformanceProfile string
+
+const (
+	// SQLiteProfileDefault uses the package default SQLite profile.
+	//
+	// SQLiteProfileDefault 使用包默认的 SQLite 配置档。
+	SQLiteProfileDefault SQLitePerformanceProfile = ""
+	// SQLiteProfileBalanced uses NORMAL synchronous mode.
+	//
+	// SQLiteProfileBalanced 使用 NORMAL synchronous 模式。
+	SQLiteProfileBalanced SQLitePerformanceProfile = "balanced"
+	// SQLiteProfilePerformance favors throughput over durability.
+	//
+	// SQLiteProfilePerformance 优先吞吐而非持久性。
+	SQLiteProfilePerformance SQLitePerformanceProfile = "performance"
+	// SQLiteProfileDurable favors durability over write throughput.
+	//
+	// SQLiteProfileDurable 优先持久性而非写入吞吐。
+	SQLiteProfileDurable SQLitePerformanceProfile = "durable"
+)
+
+// Option mutates a Config value.
+//
+// Option 是用于修改 Config 的函数式选项。
+type Option func(*Config)
+
+// DefaultConfig returns the default configuration for a backend.
+//
+// DefaultConfig 返回指定后端的默认配置。
+func DefaultConfig(driver Driver, dsn string) Config {
+	return Config{
+		Driver:               driver,
+		DSN:                  dsn,
+		TablePrefix:          "metric_",
+		DefaultRetentionDays: 90,
+		AutoMigrate:          true,
+		MaxOpenConns:         25,
+		MaxIdleConns:         5,
+		ConnMaxLifetime:      time.Hour,
+		ConnectTimeout:       10 * time.Second,
+		SQLite: SQLiteOptions{
+			PerformanceProfile: SQLiteProfileBalanced,
+			BusyTimeout:        5 * time.Second,
+			CacheSizeKB:        64 * 1024,
+			TempStoreMemory:    true,
+			MMapSizeBytes:      256 * 1024 * 1024,
+			WALAutoCheckpoint:  1000,
+		},
+	}
+}
+
+// Backend builds a generic backend configuration and applies options.
+//
+// Backend 构造通用后端配置，并应用额外选项。
+func Backend(driver Driver, dsn string, opts ...Option) Config {
+	cfg := DefaultConfig(driver, dsn)
+	applyOptions(&cfg, opts...)
+	return cfg
+}
+
+// SQLite builds a SQLite backend configuration.
+//
+// SQLite 构造 SQLite 后端配置。
+func SQLite(dsn string, opts ...Option) Config {
+	cfg := SQLiteConfig(dsn)
+	applyOptions(&cfg, opts...)
+	return cfg
+}
+
+// SQLiteInDir builds a managed SQLite file configuration rooted at a directory.
+//
+// SQLiteInDir 构造由 package 管理目录的 SQLite 文件数据库配置。
+func SQLiteInDir(dir string, opts ...Option) Config {
+	cfg := SQLiteConfig(sqliteFileDSN(filepath.Join(dir, "metrics.db")))
+	cfg.MaxOpenConns = 1
+	cfg.MaxIdleConns = 1
+	applyOptions(&cfg, opts...)
+	return cfg
+}
+
+// MySQL builds a MySQL backend configuration.
+//
+// MySQL 构造 MySQL 后端配置。
+func MySQL(dsn string, opts ...Option) Config {
+	cfg := MySQLConfig(dsn)
+	applyOptions(&cfg, opts...)
+	return cfg
+}
+
+// PostgreSQL builds a PostgreSQL backend configuration.
+//
+// PostgreSQL 构造 PostgreSQL 后端配置。
+func PostgreSQL(dsn string, opts ...Option) Config {
+	cfg := PostgreSQLConfig(dsn)
+	applyOptions(&cfg, opts...)
+	return cfg
+}
+
+// SQLiteConfig returns the default SQLite configuration.
+//
+// SQLiteConfig 返回 SQLite 后端的默认配置。
+func SQLiteConfig(dsn string) Config {
+	cfg := DefaultConfig(DriverSQLite, dsn)
+	if cfg.DSN == "" {
+		cfg.DSN = "file:metric?mode=memory&cache=shared"
+	}
+	// SQLite serializes writes, so a single connection is the safe default and
+	// avoids "database is locked" contention. Callers can still override with
+	// WithMaxOpenConns/WithMaxIdleConns. Setting it here (rather than inferring
+	// it later from the 25/5 defaults) means an explicit WithMaxOpenConns(25)
+	// is honored instead of being mistaken for "unset".
+	cfg.MaxOpenConns = 1
+	cfg.MaxIdleConns = 1
+	return cfg
+}
+
+// MySQLConfig returns the default MySQL configuration.
+//
+// MySQLConfig 返回 MySQL 后端的默认配置。
+func MySQLConfig(dsn string) Config {
+	return DefaultConfig(DriverMySQL, dsn)
+}
+
+// PostgreSQLConfig returns the default PostgreSQL configuration.
+//
+// PostgreSQLConfig 返回 PostgreSQL 后端的默认配置。
+func PostgreSQLConfig(dsn string) Config {
+	return DefaultConfig(DriverPostgreSQL, dsn)
+}
+
+// WithDB sets a caller-owned database connection pool.
+//
+// WithDB 设置由调用方提供的数据库连接池。
+func WithDB(db *sql.DB) Option {
+	return func(c *Config) {
+		c.DB = db
+	}
+}
+
+// WithTablePrefix sets the table-name prefix used by metric tables.
+//
+// WithTablePrefix 设置 metric 私有表的表名前缀。
+func WithTablePrefix(prefix string) Option {
+	return func(c *Config) {
+		c.TablePrefix = prefix
+	}
+}
+
+// WithDefaultRetention sets the default metric retention in days.
+//
+// WithDefaultRetention 设置指标默认保留天数。
+func WithDefaultRetention(days int) Option {
+	return func(c *Config) {
+		c.DefaultRetentionDays = days
+	}
+}
+
+// WithAutoMigrate controls whether Open creates the schema automatically.
+//
+// WithAutoMigrate 控制 Open 时是否自动创建表结构。
+func WithAutoMigrate(enabled bool) Option {
+	return func(c *Config) {
+		c.AutoMigrate = enabled
+	}
+}
+
+// WithMaxOpenConns sets the maximum number of open database connections.
+//
+// WithMaxOpenConns 设置底层数据库连接池的最大打开连接数。
+func WithMaxOpenConns(n int) Option {
+	return func(c *Config) {
+		c.MaxOpenConns = n
+	}
+}
+
+// WithMaxIdleConns sets the maximum number of idle database connections.
+//
+// WithMaxIdleConns 设置底层数据库连接池的最大空闲连接数。
+func WithMaxIdleConns(n int) Option {
+	return func(c *Config) {
+		c.MaxIdleConns = n
+	}
+}
+
+// WithConnMaxLifetime sets the maximum lifetime for pooled connections.
+//
+// WithConnMaxLifetime 设置连接最大复用时间。
+func WithConnMaxLifetime(d time.Duration) Option {
+	return func(c *Config) {
+		c.ConnMaxLifetime = d
+	}
+}
+
+// WithConnectTimeout sets the timeout used while opening the store.
+//
+// WithConnectTimeout 设置打开 Store 时 ping 数据库的超时时间。
+func WithConnectTimeout(d time.Duration) Option {
+	return func(c *Config) {
+		c.ConnectTimeout = d
+	}
+}
+
+// WithSQLiteProfile sets the SQLite durability and performance profile.
+//
+// WithSQLiteProfile 设置 SQLite 持久化性能预设。
+func WithSQLiteProfile(profile SQLitePerformanceProfile) Option {
+	return func(c *Config) {
+		c.SQLite.PerformanceProfile = profile
+	}
+}
+
+// WithSQLiteBusyTimeout sets SQLite busy_timeout.
+//
+// WithSQLiteBusyTimeout 设置 SQLite busy_timeout。
+func WithSQLiteBusyTimeout(d time.Duration) Option {
+	return func(c *Config) {
+		c.SQLite.BusyTimeout = d
+	}
+}
+
+// WithSQLiteCacheSizeKB sets the SQLite page cache size in KB.
+//
+// WithSQLiteCacheSizeKB 设置 SQLite 页缓存大小，单位为 KB。
+func WithSQLiteCacheSizeKB(kb int) Option {
+	return func(c *Config) {
+		c.SQLite.CacheSizeKB = kb
+	}
+}
+
+// WithSQLiteMMapSize sets SQLite mmap_size.
+//
+// WithSQLiteMMapSize 设置 SQLite mmap_size。
+func WithSQLiteMMapSize(bytes int64) Option {
+	return func(c *Config) {
+		c.SQLite.MMapSizeBytes = bytes
+	}
+}
+
+// WithSQLitePageSize sets SQLite page_size.
+//
+// WithSQLitePageSize 设置 SQLite page_size。
+func WithSQLitePageSize(bytes int) Option {
+	return func(c *Config) {
+		c.SQLite.PageSize = bytes
+	}
+}
+
+// WithSQLiteTempStoreMemory controls whether SQLite uses memory for temporary storage.
+//
+// WithSQLiteTempStoreMemory 设置 SQLite 是否使用内存临时存储。
+func WithSQLiteTempStoreMemory(enabled bool) Option {
+	return func(c *Config) {
+		c.SQLite.TempStoreMemory = enabled
+	}
+}
+
+// WithSQLiteWALAutoCheckpoint sets the SQLite WAL auto-checkpoint page count.
+//
+// WithSQLiteWALAutoCheckpoint 设置 SQLite WAL 自动 checkpoint 页数。
+func WithSQLiteWALAutoCheckpoint(pages int) Option {
+	return func(c *Config) {
+		c.SQLite.WALAutoCheckpoint = pages
+	}
+}
+
+// WithSQLiteReadPool enables a dedicated read-only connection pool of n
+// connections for SQLite. Writes stay on the single primary connection
+// (SQLite serializes them); reads fan out across the pool, which WAL mode
+// allows to run concurrently. Pass n <= 1 to disable (the default).
+//
+// WithSQLiteReadPool 为 SQLite 启用包含 n 个连接的独立只读连接池。写入仍走
+// 单主连接（SQLite 会串行化写入）；读取会分散到该连接池中，WAL 模式允许它们
+// 并发执行。传入 n <= 1 会禁用该功能（默认行为）。
+func WithSQLiteReadPool(n int) Option {
+	return func(c *Config) {
+		c.SQLite.ReadPoolSize = n
+	}
+}
+
+// WithRollupPolicy sets the downsampling/retention ladder. Compact uses it to
+// build rollup tiers (each progressively coarser and longer-lived) and to age
+// out raw points and expired tiers.
+//
+// WithRollupPolicy 设置降采样与保留时间阶梯；Compact 会据此构建逐级
+// 更粗、保留更久的 rollup，并清理过期原始点和过期层级。
+func WithRollupPolicy(p RollupPolicy) Option {
+	return func(c *Config) {
+		c.RollupPolicy = p
+	}
+}
+
+// applyOptions applies configuration options in order.
+//
+// applyOptions 将一组选项依次应用到 Config。
+func applyOptions(cfg *Config, opts ...Option) {
+	for _, opt := range opts {
+		if opt != nil {
+			opt(cfg)
+		}
+	}
+}
+
+// Validate checks whether the value is well formed.
+//
+// Validate 检查 Config 的后端、表名前缀、保留时间和 rollup 策略是否合法。
+func (c Config) Validate() error {
+	switch c.Driver {
+	case DriverSQLite, DriverMySQL, DriverPostgreSQL:
+	default:
+		return fmt.Errorf("%w: unsupported driver %q", ErrInvalidArgument, c.Driver)
+	}
+	if c.DB == nil && strings.TrimSpace(c.DSN) == "" {
+		return fmt.Errorf("%w: dsn is required when db is not supplied", ErrInvalidArgument)
+	}
+	if c.TablePrefix == "" {
+		return fmt.Errorf("%w: table prefix cannot be empty", ErrInvalidArgument)
+	}
+	for _, r := range c.TablePrefix {
+		if !(r == '_' || r >= '0' && r <= '9' || r >= 'A' && r <= 'Z' || r >= 'a' && r <= 'z') {
+			return fmt.Errorf("%w: table prefix must contain only letters, digits, and underscores", ErrInvalidArgument)
+		}
+	}
+	if c.DefaultRetentionDays <= 0 {
+		return fmt.Errorf("%w: default retention days must be positive", ErrInvalidArgument)
+	}
+	if err := c.RollupPolicy.Validate(); err != nil {
+		return err
+	}
+	return nil
+}
+
+// driverName returns the database/sql driver name for the configured backend.
+//
+// driverName 返回 database/sql 注册使用的驱动名称。
+func (c Config) driverName() string {
+	switch c.Driver {
+	case DriverSQLite:
+		return "sqlite3"
+	case DriverMySQL:
+		return "mysql"
+	case DriverPostgreSQL:
+		return "pgx"
+	default:
+		return string(c.Driver)
+	}
+}
+
+// sqliteFileDSN converts a filesystem path into a SQLite file DSN.
+//
+// sqliteFileDSN 将文件路径转换为 SQLite file: DSN。
+func sqliteFileDSN(path string) string {
+	return "file:" + filepath.ToSlash(path) + "?cache=shared&mode=rwc"
+}
+
+// appendSQLiteDSNParam appends a query parameter to a SQLite DSN.
+//
+// appendSQLiteDSNParam 向 SQLite DSN 追加查询参数。
+func appendSQLiteDSNParam(dsn, key, value string) string {
+	if strings.Contains(dsn, "?") {
+		return dsn + "&" + url.QueryEscape(key) + "=" + url.QueryEscape(value)
+	}
+	return dsn + "?" + url.QueryEscape(key) + "=" + url.QueryEscape(value)
+}

pkg/metric/config_test.go

@@ -0,0 +1,68 @@
+package metric
+
+import (
+	"testing"
+	"time"
+)
+
+// TestBackendBuildersApplyOptions verifies backend builders apply options.
+//
+// TestBackendBuildersApplyOptions 验证各后端构造器会正确应用选项。
+func TestBackendBuildersApplyOptions(t *testing.T) {
+	cfg := SQLite(
+		"file:metrics.db?cache=shared",
+		WithTablePrefix("x_metric_"),
+		WithDefaultRetention(180),
+		WithAutoMigrate(false),
+		WithMaxOpenConns(10),
+		WithMaxIdleConns(2),
+		WithConnMaxLifetime(30*time.Minute),
+		WithConnectTimeout(2*time.Second),
+	)
+
+	if cfg.Driver != DriverSQLite {
+		t.Fatalf("expected sqlite driver, got %q", cfg.Driver)
+	}
+	if cfg.TablePrefix != "x_metric_" {
+		t.Fatalf("table prefix option was not applied: %q", cfg.TablePrefix)
+	}
+	if cfg.DefaultRetentionDays != 180 {
+		t.Fatalf("default retention option was not applied: %d", cfg.DefaultRetentionDays)
+	}
+	if cfg.AutoMigrate {
+		t.Fatalf("auto migrate option was not applied")
+	}
+	if cfg.MaxOpenConns != 10 || cfg.MaxIdleConns != 2 {
+		t.Fatalf("pool options were not applied: %#v", cfg)
+	}
+	if cfg.ConnMaxLifetime != 30*time.Minute || cfg.ConnectTimeout != 2*time.Second {
+		t.Fatalf("timeout options were not applied: %#v", cfg)
+	}
+}
+
+// TestSQLiteInDirBuildsFileConfig verifies SQLiteInDir file configuration.
+//
+// TestSQLiteInDirBuildsFileConfig 验证 SQLiteInDir 会生成预期的文件数据库配置。
+func TestSQLiteInDirBuildsFileConfig(t *testing.T) {
+	cfg := SQLiteInDir(
+		"data/metrics",
+		WithSQLiteProfile(SQLiteProfilePerformance),
+		WithSQLiteCacheSizeKB(128*1024),
+	)
+
+	if cfg.Driver != DriverSQLite {
+		t.Fatalf("expected sqlite driver, got %q", cfg.Driver)
+	}
+	if cfg.MaxOpenConns != 1 || cfg.MaxIdleConns != 1 {
+		t.Fatalf("sqlite directory config should default to a single connection: %#v", cfg)
+	}
+	if cfg.SQLite.PerformanceProfile != SQLiteProfilePerformance {
+		t.Fatalf("sqlite profile option was not applied: %q", cfg.SQLite.PerformanceProfile)
+	}
+	if cfg.SQLite.CacheSizeKB != 128*1024 {
+		t.Fatalf("sqlite cache option was not applied: %d", cfg.SQLite.CacheSizeKB)
+	}
+	if cfg.DSN != "file:data/metrics/metrics.db?cache=shared&mode=rwc" {
+		t.Fatalf("unexpected sqlite dir dsn: %q", cfg.DSN)
+	}
+}

pkg/metric/dialect.go

@@ -0,0 +1,444 @@
+package metric
+
+import (
+	"database/sql"
+	"fmt"
+	"strings"
+)
+
+// dialect abstracts SQL differences between supported databases.
+//
+// dialect 抽象不同数据库后端的 SQL 差异。
+type dialect interface {
+	placeholder(n int) string
+	// jsonPlaceholder returns the bind placeholder for a JSON column value.
+	// PostgreSQL needs an explicit ::jsonb cast because the driver sends Go
+	// strings with a text OID, which the server refuses to assign to a jsonb
+	// column. SQLite and MySQL store JSON as text/native and need no cast.
+	//
+	// jsonPlaceholder 返回 JSON 列值的绑定占位符。PostgreSQL 需要显式的
+	// ::jsonb 转换，因为驱动会以 text OID 发送 Go 字符串，服务器拒绝把它赋给
+	// jsonb 列。SQLite 和 MySQL 将 JSON 存为文本或原生类型，不需要转换。
+	jsonPlaceholder(n int) string
+	jsonType() string
+	autoIncrementPrimaryKey() string
+	nowExpr() string
+	insertDefinitionSQL(t tables) string
+	upsertPointSQL(t tables, rowCount int) string
+	// jsonExtractEquals renders a boolean predicate comparing the text value at
+	// key inside a JSON column to a bind placeholder. The key is interpolated
+	// into the SQL (callers must restrict it to a safe charset); the compared
+	// value is always bound.
+	//
+	// jsonExtractEquals 渲染一个布尔谓词，用于比较 JSON 列内 key 对应的文本值和
+	// 绑定占位符。key 会插入 SQL 中（调用方必须限制到安全字符集）；被比较的值
+	// 始终通过绑定参数传入。
+	jsonExtractEquals(column, key, placeholder string) string
+	// blobType is the column type for the rollup t-digest sketch.
+	//
+	// blobType 是 rollup t-digest sketch 的列类型。
+	blobType() string
+	// upsertRollupSQL builds a single-row upsert for one rollup bucket cell.
+	//
+	// upsertRollupSQL 为单个 rollup 桶单元构造单行 upsert。
+	upsertRollupSQL(t tables) string
+	// compactTxOptions returns the transaction options used for the compaction
+	// transaction. PostgreSQL/MySQL escalate to SERIALIZABLE so the raw scan and
+	// the raw deletion share one snapshot and a point inserted between them
+	// cannot be deleted without first being rolled up. SQLite returns nil
+	// (its default) because a single connection already serializes writes.
+	//
+	// compactTxOptions 返回 compaction 事务使用的事务选项。PostgreSQL/MySQL 会
+	// 提升到 SERIALIZABLE，使 raw 扫描和 raw 删除共享同一个快照，让两者之间写入
+	// 的点不会在尚未进入 rollup 前被删除。SQLite 返回 nil（默认），因为单连接
+	// 已经串行化写入。
+	compactTxOptions() *sql.TxOptions
+}
+
+// tables stores the physical table names used by a Store.
+//
+// tables 保存 Store 使用的实际表名。
+type tables struct {
+	// definitions is the metric definitions table.
+	//
+	// definitions 是指标定义表。
+	definitions string
+	// points is the raw points table.
+	//
+	// points 是原始采样点表。
+	points string
+	// rollups is the downsampled rollups table.
+	//
+	// rollups 是降采样 rollup 表。
+	rollups string
+}
+
+// newDialect returns the SQL dialect implementation for a backend.
+//
+// newDialect 根据数据库后端创建对应 SQL 方言实现。
+func newDialect(driver Driver) dialect {
+	switch driver {
+	case DriverPostgreSQL:
+		return postgresDialect{}
+	case DriverMySQL:
+		return mysqlDialect{}
+	default:
+		return sqliteDialect{}
+	}
+}
+
+// sqliteDialect implements SQL rendering for SQLite.
+//
+// sqliteDialect 实现 SQLite 方言。
+type sqliteDialect struct{}
+
+// placeholder returns a bind placeholder for this SQL dialect.
+//
+// placeholder 返回 SQLite 的绑定参数占位符。
+func (sqliteDialect) placeholder(int) string { return "?" }
+
+// jsonPlaceholder returns a bind placeholder for JSON values.
+//
+// jsonPlaceholder 返回 SQLite JSON 值的绑定参数占位符。
+func (sqliteDialect) jsonPlaceholder(int) string { return "?" }
+
+// jsonType returns the SQL column type used for JSON values.
+//
+// jsonType 返回 SQLite 使用的 JSON 存储类型。
+func (sqliteDialect) jsonType() string { return "TEXT" }
+
+// autoIncrementPrimaryKey returns the SQL definition for an auto-incrementing primary key.
+//
+// autoIncrementPrimaryKey 返回 SQLite 自增主键定义。
+func (sqliteDialect) autoIncrementPrimaryKey() string { return "INTEGER PRIMARY KEY AUTOINCREMENT" }
+
+// nowExpr returns the SQL expression for the current time.
+//
+// nowExpr 返回 SQLite 当前时间表达式。
+func (sqliteDialect) nowExpr() string { return "CURRENT_TIMESTAMP" }
+
+// insertDefinitionSQL builds SQL for inserting or updating a metric definition.
+//
+// insertDefinitionSQL 构造 SQLite 指标定义 upsert SQL。
+func (sqliteDialect) insertDefinitionSQL(t tables) string {
+	return fmt.Sprintf(`INSERT INTO %s
+		(name, type, unit, description, retention_days, metadata, created_at, updated_at)
+		VALUES (?, ?, ?, ?, ?, ?, ?, ?)
+		ON CONFLICT(name) DO UPDATE SET
+			type = excluded.type,
+			unit = excluded.unit,
+			description = excluded.description,
+			retention_days = excluded.retention_days,
+			metadata = excluded.metadata,
+			updated_at = excluded.updated_at`, t.definitions)
+}
+
+// upsertPointSQL builds SQL for inserting or updating metric points.
+//
+// upsertPointSQL 构造 SQLite 采样点批量 upsert SQL。
+func (sqliteDialect) upsertPointSQL(t tables, rowCount int) string {
+	return buildInsertPointsSQL(t.points, sqliteDialect{}, rowCount, `ON CONFLICT(metric_name, entity_id, tags_hash, ts_nano) DO UPDATE SET
+		value = excluded.value,
+		tags = excluded.tags,
+		labels = excluded.labels,
+		created_at = excluded.created_at`)
+}
+
+// mysqlDialect implements SQL rendering for MySQL.
+//
+// mysqlDialect 实现 MySQL 方言。
+type mysqlDialect struct{}
+
+// placeholder returns a bind placeholder for this SQL dialect.
+//
+// placeholder 返回 MySQL 的绑定参数占位符。
+func (mysqlDialect) placeholder(int) string { return "?" }
+
+// jsonPlaceholder returns a bind placeholder for JSON values.
+//
+// jsonPlaceholder 返回 MySQL JSON 值的绑定参数占位符。
+func (mysqlDialect) jsonPlaceholder(int) string { return "?" }
+
+// jsonType returns the SQL column type used for JSON values.
+//
+// jsonType 返回 MySQL 使用的 JSON 存储类型。
+func (mysqlDialect) jsonType() string { return "JSON" }
+
+// autoIncrementPrimaryKey returns the SQL definition for an auto-incrementing primary key.
+//
+// autoIncrementPrimaryKey 返回 MySQL 自增主键定义。
+func (mysqlDialect) autoIncrementPrimaryKey() string { return "BIGINT AUTO_INCREMENT PRIMARY KEY" }
+
+// nowExpr returns the SQL expression for the current time.
+//
+// nowExpr 返回 MySQL 当前时间表达式。
+func (mysqlDialect) nowExpr() string { return "CURRENT_TIMESTAMP" }
+
+// insertDefinitionSQL builds SQL for inserting or updating a metric definition.
+//
+// insertDefinitionSQL 构造 MySQL 指标定义 upsert SQL。
+func (mysqlDialect) insertDefinitionSQL(t tables) string {
+	return fmt.Sprintf(`INSERT INTO %s
+		(name, type, unit, description, retention_days, metadata, created_at, updated_at)
+		VALUES (?, ?, ?, ?, ?, ?, ?, ?)
+		ON DUPLICATE KEY UPDATE
+			type = VALUES(type),
+			unit = VALUES(unit),
+			description = VALUES(description),
+			retention_days = VALUES(retention_days),
+			metadata = VALUES(metadata),
+			updated_at = VALUES(updated_at)`, t.definitions)
+}
+
+// upsertPointSQL builds SQL for inserting or updating metric points.
+//
+// upsertPointSQL 构造 MySQL 采样点批量 upsert SQL。
+func (mysqlDialect) upsertPointSQL(t tables, rowCount int) string {
+	return buildInsertPointsSQL(t.points, mysqlDialect{}, rowCount, `ON DUPLICATE KEY UPDATE
+		value = VALUES(value),
+		tags = VALUES(tags),
+		labels = VALUES(labels),
+		created_at = VALUES(created_at)`)
+}
+
+// postgresDialect implements SQL rendering for PostgreSQL.
+//
+// postgresDialect 实现 PostgreSQL 方言。
+type postgresDialect struct{}
+
+// placeholder returns a bind placeholder for this SQL dialect.
+//
+// placeholder 返回 PostgreSQL 的编号绑定参数占位符。
+func (postgresDialect) placeholder(n int) string { return fmt.Sprintf("$%d", n) }
+
+// jsonPlaceholder returns a bind placeholder for JSON values.
+//
+// jsonPlaceholder 返回 PostgreSQL JSONB 值的绑定参数占位符。
+func (postgresDialect) jsonPlaceholder(n int) string { return fmt.Sprintf("$%d::jsonb", n) }
+
+// jsonType returns the SQL column type used for JSON values.
+//
+// jsonType 返回 PostgreSQL 使用的 JSONB 存储类型。
+func (postgresDialect) jsonType() string { return "JSONB" }
+
+// autoIncrementPrimaryKey returns the SQL definition for an auto-incrementing primary key.
+//
+// autoIncrementPrimaryKey 返回 PostgreSQL 自增主键定义。
+func (postgresDialect) autoIncrementPrimaryKey() string { return "BIGSERIAL PRIMARY KEY" }
+
+// nowExpr returns the SQL expression for the current time.
+//
+// nowExpr 返回 PostgreSQL 当前时间表达式。
+func (postgresDialect) nowExpr() string { return "CURRENT_TIMESTAMP" }
+
+// insertDefinitionSQL builds SQL for inserting or updating a metric definition.
+//
+// insertDefinitionSQL 构造 PostgreSQL 指标定义 upsert SQL。
+func (postgresDialect) insertDefinitionSQL(t tables) string {
+	return fmt.Sprintf(`INSERT INTO %s
+		(name, type, unit, description, retention_days, metadata, created_at, updated_at)
+		VALUES ($1, $2, $3, $4, $5, $6::jsonb, $7, $8)
+		ON CONFLICT(name) DO UPDATE SET
+			type = EXCLUDED.type,
+			unit = EXCLUDED.unit,
+			description = EXCLUDED.description,
+			retention_days = EXCLUDED.retention_days,
+			metadata = EXCLUDED.metadata,
+			updated_at = EXCLUDED.updated_at`, t.definitions)
+}
+
+// upsertPointSQL builds SQL for inserting or updating metric points.
+//
+// upsertPointSQL 构造 PostgreSQL 采样点批量 upsert SQL。
+func (postgresDialect) upsertPointSQL(t tables, rowCount int) string {
+	return buildInsertPointsSQL(t.points, postgresDialect{}, rowCount, `ON CONFLICT(metric_name, entity_id, tags_hash, ts_nano) DO UPDATE SET
+		value = EXCLUDED.value,
+		tags = EXCLUDED.tags,
+		labels = EXCLUDED.labels,
+		created_at = EXCLUDED.created_at`)
+}
+
+// buildInsertPointsSQL builds a multi-row insert statement for metric points.
+//
+// buildInsertPointsSQL 构造多行采样点 INSERT/UPSERT SQL。
+func buildInsertPointsSQL(table string, d dialect, rowCount int, suffix string) string {
+	var b strings.Builder
+	b.WriteString("INSERT INTO ")
+	b.WriteString(table)
+	b.WriteString(" (metric_name, entity_id, tags_hash, ts_nano, value, tags, labels, created_at) VALUES ")
+	arg := 1
+	rows := make([]string, rowCount)
+	for i := 0; i < rowCount; i++ {
+		parts := make([]string, 8)
+		for j := range parts {
+			// Columns are (metric_name, entity_id, tags_hash, ts_nano, value, tags,
+			// labels, created_at); positions 5 and 6 (0-indexed) are JSON values.
+			// tags_hash makes the tag set part of a point's identity, so points
+			// that share entity+timestamp but differ in tags no longer collide.
+			if j == 5 || j == 6 {
+				parts[j] = d.jsonPlaceholder(arg)
+			} else {
+				parts[j] = d.placeholder(arg)
+			}
+			arg++
+		}
+		rows[i] = "(" + strings.Join(parts, ", ") + ")"
+	}
+	b.WriteString(strings.Join(rows, ", "))
+	if suffix != "" {
+		b.WriteByte(' ')
+		b.WriteString(suffix)
+	}
+	return b.String()
+}
+
+// tableName combines a table prefix and logical table name.
+//
+// tableName 用表名前缀和逻辑表名生成实际表名。
+func tableName(prefix, name string) string {
+	return prefix + name
+}
+
+// insertDefinitionOnlySQL builds a plain INSERT (no upsert clause) for a metric
+// definition. CreateMetric uses it so a duplicate name surfaces as a unique
+// constraint violation instead of silently overwriting the existing row.
+//
+// insertDefinitionOnlySQL 构造普通 INSERT（不带 upsert），供
+// CreateMetric 保持“只创建”语义；重复名称会暴露为唯一约束错误。
+func insertDefinitionOnlySQL(d dialect, t tables) string {
+	cols := "(name, type, unit, description, retention_days, metadata, created_at, updated_at)"
+	ph := []string{
+		d.placeholder(1),
+		d.placeholder(2),
+		d.placeholder(3),
+		d.placeholder(4),
+		d.placeholder(5),
+		d.jsonPlaceholder(6),
+		d.placeholder(7),
+		d.placeholder(8),
+	}
+	return fmt.Sprintf("INSERT INTO %s %s VALUES (%s)", t.definitions, cols, strings.Join(ph, ", "))
+}
+
+// sqlSingleQuote escapes a string for safe inclusion inside a single-quoted SQL
+// string literal by doubling embedded single quotes.
+//
+// sqlSingleQuote 通过把单引号加倍，安全地转义可放入 SQL 单引号字符串
+// 字面量的内容。
+func sqlSingleQuote(s string) string {
+	return strings.ReplaceAll(s, "'", "''")
+}
+
+// sqlJSONPathQuoted builds a quoted JSON path member access ($."key") for the
+// given object key and returns it ready to embed inside a single-quoted SQL
+// string literal. Using the quoted form ($."key") rather than the bare form
+// ($.key) is required for keys that contain dots, hyphens, spaces or other
+// characters the JSON path grammar would otherwise interpret (e.g. a dot is a
+// member separator, so the bare path "$.region.zone" would look up nested
+// member "zone" of object "region" instead of the flat key "region.zone").
+//
+// Escaping is applied in two layers, innermost first:
+//  1. JSON path string escaping inside the double quotes: backslash and double
+//     quote are backslash-escaped.
+//  2. SQL string-literal escaping of the whole path: single quotes doubled.
+//
+// sqlJSONPathQuoted 为给定对象键构造带引号的 JSON path 成员访问（$."key"），
+// 并返回可嵌入 SQL 单引号字符串字面量的结果。必须使用带引号形式（$."key"），
+// 而不是裸形式（$.key），因为包含点号、连字符、空格或其他 JSON path 语法字符的
+// key 会被误解；例如点号是成员分隔符，所以裸路径 "$.region.zone" 会查找对象
+// "region" 的嵌套成员 "zone"，而不是平铺 key "region.zone"。
+//
+// 转义按从内到外两层应用：
+//  1. 双引号内的 JSON path 字符串转义：反斜杠和双引号使用反斜杠转义。
+//  2. 整个 path 的 SQL 字符串字面量转义：单引号加倍。
+func sqlJSONPathQuoted(key string) string {
+	escaped := strings.ReplaceAll(key, "\\", "\\\\")
+	escaped = strings.ReplaceAll(escaped, "\"", "\\\"")
+	return sqlSingleQuote("$.\"" + escaped + "\"")
+}
+
+// jsonExtractEquals renders a JSON equality predicate.
+//
+// jsonExtractEquals 构造 SQLite JSON 字段等值过滤表达式。
+func (sqliteDialect) jsonExtractEquals(column, key, placeholder string) string {
+	return fmt.Sprintf("json_extract(%s, '%s') = %s", column, sqlJSONPathQuoted(key), placeholder)
+}
+
+// jsonExtractEquals renders a JSON equality predicate.
+//
+// jsonExtractEquals 构造 MySQL JSON 字段等值过滤表达式。
+func (mysqlDialect) jsonExtractEquals(column, key, placeholder string) string {
+	return fmt.Sprintf("JSON_UNQUOTE(JSON_EXTRACT(%s, '%s')) = %s", column, sqlJSONPathQuoted(key), placeholder)
+}
+
+// jsonExtractEquals renders a JSON equality predicate.
+//
+// jsonExtractEquals 构造 PostgreSQL JSONB 字段等值过滤表达式。
+func (postgresDialect) jsonExtractEquals(column, key, placeholder string) string {
+	// PostgreSQL's ->> takes the key as a plain text literal (not a JSON path),
+	// so a dot or hyphen in the key is harmless; only single quotes need SQL
+	// escaping.
+	return fmt.Sprintf("%s->>'%s' = %s", column, sqlSingleQuote(key), placeholder)
+}
+
+// sqlAggValueExpr returns the SQL value expression that computes agg over the
+// "value" column for the given driver, and whether that aggregation can be
+// pushed down to this backend. Aggregations that need the ordered raw series
+// (first/last/rate) are never pushed down here. Percentiles and population
+// standard deviation are pushed down only where the backend has a portable,
+// semantics-matching function:
+//
+//   - STDDEV_POP: MySQL and PostgreSQL (matches the in-memory population stddev,
+//     which divides by N). SQLite has no built-in, so it falls back to memory.
+//   - percentile_cont(p) WITHIN GROUP (ORDER BY value): PostgreSQL only. This
+//     matches the in-memory linear-interpolation percentile for ANY p (p50,
+//     p95, p99, p99.9, ...). MySQL lacks a portable continuous-percentile
+//     function and SQLite has none, so both fall back to memory.
+//
+// The simple reductions (avg/min/max/sum/count) are portable everywhere.
+//
+// sqlAggValueExpr 返回给定驱动上对 "value" 列计算 agg 的 SQL 值表达式，
+// 以及该聚合是否能下推到该后端。需要有序原始序列的聚合（first/last/rate）
+// 永远不会在这里下推。百分位和总体标准差只会在后端提供可移植且语义匹配的函数时
+// 下推：
+//
+//   - STDDEV_POP：MySQL 和 PostgreSQL 支持（匹配内存中的总体标准差，
+//     即除以 N）。SQLite 没有内置函数，因此回退到内存计算。
+//   - percentile_cont(p) WITHIN GROUP (ORDER BY value)：仅 PostgreSQL 支持。
+//     它匹配内存中任意 p（p50、p95、p99、p99.9 等）的线性插值百分位。
+//     MySQL 没有可移植的连续百分位函数，SQLite 也没有，因此两者都回退到内存计算。
+//
+// 简单归约（avg/min/max/sum/count）在所有后端都可移植。
+func sqlAggValueExpr(driver Driver, agg Aggregation) (string, bool) {
+	switch agg {
+	case AggAvg:
+		return "AVG(value)", true
+	case AggMin:
+		return "MIN(value)", true
+	case AggMax:
+		return "MAX(value)", true
+	case AggSum:
+		return "SUM(value)", true
+	case AggCount:
+		return "COUNT(*)", true
+	case AggStdDev:
+		switch driver {
+		case DriverMySQL, DriverPostgreSQL:
+			return "STDDEV_POP(value)", true
+		default:
+			return "", false
+		}
+	default:
+		// Any percentile (fixed p50/p95/p99 and arbitrary pXX / pXX.X all land
+		// here). PostgreSQL has a portable continuous percentile that matches the
+		// in-memory linear-interpolation percentile; other backends fall back to
+		// the in-memory path.
+		if frac, ok := percentileFractionString(agg); ok {
+			if driver == DriverPostgreSQL {
+				return fmt.Sprintf("percentile_cont(%s) WITHIN GROUP (ORDER BY value)", frac), true
+			}
+			return "", false
+		}
+		return "", false
+	}
+}

pkg/metric/dialect_rollup.go

@@ -0,0 +1,160 @@
+package metric
+
+import (
+	"database/sql"
+	"fmt"
+	"strings"
+)
+
+// rollupColumns is the fixed column list for the rollups table, shared by the
+// upsert builder and the row scanner so they cannot drift apart. tags_hash is a
+// stable fingerprint of the (canonical) tag set; tags carries the tag map itself
+// so reads can return it and tag filters can be pushed down. Including the tag
+// dimension in the key is what keeps each distinct tag combination (e.g. a GPU
+// device_index) as its own rollup series instead of being merged together.
+//
+// rollupColumns 是 rollups 表固定列清单，由 upsert 构造器和行扫描器共用，
+// 避免二者漂移。tags_hash 是（规范化）标签集合的稳定指纹；tags 携带标签 map
+// 本身，因此读取时可以返回它，也能下推标签过滤。把标签维度纳入 key，正是让
+// 每个不同标签组合（例如 GPU 的 device_index）保留为自己的 rollup 序列、
+// 而不是被合并在一起的原因。
+const rollupColumns = "metric_name, entity_id, tags_hash, tags, resolution_nano, bucket_nano, " +
+	"count, sum, sum_sq, min_val, max_val, first_val, first_ts, last_val, last_ts, digest, created_at"
+
+// rollupColumnCount is the number of columns in rollupColumns.
+//
+// rollupColumnCount 是 rollupColumns 中的列数量。
+const rollupColumnCount = 17
+
+// rollupTagsArgIndex is the 1-based position of the `tags` JSON column in
+// rollupColumns. PostgreSQL needs a ::jsonb cast on that bind placeholder.
+//
+// rollupTagsArgIndex 是 tags JSON 列在 rollupColumns 中的 1 基位置；
+// PostgreSQL 需要在对应占位符上追加 ::jsonb 转换。
+const rollupTagsArgIndex = 4
+
+// blobType returns the column type used to store the t-digest sketch.
+//
+// blobType 返回用于保存 t-digest 摘要的二进制列类型。
+func (sqliteDialect) blobType() string { return "BLOB" }
+
+// blobType returns the binary column type used for t-digest blobs.
+//
+// blobType 返回 MySQL 用于保存 t-digest 摘要的二进制列类型。
+func (mysqlDialect) blobType() string { return "LONGBLOB" }
+
+// blobType returns the binary column type used for t-digest blobs.
+//
+// blobType 返回 PostgreSQL 用于保存 t-digest 摘要的二进制列类型。
+func (postgresDialect) blobType() string { return "BYTEA" }
+
+// compactTxOptions returns nil so SQLite uses its default isolation. A single
+// write connection already serializes the compaction transaction against
+// concurrent writers, so the raw scan and the raw deletion observe a stable
+// view without escalating the isolation level.
+//
+// compactTxOptions 返回 nil，让 SQLite 使用默认隔离级别。单个写连接已经把
+// compaction 事务与并发写入串行化，因此 raw 扫描和 raw 删除无需提升隔离级别
+// 即可观察到稳定视图。
+func (sqliteDialect) compactTxOptions() *sql.TxOptions { return nil }
+
+// compactTxOptions escalates to SERIALIZABLE so a point inserted between the raw
+// scan and the raw deletion cannot be deleted without first being rolled up.
+// InnoDB's SERIALIZABLE turns the scan into a locking read that blocks such a
+// concurrent insert in the scanned range until the compaction commits.
+//
+// compactTxOptions 提升到 SERIALIZABLE，使在 raw 扫描和 raw 删除之间写入的点
+// 不会在尚未进入 rollup 前被删除。InnoDB 的 SERIALIZABLE 会把扫描变为加锁读，
+// 在 compaction 提交前阻塞扫描范围内的此类并发插入。
+func (mysqlDialect) compactTxOptions() *sql.TxOptions {
+	return &sql.TxOptions{Isolation: sql.LevelSerializable}
+}
+
+// compactTxOptions escalates to SERIALIZABLE so the raw scan and the raw
+// deletion share one snapshot. A concurrent insert into the scanned range that
+// would otherwise be deleted without a rollup instead triggers a serialization
+// failure that CompactMetric retries on a fresh snapshot.
+//
+// compactTxOptions 提升到 SERIALIZABLE，使 raw 扫描和 raw 删除共享同一个快照。
+// 对扫描范围内的并发插入（否则会被删除却没有进入 rollup）会触发序列化失败，
+// 由 CompactMetric 在新快照上重试。
+func (postgresDialect) compactTxOptions() *sql.TxOptions {
+	return &sql.TxOptions{Isolation: sql.LevelSerializable}
+}
+
+// upsertRollupSQL builds a single-row upsert for a rollup cell, keyed by
+// (metric_name, entity_id, tags_hash, resolution_nano, bucket_nano). Compact
+// recomputes a bucket wholesale from its inputs each run, so the conflict action
+// replaces every summary column rather than trying to merge in SQL.
+//
+// upsertRollupSQL 构造单个 rollup 单元的 upsert SQL，其 key 为
+// (metric_name, entity_id, tags_hash, resolution_nano, bucket_nano)。Compact
+// 每次都会从输入整体重新计算一个桶，因此冲突动作会替换每个摘要列，
+// 而不是尝试在 SQL 中合并。
+func (d sqliteDialect) upsertRollupSQL(t tables) string {
+	return buildUpsertRollupSQL(t.rollups, d, sqliteRollupConflict)
+}
+
+// upsertRollupSQL builds backend-specific SQL for upserting a rollup cell.
+//
+// upsertRollupSQL 构造 MySQL rollup 单元 upsert SQL。
+func (d mysqlDialect) upsertRollupSQL(t tables) string {
+	return buildUpsertRollupSQL(t.rollups, d, mysqlRollupConflict)
+}
+
+// upsertRollupSQL builds backend-specific SQL for upserting a rollup cell.
+//
+// upsertRollupSQL 构造 PostgreSQL rollup 单元 upsert SQL。
+func (d postgresDialect) upsertRollupSQL(t tables) string {
+	return buildUpsertRollupSQL(t.rollups, d, postgresRollupConflict)
+}
+
+// sqliteRollupConflict is the SQLite conflict clause for replacing rollup cells.
+//
+// sqliteRollupConflict 是 SQLite 用于替换 rollup 单元的冲突处理子句。
+const sqliteRollupConflict = `ON CONFLICT(metric_name, entity_id, tags_hash, resolution_nano, bucket_nano) DO UPDATE SET
+	tags = excluded.tags,
+	count = excluded.count, sum = excluded.sum, sum_sq = excluded.sum_sq,
+	min_val = excluded.min_val, max_val = excluded.max_val,
+	first_val = excluded.first_val, first_ts = excluded.first_ts,
+	last_val = excluded.last_val, last_ts = excluded.last_ts,
+	digest = excluded.digest, created_at = excluded.created_at`
+
+// postgresRollupConflict is the PostgreSQL conflict clause for replacing rollup cells.
+//
+// postgresRollupConflict 是 PostgreSQL 用于替换 rollup 单元的冲突处理子句。
+const postgresRollupConflict = `ON CONFLICT(metric_name, entity_id, tags_hash, resolution_nano, bucket_nano) DO UPDATE SET
+	tags = EXCLUDED.tags,
+	count = EXCLUDED.count, sum = EXCLUDED.sum, sum_sq = EXCLUDED.sum_sq,
+	min_val = EXCLUDED.min_val, max_val = EXCLUDED.max_val,
+	first_val = EXCLUDED.first_val, first_ts = EXCLUDED.first_ts,
+	last_val = EXCLUDED.last_val, last_ts = EXCLUDED.last_ts,
+	digest = EXCLUDED.digest, created_at = EXCLUDED.created_at`
+
+// mysqlRollupConflict is the MySQL conflict clause for replacing rollup cells.
+//
+// mysqlRollupConflict 是 MySQL 用于替换 rollup 单元的冲突处理子句。
+const mysqlRollupConflict = `ON DUPLICATE KEY UPDATE
+	tags = VALUES(tags),
+	count = VALUES(count), sum = VALUES(sum), sum_sq = VALUES(sum_sq),
+	min_val = VALUES(min_val), max_val = VALUES(max_val),
+	first_val = VALUES(first_val), first_ts = VALUES(first_ts),
+	last_val = VALUES(last_val), last_ts = VALUES(last_ts),
+	digest = VALUES(digest), created_at = VALUES(created_at)`
+
+// buildUpsertRollupSQL builds a single-row rollup upsert statement.
+//
+// buildUpsertRollupSQL 构造单行 rollup upsert SQL。
+func buildUpsertRollupSQL(table string, d dialect, conflict string) string {
+	ph := make([]string, rollupColumnCount)
+	for i := 0; i < rollupColumnCount; i++ {
+		if i+1 == rollupTagsArgIndex {
+			// The tags column is JSON; PostgreSQL requires an explicit ::jsonb cast.
+			ph[i] = d.jsonPlaceholder(i + 1)
+		} else {
+			ph[i] = d.placeholder(i + 1)
+		}
+	}
+	return fmt.Sprintf("INSERT INTO %s (%s) VALUES (%s) %s",
+		table, rollupColumns, strings.Join(ph, ", "), conflict)
+}

pkg/metric/dialect_test.go

@@ -0,0 +1,60 @@
+package metric
+
+import (
+	"strings"
+	"testing"
+)
+
+// TestDialectsGenerateBackendSpecificSQL verifies backend-specific SQL rendering.
+//
+// TestDialectsGenerateBackendSpecificSQL 验证不同数据库后端生成各自的 SQL。
+func TestDialectsGenerateBackendSpecificSQL(t *testing.T) {
+	tests := []struct {
+		name        string
+		driver      Driver
+		placeholder string
+		upsert      string
+		jsonType    string
+	}{
+		{
+			name:        "sqlite",
+			driver:      DriverSQLite,
+			placeholder: "?",
+			upsert:      "ON CONFLICT(metric_name, entity_id, tags_hash, ts_nano)",
+			jsonType:    "TEXT",
+		},
+		{
+			name:        "mysql",
+			driver:      DriverMySQL,
+			placeholder: "?",
+			upsert:      "ON DUPLICATE KEY UPDATE",
+			jsonType:    "JSON",
+		},
+		{
+			name:        "postgresql",
+			driver:      DriverPostgreSQL,
+			placeholder: "$1",
+			upsert:      "ON CONFLICT(metric_name, entity_id, tags_hash, ts_nano)",
+			jsonType:    "JSONB",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			d := newDialect(tt.driver)
+			if got := d.placeholder(1); got != tt.placeholder {
+				t.Fatalf("placeholder: expected %q, got %q", tt.placeholder, got)
+			}
+			if got := d.jsonType(); got != tt.jsonType {
+				t.Fatalf("json type: expected %q, got %q", tt.jsonType, got)
+			}
+			sql := d.upsertPointSQL(tables{points: "metric_points"}, 2)
+			if !strings.Contains(sql, "metric_points") {
+				t.Fatalf("upsert sql should reference metric_points: %s", sql)
+			}
+			if !strings.Contains(sql, tt.upsert) {
+				t.Fatalf("upsert sql should contain %q: %s", tt.upsert, sql)
+			}
+		})
+	}
+}

pkg/metric/errors.go

@@ -0,0 +1,26 @@
+package metric
+
+import "errors"
+
+var (
+	// ErrInvalidArgument reports an invalid caller-supplied argument.
+	//
+	// ErrInvalidArgument 表示调用方传入了无效参数。
+	ErrInvalidArgument = errors.New("metric: invalid argument")
+	// ErrNotFound reports a missing metric or record.
+	//
+	// ErrNotFound 表示指标或记录不存在。
+	ErrNotFound = errors.New("metric: not found")
+	// ErrNoData reports that a valid query range contained no samples.
+	//
+	// ErrNoData 表示有效查询范围内没有采样。
+	ErrNoData = errors.New("metric: no data in range")
+	// ErrAlreadyExists reports that a create-only operation found an existing row.
+	//
+	// ErrAlreadyExists 表示只创建操作遇到了已存在的行。
+	ErrAlreadyExists = errors.New("metric: already exists")
+	// ErrClosed reports that the store has already been closed.
+	//
+	// ErrClosed 表示 Store 已经关闭。
+	ErrClosed = errors.New("metric: store is closed")
+)

pkg/metric/example_test.go

@@ -0,0 +1,347 @@
+package metric_test
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"strings"
+	"time"
+
+	"github.com/komari-monitor/komari/pkg/metric"
+)
+
+// Example_rollupTags demonstrates tagged rollups with automatic series routing.
+//
+// Example_rollupTags 演示带标签的 rollup 以及 Series 自动路由读取。
+func Example_rollupTags() {
+	ctx := context.Background()
+	base := time.Date(2026, 6, 18, 10, 0, 0, 0, time.UTC)
+
+	store, err := metric.Open(ctx, metric.SQLite(
+		"file:metric-example?mode=memory&cache=shared",
+		metric.WithRollupPolicy(metric.RollupPolicy{
+			RawRetention: 2 * time.Minute,
+			Tiers: []metric.RollupTier{
+				{Interval: time.Minute, Retention: 24 * time.Hour},
+			},
+		}),
+	))
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer store.Close()
+
+	if err := store.CreateMetric(ctx, metric.Definition{
+		Name: "gpu.usage",
+		Type: metric.TypeGauge,
+		Unit: "%",
+	}); err != nil {
+		log.Fatal(err)
+	}
+
+	var points []metric.Point
+	for i := 0; i < 10; i++ {
+		ts := base.Add(time.Duration(i) * time.Second)
+		points = append(points,
+			metric.Point{
+				MetricName: "gpu.usage",
+				EntityID:   "host-1",
+				Timestamp:  ts,
+				Value:      float64(10 + i),
+				Tags:       map[string]string{"device_index": "0"},
+			},
+			metric.Point{
+				MetricName: "gpu.usage",
+				EntityID:   "host-1",
+				Timestamp:  ts,
+				Value:      float64(80 + i),
+				Tags:       map[string]string{"device_index": "1"},
+			},
+		)
+	}
+	if err := store.WriteBatch(ctx, points); err != nil {
+		log.Fatal(err)
+	}
+
+	// Compact builds one rollup series per tag set. Raw points older than the
+	// policy's RawRetention are deleted after their rollups are materialized.
+	now := base.Add(time.Hour)
+	if _, err := store.Compact(ctx, now); err != nil {
+		log.Fatal(err)
+	}
+
+	series, err := store.Series(ctx, metric.AggregateQuery{
+		Query: metric.Query{
+			MetricName: "gpu.usage",
+			EntityID:   "host-1",
+			Start:      base,
+			End:        base.Add(time.Minute),
+			Tags:       map[string]string{"device_index": "0"},
+		},
+		Aggregation: metric.AggAvg,
+		Interval:    time.Minute,
+	}, now)
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	fmt.Printf("device 0: count=%d avg=%.1f\n", series[0].Count, series[0].Value)
+
+	// Output:
+	// device 0: count=10 avg=14.5
+}
+
+// Example_agentFleetLifecycle demonstrates a realistic Komari-style fleet:
+// 50 agents report system metrics every second, ping tasks are represented as a
+// tagged ping metric, and lifecycle events clean up task and agent series.
+//
+// Example_agentFleetLifecycle 演示贴近 Komari 的实际场景：50 个 agent 每秒上报系统
+// 指标，ping task 用带标签的 ping 指标表示，并通过生命周期事件清理 task 和 agent 序列。
+func Example_agentFleetLifecycle() {
+	ctx := context.Background()
+	base := time.Date(2026, 6, 18, 10, 0, 0, 0, time.UTC)
+
+	store, err := metric.Open(ctx, metric.SQLite(
+		"file:metric-agent-fleet?mode=memory&cache=shared",
+		metric.WithRollupPolicy(metric.RollupPolicy{
+			RawRetention: 2 * time.Minute,
+			Tiers: []metric.RollupTier{
+				{Interval: time.Minute, Retention: 24 * time.Hour},
+				{Interval: 5 * time.Minute, Retention: 30 * 24 * time.Hour},
+			},
+		}),
+	))
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer store.Close()
+
+	for _, def := range []metric.Definition{
+		{Name: "cpu.usage", Type: metric.TypeGauge, Unit: "%"},
+		{Name: "mem.usage", Type: metric.TypeGauge, Unit: "%"},
+		{Name: "disk.usage", Type: metric.TypeGauge, Unit: "%"},
+		{Name: "swap.usage", Type: metric.TypeGauge, Unit: "%"},
+		{Name: "net.in.bytes", Type: metric.TypeCounter, Unit: "bytes"},
+		{Name: "net.out.bytes", Type: metric.TypeCounter, Unit: "bytes"},
+		{Name: "process.count", Type: metric.TypeGauge, Unit: "count"},
+		{Name: "gpu.usage", Type: metric.TypeGauge, Unit: "%"},
+		{Name: "ping.latency_ms", Type: metric.TypeGauge, Unit: "ms"},
+	} {
+		if err := store.UpsertMetric(ctx, def); err != nil {
+			log.Fatal(err)
+		}
+	}
+
+	agents := makeAgents(50)
+	pingTasks := []string{"pingtask1", "pingtask2"}
+	for sec := 0; sec < 120; sec++ {
+		ts := base.Add(time.Duration(sec) * time.Second)
+
+		switch sec {
+		case 45:
+			// A ping task is deleted by the admin. Because task_id is a tag, one
+			// DeleteSeries removes that task's records for every agent, including
+			// already materialized rollups.
+			if _, err := store.DeleteSeries(ctx, metric.Query{
+				MetricName: "ping.latency_ms",
+				Tags:       map[string]string{"task_id": "pingtask2"},
+			}); err != nil {
+				log.Fatal(err)
+			}
+			pingTasks = []string{"pingtask1"}
+		case 70:
+			// The admin clears all ping task records while keeping the task
+			// definitions themselves. Deleting the metric data is the simplest
+			// reset; UpsertMetric recreates the definition before new samples land.
+			if err := store.DeleteMetric(ctx, "ping.latency_ms"); err != nil {
+				log.Fatal(err)
+			}
+			if err := store.UpsertMetric(ctx, metric.Definition{
+				Name: "ping.latency_ms", Type: metric.TypeGauge, Unit: "ms",
+			}); err != nil {
+				log.Fatal(err)
+			}
+		case 85:
+			// An agent is removed. DeleteEntity clears every metric series for
+			// that agent, raw and rollup, across the whole store.
+			if _, err := store.DeleteEntity(ctx, "agent-007"); err != nil {
+				log.Fatal(err)
+			}
+			agents = removeAgent(agents, "agent-007")
+		case 95:
+			// A new agent is added. No schema change is needed; the first write
+			// creates its series naturally.
+			agents = append(agents, "agent-051")
+		}
+
+		points := make([]metric.Point, 0, len(agents)*(7+2+len(pingTasks)))
+		for _, agentID := range agents {
+			points = append(points, systemPoints(agentID, ts, sec)...)
+			for _, gpu := range []string{"0", "1"} {
+				points = append(points, metric.Point{
+					MetricName: "gpu.usage",
+					EntityID:   agentID,
+					Timestamp:  ts,
+					Value:      gpuUsage(agentID, gpu, sec),
+					Tags:       map[string]string{"device_index": gpu},
+				})
+			}
+			for _, taskID := range pingTasks {
+				points = append(points, metric.Point{
+					MetricName: "ping.latency_ms",
+					EntityID:   agentID,
+					Timestamp:  ts,
+					Value:      pingLatency(agentID, taskID, sec),
+					Tags:       map[string]string{"task_id": taskID},
+				})
+			}
+		}
+		if err := store.WriteBatch(ctx, points); err != nil {
+			log.Fatal(err)
+		}
+		if sec%30 == 29 {
+			if _, err := store.Compact(ctx, ts); err != nil {
+				log.Fatal(err)
+			}
+		}
+	}
+
+	now := base.Add(2 * time.Hour)
+	if _, err := store.Compact(ctx, now); err != nil {
+		log.Fatal(err)
+	}
+
+	activeCPU, err := store.Series(ctx, metric.AggregateQuery{
+		Query: metric.Query{
+			MetricName: "cpu.usage",
+			Start:      base,
+			End:        base.Add(2 * time.Minute),
+		},
+		Aggregation: metric.AggAvg,
+		Interval:    time.Minute,
+	}, now)
+	if err != nil {
+		log.Fatal(err)
+	}
+	deletedAgentCPU, err := store.Series(ctx, metric.AggregateQuery{
+		Query: metric.Query{
+			MetricName: "cpu.usage",
+			EntityID:   "agent-007",
+			Start:      base,
+			End:        base.Add(2 * time.Minute),
+		},
+		Aggregation: metric.AggAvg,
+		Interval:    time.Minute,
+	}, now)
+	if err != nil {
+		log.Fatal(err)
+	}
+	task1, err := store.Series(ctx, metric.AggregateQuery{
+		Query: metric.Query{
+			MetricName: "ping.latency_ms",
+			Start:      base,
+			End:        base.Add(2 * time.Minute),
+			Tags:       map[string]string{"task_id": "pingtask1"},
+		},
+		Aggregation: metric.AggAvg,
+		Interval:    time.Minute,
+	}, now)
+	if err != nil {
+		log.Fatal(err)
+	}
+	task2, err := store.Series(ctx, metric.AggregateQuery{
+		Query: metric.Query{
+			MetricName: "ping.latency_ms",
+			Start:      base,
+			End:        base.Add(2 * time.Minute),
+			Tags:       map[string]string{"task_id": "pingtask2"},
+		},
+		Aggregation: metric.AggAvg,
+		Interval:    time.Minute,
+	}, now)
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	fmt.Printf("agents now=%d\n", len(agents))
+	fmt.Printf("fleet cpu buckets=%d\n", len(nonEmpty(activeCPU)))
+	fmt.Printf("deleted agent buckets=%d\n", len(nonEmpty(deletedAgentCPU)))
+	fmt.Printf("pingtask1 buckets=%d\n", len(nonEmpty(task1)))
+	fmt.Printf("pingtask2 buckets=%d\n", len(nonEmpty(task2)))
+
+	// Output:
+	// agents now=50
+	// fleet cpu buckets=2
+	// deleted agent buckets=0
+	// pingtask1 buckets=1
+	// pingtask2 buckets=0
+}
+
+func makeAgents(n int) []string {
+	out := make([]string, 0, n)
+	for i := 1; i <= n; i++ {
+		out = append(out, fmt.Sprintf("agent-%03d", i))
+	}
+	return out
+}
+
+func removeAgent(agents []string, deleted string) []string {
+	out := agents[:0]
+	for _, agentID := range agents {
+		if agentID != deleted {
+			out = append(out, agentID)
+		}
+	}
+	return out
+}
+
+func systemPoints(agentID string, ts time.Time, sec int) []metric.Point {
+	id := numericSuffix(agentID)
+	return []metric.Point{
+		{MetricName: "cpu.usage", EntityID: agentID, Timestamp: ts, Value: float64((id*3 + sec) % 100)},
+		{MetricName: "mem.usage", EntityID: agentID, Timestamp: ts, Value: 35 + float64((id+sec)%40)},
+		{MetricName: "disk.usage", EntityID: agentID, Timestamp: ts, Value: 50 + float64(id%20)/2},
+		{MetricName: "swap.usage", EntityID: agentID, Timestamp: ts, Value: float64((id + sec/10) % 12)},
+		{MetricName: "net.in.bytes", EntityID: agentID, Timestamp: ts, Value: float64(sec * (1000 + id))},
+		{MetricName: "net.out.bytes", EntityID: agentID, Timestamp: ts, Value: float64(sec * (700 + id))},
+		{MetricName: "process.count", EntityID: agentID, Timestamp: ts, Value: float64(90 + id%25)},
+	}
+}
+
+func gpuUsage(agentID, deviceIndex string, sec int) float64 {
+	id := numericSuffix(agentID)
+	device := 0
+	if deviceIndex == "1" {
+		device = 37
+	}
+	return float64((id*5 + device + sec*2) % 100)
+}
+
+func pingLatency(agentID, taskID string, sec int) float64 {
+	id := numericSuffix(agentID)
+	offset := 0
+	if strings.HasSuffix(taskID, "2") {
+		offset = 35
+	}
+	return float64(20 + (id+sec+offset)%80)
+}
+
+func numericSuffix(agentID string) int {
+	var n int
+	for _, r := range agentID {
+		if r >= '0' && r <= '9' {
+			n = n*10 + int(r-'0')
+		}
+	}
+	return n
+}
+
+func nonEmpty(points []metric.AggregatePoint) []metric.AggregatePoint {
+	out := points[:0]
+	for _, p := range points {
+		if p.Count > 0 {
+			out = append(out, p)
+		}
+	}
+	return out
+}

pkg/metric/improvements_test.go

@@ -0,0 +1,565 @@
+package metric
+
+import (
+	"context"
+	"errors"
+	"math"
+	"path/filepath"
+	"testing"
+	"time"
+)
+
+// newMemStore opens an isolated in-memory store for tests.
+//
+// newMemStore 打开一个用于测试的隔离内存 Store。
+func newMemStore(t *testing.T) *Store {
+	t.Helper()
+	store, err := Open(context.Background(), SQLite("file:imp-test?mode=memory&cache=shared"))
+	if err != nil {
+		t.Fatalf("open store: %v", err)
+	}
+	t.Cleanup(func() { store.Close() })
+	return store
+}
+
+// TestCreateMetricRejectsDuplicate verifies create-only metric semantics.
+//
+// TestCreateMetricRejectsDuplicate 验证 CreateMetric 遇到重复指标时会拒绝。
+func TestCreateMetricRejectsDuplicate(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	def := Definition{Name: "dup.metric", Type: TypeGauge}
+	if err := s.CreateMetric(ctx, def); err != nil {
+		t.Fatalf("first create: %v", err)
+	}
+	err := s.CreateMetric(ctx, def)
+	if !errors.Is(err, ErrAlreadyExists) {
+		t.Fatalf("expected ErrAlreadyExists on duplicate, got %v", err)
+	}
+}
+
+// TestUpsertMetricOverwrites verifies upsert updates mutable definition fields.
+//
+// TestUpsertMetricOverwrites 验证 UpsertMetric 会更新指标定义的可变字段。
+func TestUpsertMetricOverwrites(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	if err := s.CreateMetric(ctx, Definition{Name: "m", Type: TypeGauge, Unit: "ms"}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	if err := s.UpsertMetric(ctx, Definition{Name: "m", Type: TypeCounter, Unit: "count"}); err != nil {
+		t.Fatalf("upsert: %v", err)
+	}
+	got, err := s.GetMetric(ctx, "m")
+	if err != nil {
+		t.Fatalf("get: %v", err)
+	}
+	if got.Type != TypeCounter || got.Unit != "count" {
+		t.Fatalf("upsert did not overwrite: %#v", got)
+	}
+}
+
+// TestTagFilterPushdownWithPaging verifies tag filtering happens before paging.
+//
+// TestTagFilterPushdownWithPaging 验证标签过滤会先于分页在 SQL 中执行。
+func TestTagFilterPushdownWithPaging(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	if err := s.CreateMetric(ctx, Definition{Name: "req", Type: TypeCounter}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var batch []Point
+	for i := 0; i < 6; i++ {
+		env := "prod"
+		if i%2 == 1 {
+			env = "stage"
+		}
+		batch = append(batch, Point{
+			MetricName: "req", EntityID: "n1",
+			Timestamp: base.Add(time.Duration(i) * time.Minute),
+			Value:     float64(i),
+			Tags:      map[string]string{"env": env},
+		})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	// 3 prod points (i=0,2,4); with Limit=2 the SQL LIMIT must apply to the
+	// tag-filtered set, returning the first two prod points in time order.
+	got, err := s.Query(ctx, Query{
+		MetricName: "req", EntityID: "n1",
+		Start: base, End: base.Add(time.Hour),
+		Tags:  map[string]string{"env": "prod"},
+		Limit: 2,
+	})
+	if err != nil {
+		t.Fatalf("query: %v", err)
+	}
+	if len(got) != 2 {
+		t.Fatalf("expected 2 paged prod points, got %d", len(got))
+	}
+	if got[0].Value != 0 || got[1].Value != 2 {
+		t.Fatalf("unexpected paged values: %#v", got)
+	}
+}
+
+// TestSQLAggregateMatchesInMemory compares SQL and in-memory aggregation.
+//
+// TestSQLAggregateMatchesInMemory 对比 SQL 下推聚合和内存聚合结果。
+func TestSQLAggregateMatchesInMemory(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	if err := s.CreateMetric(ctx, Definition{Name: "g", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var batch []Point
+	vals := []float64{5, 15, 25, 35, 100, 200}
+	for i, v := range vals {
+		batch = append(batch, Point{
+			MetricName: "g", EntityID: "n1",
+			Timestamp: base.Add(time.Duration(i) * time.Minute),
+			Value:     v,
+		})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	q := AggregateQuery{
+		Query:       Query{MetricName: "g", EntityID: "n1", Start: base, End: base.Add(time.Hour)},
+		Aggregation: AggAvg,
+		Interval:    3 * time.Minute,
+	}
+	// SQL pushdown path.
+	sqlRes, err := s.Aggregate(ctx, q)
+	if err != nil {
+		t.Fatalf("sql aggregate: %v", err)
+	}
+	// In-memory reference over the same raw points.
+	pts, err := s.Query(ctx, q.Query)
+	if err != nil {
+		t.Fatalf("query: %v", err)
+	}
+	memRes, err := AggregatePoints(pts, q)
+	if err != nil {
+		t.Fatalf("mem aggregate: %v", err)
+	}
+	if len(sqlRes) != len(memRes) {
+		t.Fatalf("bucket count mismatch: sql=%d mem=%d", len(sqlRes), len(memRes))
+	}
+	for i := range sqlRes {
+		if !sqlRes[i].Bucket.Equal(memRes[i].Bucket) || sqlRes[i].Value != memRes[i].Value || sqlRes[i].Count != memRes[i].Count {
+			t.Fatalf("bucket %d mismatch: sql=%#v mem=%#v", i, sqlRes[i], memRes[i])
+		}
+	}
+}
+
+// TestCounterRateHandlesReset verifies reset-aware counter rate calculation.
+//
+// TestCounterRateHandlesReset 验证计数器重置时速率计算仍然稳定。
+func TestCounterRateHandlesReset(t *testing.T) {
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	// Counter goes 0 -> 10 -> reset -> 5; naive (last-first)/sec would give a
+	// positive-but-wrong 5/30s. Correct counter rate sums positive deltas:
+	// (10-0) + (5-0 after reset) = 15 over 30s = 0.5/s.
+	pts := []Point{
+		{Timestamp: base, Value: 0},
+		{Timestamp: base.Add(10 * time.Second), Value: 10},
+		{Timestamp: base.Add(20 * time.Second), Value: 0},
+		{Timestamp: base.Add(30 * time.Second), Value: 5},
+	}
+	rate := counterRate(pts)
+	if rate != 0.5 {
+		t.Fatalf("expected reset-aware rate 0.5/s, got %v", rate)
+	}
+}
+
+// TestAlignTimeNegativeTimestamp verifies pre-epoch bucket alignment.
+//
+// TestAlignTimeNegativeTimestamp 验证 Unix epoch 之前的时间也能正确对齐桶。
+func TestAlignTimeNegativeTimestamp(t *testing.T) {
+	interval := time.Minute
+	// 30s before the epoch should align down to -60s, not up to 0.
+	tm := time.Unix(-30, 0).UTC()
+	got := alignTime(tm, interval)
+	want := time.Unix(-60, 0).UTC()
+	if !got.Equal(want) {
+		t.Fatalf("alignTime negative: got %v want %v", got, want)
+	}
+}
+
+// TestLatestReturnsMostRecent verifies latest-point ordering.
+//
+// TestLatestReturnsMostRecent 验证 Latest 返回最新采样点。
+func TestLatestReturnsMostRecent(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	if err := s.CreateMetric(ctx, Definition{Name: "l", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	for i := 0; i < 5; i++ {
+		if err := s.Write(ctx, Point{MetricName: "l", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Minute), Value: float64(i)}); err != nil {
+			t.Fatalf("write: %v", err)
+		}
+	}
+	got, err := s.Latest(ctx, "l", "n1", 2)
+	if err != nil {
+		t.Fatalf("latest: %v", err)
+	}
+	if len(got) != 2 || got[0].Value != 4 || got[1].Value != 3 {
+		t.Fatalf("unexpected latest: %#v", got)
+	}
+}
+
+// TestStatsDistinguishesNoDataFromUnknownMetric verifies stats error semantics.
+//
+// TestStatsDistinguishesNoDataFromUnknownMetric 验证 Stats 能区分无数据和未知指标。
+func TestStatsDistinguishesNoDataFromUnknownMetric(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+
+	// Unknown metric -> ErrNotFound.
+	_, err := s.Stats(ctx, Query{MetricName: "nope", Start: base, End: base.Add(time.Hour)})
+	if !errors.Is(err, ErrNotFound) {
+		t.Fatalf("unknown metric: expected ErrNotFound, got %v", err)
+	}
+
+	// Known metric but empty window -> ErrNoData.
+	if err := s.CreateMetric(ctx, Definition{Name: "known", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	_, err = s.Stats(ctx, Query{MetricName: "known", Start: base, End: base.Add(time.Hour)})
+	if !errors.Is(err, ErrNoData) {
+		t.Fatalf("empty window: expected ErrNoData, got %v", err)
+	}
+}
+
+// TestStdDevPopMatchesCalculateStats verifies population standard deviation.
+//
+// TestStdDevPopMatchesCalculateStats 验证总体标准差与统计摘要一致。
+func TestStdDevPopMatchesCalculateStats(t *testing.T) {
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var pts []Point
+	for i, v := range []float64{10, 20, 30, 40, 50} {
+		pts = append(pts, Point{Timestamp: base.Add(time.Duration(i) * time.Minute), Value: v})
+	}
+	st, err := CalculateStats(pts)
+	if err != nil {
+		t.Fatalf("stats: %v", err)
+	}
+	got, _ := aggregateValue(pts, AggStdDev)
+	if math.Abs(got-st.StdDev) > 1e-9 {
+		t.Fatalf("AggStdDev %v != Stats.StdDev %v", got, st.StdDev)
+	}
+	// Known value: population stddev of 10..50 step 10 is sqrt(200) ~= 14.142135.
+	if math.Abs(got-14.142135623730951) > 1e-9 {
+		t.Fatalf("unexpected population stddev: %v", got)
+	}
+}
+
+// TestAggregateStdDevSQLiteUsesMemoryPath verifies SQLite stddev fallback.
+//
+// TestAggregateStdDevSQLiteUsesMemoryPath 验证 SQLite 标准差聚合会回退到内存路径。
+func TestAggregateStdDevSQLiteUsesMemoryPath(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	if err := s.CreateMetric(ctx, Definition{Name: "sd", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var batch []Point
+	for i, v := range []float64{10, 20, 30} {
+		batch = append(batch, Point{MetricName: "sd", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Minute), Value: v})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	res, err := s.Aggregate(ctx, AggregateQuery{
+		Query:       Query{MetricName: "sd", EntityID: "n1", Start: base, End: base.Add(time.Hour)},
+		Aggregation: AggStdDev,
+		Interval:    time.Hour,
+	})
+	if err != nil {
+		t.Fatalf("aggregate stddev: %v", err)
+	}
+	if len(res) != 1 {
+		t.Fatalf("expected 1 bucket, got %d", len(res))
+	}
+	// Population stddev of {10,20,30} = sqrt(200/3) ~= 8.16496.
+	if math.Abs(res[0].Value-8.16496580927726) > 1e-9 {
+		t.Fatalf("unexpected stddev bucket value: %v", res[0].Value)
+	}
+}
+
+// TestSQLAggValueExprPushdownMatrix verifies aggregation pushdown support.
+//
+// TestSQLAggValueExprPushdownMatrix 验证各后端支持的聚合下推矩阵。
+func TestSQLAggValueExprPushdownMatrix(t *testing.T) {
+	cases := []struct {
+		driver Driver
+		agg    Aggregation
+		ok     bool
+	}{
+		{DriverSQLite, AggAvg, true},
+		{DriverSQLite, AggStdDev, false}, // no STDDEV_POP in sqlite -> memory
+		{DriverSQLite, AggP95, false},    // no percentile in sqlite -> memory
+		{DriverMySQL, AggStdDev, true},   // STDDEV_POP
+		{DriverMySQL, AggP95, false},     // no portable continuous percentile
+		{DriverPostgreSQL, AggStdDev, true},
+		{DriverPostgreSQL, AggP50, true}, // percentile_cont
+		{DriverPostgreSQL, AggP99, true},
+		{DriverPostgreSQL, AggFirst, false}, // needs ordered raw series
+	}
+	for _, c := range cases {
+		expr, ok := sqlAggValueExpr(c.driver, c.agg)
+		if ok != c.ok {
+			t.Fatalf("%s/%s: pushdown=%v want %v (expr=%q)", c.driver, c.agg, ok, c.ok, expr)
+		}
+		if ok && expr == "" {
+			t.Fatalf("%s/%s: pushdown ok but empty expr", c.driver, c.agg)
+		}
+	}
+}
+
+// TestSQLiteReadPoolOpens verifies SQLite read-pool creation.
+//
+// TestSQLiteReadPoolOpens 验证 SQLite 只读连接池会按配置打开。
+func TestSQLiteReadPoolOpens(t *testing.T) {
+	ctx := context.Background()
+	dir := filepath.Join(t.TempDir(), "rp")
+	store, err := Open(ctx, SQLiteInDir(dir, WithSQLiteReadPool(4)))
+	if err != nil {
+		t.Fatalf("open with read pool: %v", err)
+	}
+	t.Cleanup(func() { store.Close() })
+	if store.readDB == nil {
+		t.Fatalf("expected a dedicated read pool to be opened")
+	}
+	if store.reader() != store.readDB {
+		t.Fatalf("reader() should return the dedicated read pool")
+	}
+	// Round-trip a write (primary) and a read (read pool).
+	if err := store.CreateMetric(ctx, Definition{Name: "rp", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	if err := store.Write(ctx, Point{MetricName: "rp", EntityID: "n1", Timestamp: base, Value: 7}); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	got, err := store.Latest(ctx, "rp", "n1", 1)
+	if err != nil {
+		t.Fatalf("latest: %v", err)
+	}
+	if len(got) != 1 || got[0].Value != 7 {
+		t.Fatalf("read pool round-trip failed: %#v", got)
+	}
+}
+
+// TestMemoryDSNSkipsReadPool verifies memory SQLite skips read pools.
+//
+// TestMemoryDSNSkipsReadPool 验证内存 SQLite 不会打开独立读池。
+func TestMemoryDSNSkipsReadPool(t *testing.T) {
+	ctx := context.Background()
+	store, err := Open(ctx, SQLite("file:rp-mem?mode=memory&cache=shared", WithSQLiteReadPool(4)))
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+	t.Cleanup(func() { store.Close() })
+	if store.readDB != nil {
+		t.Fatalf("in-memory database must not open a second read pool")
+	}
+}
+
+// TestWriteBatchAtomicAcrossChunks verifies chunked writes are atomic.
+//
+// TestWriteBatchAtomicAcrossChunks 验证分块批量写入仍保持整体原子性。
+func TestWriteBatchAtomicAcrossChunks(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	if err := s.CreateMetric(ctx, Definition{Name: "atom", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	// 1500 valid points, then one invalid (missing entity) to force a mid-batch
+	// failure on the second chunk. The whole batch must roll back.
+	var batch []Point
+	for i := 0; i < 1500; i++ {
+		batch = append(batch, Point{MetricName: "atom", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Millisecond), Value: float64(i)})
+	}
+	batch = append(batch, Point{MetricName: "atom", EntityID: "", Timestamp: base.Add(2 * time.Second), Value: 1}) // invalid
+	if err := s.WriteBatch(ctx, batch); err == nil {
+		t.Fatalf("expected WriteBatch to fail on invalid point")
+	}
+	// Nothing should have been committed.
+	pts, err := s.Query(ctx, Query{MetricName: "atom", EntityID: "n1", Start: base, End: base.Add(time.Hour)})
+	if err != nil {
+		t.Fatalf("query: %v", err)
+	}
+	if len(pts) != 0 {
+		t.Fatalf("expected 0 committed points after rolled-back batch, got %d", len(pts))
+	}
+}
+
+// TestAggregateBucketPagingSQLPath verifies bucket paging in SQL aggregation.
+//
+// TestAggregateBucketPagingSQLPath 验证 SQL 聚合路径按桶分页。
+func TestAggregateBucketPagingSQLPath(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	if err := s.CreateMetric(ctx, Definition{Name: "bp", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	// 6 points across 6 one-minute buckets, values 0..5.
+	var batch []Point
+	for i := 0; i < 6; i++ {
+		batch = append(batch, Point{MetricName: "bp", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Minute), Value: float64(i)})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	// AggAvg -> SQL pushdown path. One bucket per minute. Page: offset 1, limit 2.
+	res, err := s.Aggregate(ctx, AggregateQuery{
+		Query:        Query{MetricName: "bp", EntityID: "n1", Start: base, End: base.Add(time.Hour)},
+		Aggregation:  AggAvg,
+		Interval:     time.Minute,
+		BucketLimit:  2,
+		BucketOffset: 1,
+	})
+	if err != nil {
+		t.Fatalf("aggregate: %v", err)
+	}
+	if len(res) != 2 {
+		t.Fatalf("expected 2 paged buckets, got %d", len(res))
+	}
+	// Buckets are values 0..5; offset 1 skips the value-0 bucket, so we expect 1,2.
+	if res[0].Value != 1 || res[1].Value != 2 {
+		t.Fatalf("unexpected paged bucket values: %#v", res)
+	}
+}
+
+// TestAggregateBucketPagingMemoryPathMatchesSQL compares bucket paging paths.
+//
+// TestAggregateBucketPagingMemoryPathMatchesSQL 对比内存聚合和 SQL 聚合的桶分页语义。
+func TestAggregateBucketPagingMemoryPathMatchesSQL(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	if err := s.CreateMetric(ctx, Definition{Name: "bp2", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var batch []Point
+	for i := 0; i < 6; i++ {
+		batch = append(batch, Point{MetricName: "bp2", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Minute), Value: float64(i * 10)})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	q := Query{MetricName: "bp2", EntityID: "n1", Start: base, End: base.Add(time.Hour)}
+	// AggMin (SQL pushdown) and AggP50 (memory fallback on sqlite) with identical
+	// bucket paging must select the same bucket window (the same timestamps).
+	sqlPaged, err := s.Aggregate(ctx, AggregateQuery{Query: q, Aggregation: AggMin, Interval: time.Minute, BucketLimit: 3, BucketOffset: 2})
+	if err != nil {
+		t.Fatalf("sql aggregate: %v", err)
+	}
+	memPaged, err := s.Aggregate(ctx, AggregateQuery{Query: q, Aggregation: AggP50, Interval: time.Minute, BucketLimit: 3, BucketOffset: 2})
+	if err != nil {
+		t.Fatalf("mem aggregate: %v", err)
+	}
+	if len(sqlPaged) != 3 || len(memPaged) != 3 {
+		t.Fatalf("expected 3 buckets each, got sql=%d mem=%d", len(sqlPaged), len(memPaged))
+	}
+	for i := range sqlPaged {
+		if !sqlPaged[i].Bucket.Equal(memPaged[i].Bucket) {
+			t.Fatalf("bucket window mismatch at %d: sql=%v mem=%v", i, sqlPaged[i].Bucket, memPaged[i].Bucket)
+		}
+	}
+	// Per-bucket single point, so min == p50 == the value.
+	for i := range sqlPaged {
+		if sqlPaged[i].Value != memPaged[i].Value {
+			t.Fatalf("value mismatch at %d: sql=%v mem=%v", i, sqlPaged[i].Value, memPaged[i].Value)
+		}
+	}
+}
+
+// TestAggregateIgnoresRawPointLimit verifies raw limits do not affect aggregation.
+//
+// TestAggregateIgnoresRawPointLimit 验证原始点分页参数不会影响聚合输入。
+func TestAggregateIgnoresRawPointLimit(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	if err := s.CreateMetric(ctx, Definition{Name: "ig", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var batch []Point
+	for i := 0; i < 6; i++ {
+		batch = append(batch, Point{MetricName: "ig", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Minute), Value: float64(i)})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	// The embedded Query.Limit must be ignored for aggregation: a single 1-hour
+	// bucket should aggregate ALL 6 points (count 6), not just 2.
+	res, err := s.Aggregate(ctx, AggregateQuery{
+		Query:       Query{MetricName: "ig", EntityID: "n1", Start: base, End: base.Add(time.Hour), Limit: 2},
+		Aggregation: AggP95, // memory path, where the bug would have surfaced
+		Interval:    time.Hour,
+	})
+	if err != nil {
+		t.Fatalf("aggregate: %v", err)
+	}
+	if len(res) != 1 || res[0].Count != 6 {
+		t.Fatalf("embedded Limit leaked into aggregation: %#v", res)
+	}
+}
+
+// TestJSONTagKeyWithSpecialChars verifies JSON tag keys with special characters.
+//
+// TestJSONTagKeyWithSpecialChars 验证包含特殊字符的标签键可正确查询。
+func TestJSONTagKeyWithSpecialChars(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	if err := s.CreateMetric(ctx, Definition{Name: "tk", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	// Keys containing a dot, a hyphen, a space, and a quote.
+	tags := map[string]string{
+		"region.zone": "ap-1",
+		"a-b":         "yes",
+		"with space":  "ok",
+	}
+	if err := s.Write(ctx, Point{MetricName: "tk", EntityID: "n1", Timestamp: base, Value: 1, Tags: tags}); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	// A second point that should NOT match the filter below.
+	if err := s.Write(ctx, Point{MetricName: "tk", EntityID: "n1", Timestamp: base.Add(time.Minute), Value: 2, Tags: map[string]string{"region.zone": "eu-1"}}); err != nil {
+		t.Fatalf("write 2: %v", err)
+	}
+
+	for _, tc := range []struct {
+		key, val string
+		want     int
+	}{
+		{"region.zone", "ap-1", 1}, // dotted key must match the flat key, not nested
+		{"a-b", "yes", 1},          // hyphen
+		{"with space", "ok", 1},    // space
+		{"region.zone", "eu-1", 1}, // the other point
+	} {
+		got, err := s.Query(ctx, Query{
+			MetricName: "tk", EntityID: "n1",
+			Start: base.Add(-time.Second), End: base.Add(time.Hour),
+			Tags: map[string]string{tc.key: tc.val},
+		})
+		if err != nil {
+			t.Fatalf("query %s=%s: %v", tc.key, tc.val, err)
+		}
+		if len(got) != tc.want {
+			t.Fatalf("tag %q=%q: expected %d points, got %d", tc.key, tc.val, tc.want, len(got))
+		}
+	}
+}

pkg/metric/json.go

@@ -0,0 +1,83 @@
+package metric
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"encoding/json"
+	"fmt"
+)
+
+// encodeMap encodes a string map as JSON.
+//
+// encodeMap 将字符串 map 编码为 JSON 字符串，nil 会按空对象处理。
+func encodeMap(m map[string]string) (string, error) {
+	if m == nil {
+		m = map[string]string{}
+	}
+	b, err := json.Marshal(m)
+	if err != nil {
+		return "", err
+	}
+	return string(b), nil
+}
+
+// decodeMap decodes a scanned JSON value into a string map.
+//
+// decodeMap 将数据库扫描出的 JSON 值还原为字符串 map。
+func decodeMap(v any) (map[string]string, error) {
+	switch x := v.(type) {
+	case nil:
+		return map[string]string{}, nil
+	case string:
+		return decodeMapString(x)
+	case []byte:
+		return decodeMapString(string(x))
+	default:
+		return nil, fmt.Errorf("unsupported json value type %T", v)
+	}
+}
+
+// decodeMapString decodes a JSON object string into a string map.
+//
+// decodeMapString 将 JSON 字符串解码为字符串 map。
+func decodeMapString(s string) (map[string]string, error) {
+	if s == "" {
+		return map[string]string{}, nil
+	}
+	var out map[string]string
+	if err := json.Unmarshal([]byte(s), &out); err != nil {
+		return nil, err
+	}
+	if out == nil {
+		out = map[string]string{}
+	}
+	return out, nil
+}
+
+// canonicalTags returns a deterministic JSON encoding of a tag map. encodeMap
+// already relies on encoding/json, which sorts map keys, so equal tag maps
+// always produce identical bytes — the property tagsFingerprint depends on.
+//
+// canonicalTags 返回标签 map 的确定性 JSON 编码；相同标签集合总会得到
+// 相同字节序列，tagsFingerprint 依赖这个性质。
+func canonicalTags(m map[string]string) (string, error) {
+	return encodeMap(m)
+}
+
+// tagsFingerprint returns a stable hex fingerprint of a tag set, used as the
+// rollups table's tags_hash key column. Equal tag maps (regardless of Go map
+// iteration order) hash identically, so each distinct tag combination becomes
+// its own rollup series; an empty/nil tag map hashes to the fingerprint of "{}".
+//
+// tagsFingerprint 为标签集合生成稳定的十六进制指纹，用作 rollups 表的
+// tags_hash key 列。相同标签 map（无论 Go map 迭代顺序如何）都会得到相同 hash，
+// 因此每种不同标签组合都会成为自己的 rollup 序列；空或 nil 标签 map 会得到 "{}"
+// 的指纹。
+func tagsFingerprint(m map[string]string) (hash string, canonical string, err error) {
+	canonical, err = canonicalTags(m)
+	if err != nil {
+		return "", "", err
+	}
+	sum := sha256.Sum256([]byte(canonical))
+	return hex.EncodeToString(sum[:]), canonical, nil
+}

pkg/metric/migrations.go

@@ -0,0 +1,182 @@
+package metric
+
+import (
+	"context"
+	"fmt"
+)
+
+// Migrate creates the tables and indexes used by the metric store.
+//
+// Migrate 创建 metric 包需要的表和索引。
+func (s *Store) Migrate(ctx context.Context) error {
+	if err := s.ensureOpen(); err != nil {
+		return err
+	}
+	d := s.dialect
+	jsonType := d.jsonType()
+	pk := d.autoIncrementPrimaryKey()
+	blob := d.blobType()
+
+	statements := []string{
+		fmt.Sprintf(`CREATE TABLE IF NOT EXISTS %s (
+			name VARCHAR(191) PRIMARY KEY,
+			type VARCHAR(32) NOT NULL,
+			unit VARCHAR(64) NOT NULL DEFAULT '',
+			description TEXT NOT NULL DEFAULT '',
+			retention_days INTEGER NOT NULL DEFAULT 0,
+			metadata %s NOT NULL,
+			created_at BIGINT NOT NULL,
+			updated_at BIGINT NOT NULL
+		)`, s.tables.definitions, jsonType),
+		fmt.Sprintf(`CREATE TABLE IF NOT EXISTS %s (
+			id %s,
+			metric_name VARCHAR(191) NOT NULL,
+			entity_id VARCHAR(191) NOT NULL,
+			tags_hash VARCHAR(64) NOT NULL,
+			ts_nano BIGINT NOT NULL,
+			value DOUBLE PRECISION NOT NULL,
+			tags %s NOT NULL,
+			labels %s NOT NULL,
+			created_at BIGINT NOT NULL,
+			UNIQUE(metric_name, entity_id, tags_hash, ts_nano)
+		)`, s.tables.points, pk, jsonType, jsonType),
+		fmt.Sprintf(`CREATE INDEX IF NOT EXISTS %s_points_metric_entity_time_idx ON %s (metric_name, entity_id, ts_nano)`, s.cfg.TablePrefix, s.tables.points),
+		fmt.Sprintf(`CREATE INDEX IF NOT EXISTS %s_points_metric_time_idx ON %s (metric_name, ts_nano)`, s.cfg.TablePrefix, s.tables.points),
+		fmt.Sprintf(`CREATE INDEX IF NOT EXISTS %s_points_entity_time_idx ON %s (entity_id, ts_nano)`, s.cfg.TablePrefix, s.tables.points),
+		fmt.Sprintf(`CREATE TABLE IF NOT EXISTS %s (
+			id %s,
+			metric_name VARCHAR(191) NOT NULL,
+			entity_id VARCHAR(191) NOT NULL,
+			tags_hash VARCHAR(64) NOT NULL,
+			tags %s NOT NULL,
+			resolution_nano BIGINT NOT NULL,
+			bucket_nano BIGINT NOT NULL,
+			count BIGINT NOT NULL,
+			sum DOUBLE PRECISION NOT NULL,
+			sum_sq DOUBLE PRECISION NOT NULL,
+			min_val DOUBLE PRECISION NOT NULL,
+			max_val DOUBLE PRECISION NOT NULL,
+			first_val DOUBLE PRECISION NOT NULL,
+			first_ts BIGINT NOT NULL,
+			last_val DOUBLE PRECISION NOT NULL,
+			last_ts BIGINT NOT NULL,
+			digest %s,
+			created_at BIGINT NOT NULL,
+			UNIQUE(metric_name, entity_id, tags_hash, resolution_nano, bucket_nano)
+		)`, s.tables.rollups, pk, jsonType, blob),
+		fmt.Sprintf(`CREATE INDEX IF NOT EXISTS %s_rollups_lookup_idx ON %s (metric_name, entity_id, tags_hash, resolution_nano, bucket_nano)`, s.cfg.TablePrefix, s.tables.rollups),
+		fmt.Sprintf(`CREATE INDEX IF NOT EXISTS %s_rollups_res_time_idx ON %s (metric_name, resolution_nano, bucket_nano)`, s.cfg.TablePrefix, s.tables.rollups),
+	}
+
+	if s.cfg.Driver == DriverMySQL {
+		statements = []string{
+			fmt.Sprintf(`CREATE TABLE IF NOT EXISTS %s (
+				name VARCHAR(191) PRIMARY KEY,
+				type VARCHAR(32) NOT NULL,
+				unit VARCHAR(64) NOT NULL DEFAULT '',
+				description TEXT NOT NULL,
+				retention_days INT NOT NULL DEFAULT 0,
+				metadata %s NOT NULL,
+				created_at BIGINT NOT NULL,
+				updated_at BIGINT NOT NULL
+			) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci`, s.tables.definitions, jsonType),
+			fmt.Sprintf(`CREATE TABLE IF NOT EXISTS %s (
+				id %s,
+				metric_name VARCHAR(191) NOT NULL,
+				entity_id VARCHAR(191) NOT NULL,
+				tags_hash VARCHAR(64) NOT NULL,
+				ts_nano BIGINT NOT NULL,
+				value DOUBLE NOT NULL,
+				tags %s NOT NULL,
+				labels %s NOT NULL,
+				created_at BIGINT NOT NULL,
+				UNIQUE KEY uq_metric_entity_time (metric_name, entity_id, tags_hash, ts_nano),
+				INDEX idx_metric_entity_time (metric_name, entity_id, ts_nano),
+				INDEX idx_metric_time (metric_name, ts_nano),
+				INDEX idx_entity_time (entity_id, ts_nano)
+			) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci`, s.tables.points, pk, jsonType, jsonType),
+			fmt.Sprintf(`CREATE TABLE IF NOT EXISTS %s (
+				id %s,
+				metric_name VARCHAR(191) NOT NULL,
+				entity_id VARCHAR(191) NOT NULL,
+				tags_hash VARCHAR(64) NOT NULL,
+				tags %s NOT NULL,
+				resolution_nano BIGINT NOT NULL,
+				bucket_nano BIGINT NOT NULL,
+				count BIGINT NOT NULL,
+				sum DOUBLE NOT NULL,
+				sum_sq DOUBLE NOT NULL,
+				min_val DOUBLE NOT NULL,
+				max_val DOUBLE NOT NULL,
+				first_val DOUBLE NOT NULL,
+				first_ts BIGINT NOT NULL,
+				last_val DOUBLE NOT NULL,
+				last_ts BIGINT NOT NULL,
+				digest %s,
+				created_at BIGINT NOT NULL,
+				UNIQUE KEY uq_rollup (metric_name, entity_id, tags_hash, resolution_nano, bucket_nano),
+				INDEX idx_rollup_res_time (metric_name, resolution_nano, bucket_nano)
+			) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci`, s.tables.rollups, pk, jsonType, blob),
+		}
+	}
+
+	if s.cfg.Driver == DriverPostgreSQL {
+		statements = []string{
+			fmt.Sprintf(`CREATE TABLE IF NOT EXISTS %s (
+				name VARCHAR(191) PRIMARY KEY,
+				type VARCHAR(32) NOT NULL,
+				unit VARCHAR(64) NOT NULL DEFAULT '',
+				description TEXT NOT NULL DEFAULT '',
+				retention_days INTEGER NOT NULL DEFAULT 0,
+				metadata %s NOT NULL,
+				created_at BIGINT NOT NULL,
+				updated_at BIGINT NOT NULL
+			)`, s.tables.definitions, jsonType),
+			fmt.Sprintf(`CREATE TABLE IF NOT EXISTS %s (
+				id %s,
+				metric_name VARCHAR(191) NOT NULL,
+				entity_id VARCHAR(191) NOT NULL,
+				tags_hash VARCHAR(64) NOT NULL,
+				ts_nano BIGINT NOT NULL,
+				value DOUBLE PRECISION NOT NULL,
+				tags %s NOT NULL,
+				labels %s NOT NULL,
+				created_at BIGINT NOT NULL,
+				UNIQUE(metric_name, entity_id, tags_hash, ts_nano)
+			)`, s.tables.points, pk, jsonType, jsonType),
+			fmt.Sprintf(`CREATE INDEX IF NOT EXISTS %s_points_metric_entity_time_idx ON %s (metric_name, entity_id, ts_nano)`, s.cfg.TablePrefix, s.tables.points),
+			fmt.Sprintf(`CREATE INDEX IF NOT EXISTS %s_points_metric_time_idx ON %s (metric_name, ts_nano)`, s.cfg.TablePrefix, s.tables.points),
+			fmt.Sprintf(`CREATE INDEX IF NOT EXISTS %s_points_entity_time_idx ON %s (entity_id, ts_nano)`, s.cfg.TablePrefix, s.tables.points),
+			fmt.Sprintf(`CREATE TABLE IF NOT EXISTS %s (
+				id %s,
+				metric_name VARCHAR(191) NOT NULL,
+				entity_id VARCHAR(191) NOT NULL,
+				tags_hash VARCHAR(64) NOT NULL,
+				tags %s NOT NULL,
+				resolution_nano BIGINT NOT NULL,
+				bucket_nano BIGINT NOT NULL,
+				count BIGINT NOT NULL,
+				sum DOUBLE PRECISION NOT NULL,
+				sum_sq DOUBLE PRECISION NOT NULL,
+				min_val DOUBLE PRECISION NOT NULL,
+				max_val DOUBLE PRECISION NOT NULL,
+				first_val DOUBLE PRECISION NOT NULL,
+				first_ts BIGINT NOT NULL,
+				last_val DOUBLE PRECISION NOT NULL,
+				last_ts BIGINT NOT NULL,
+				digest %s,
+				created_at BIGINT NOT NULL,
+				UNIQUE(metric_name, entity_id, tags_hash, resolution_nano, bucket_nano)
+			)`, s.tables.rollups, pk, jsonType, blob),
+			fmt.Sprintf(`CREATE INDEX IF NOT EXISTS %s_rollups_lookup_idx ON %s (metric_name, entity_id, tags_hash, resolution_nano, bucket_nano)`, s.cfg.TablePrefix, s.tables.rollups),
+			fmt.Sprintf(`CREATE INDEX IF NOT EXISTS %s_rollups_res_time_idx ON %s (metric_name, resolution_nano, bucket_nano)`, s.cfg.TablePrefix, s.tables.rollups),
+		}
+	}
+
+	for _, stmt := range statements {
+		if _, err := s.db.ExecContext(ctx, stmt); err != nil {
+			return err
+		}
+	}
+	return nil
+}

pkg/metric/percentile.go

@@ -0,0 +1,76 @@
+package metric
+
+import (
+	"strconv"
+	"strings"
+)
+
+// Pxx builds the Aggregation for an arbitrary percentile. The argument is a
+// percentage in (0,100): Pxx(99.9) -> "p99.9", Pxx(50) -> "p50". The fixed
+// AggP50/AggP95/AggP99 constants are just the common cases of this same string
+// form, so they keep working unchanged.
+//
+// This is what turns the package from "p50/p95/p99 only" into "any percentile":
+// callers can ask for p75, p90, p99.99, etc., and every path (in-memory,
+// SQL pushdown, and rollup-via-t-digest) understands it.
+//
+// Pxx 根据任意百分位构造 Aggregation。参数是 (0,100) 内的百分比：
+// Pxx(99.9) -> "p99.9"，Pxx(50) -> "p50"。固定的 AggP50/AggP95/AggP99
+// 常量只是同一字符串形式的常用情况，因此会保持原有行为。
+//
+// 这让 package 从“只支持 p50/p95/p99”变成“支持任意百分位”：调用方可以请求
+// p75、p90、p99.99 等，并且每条路径（内存、SQL 下推、基于 t-digest 的 rollup）
+// 都能理解它。
+func Pxx(p float64) Aggregation {
+	// Trim trailing zeros so Pxx(95) == AggP95 ("p95"), not "p95.000000".
+	s := strconv.FormatFloat(p, 'f', -1, 64)
+	return Aggregation("p" + s)
+}
+
+// parsePercentile reports whether agg names a percentile and, if so, returns
+// the corresponding fraction in [0,1]. "p99.9" -> 0.999. Out-of-range
+// percentages (<=0 or >=100) are rejected so validation can reject them.
+//
+// parsePercentile 判断 agg 是否命名了百分位；如果是，则返回对应的 [0,1] 小数。
+// 例如 "p99.9" -> 0.999。越界百分比（<=0 或 >=100）会被拒绝，以便校验逻辑
+// 能拒绝它们。
+func parsePercentile(agg Aggregation) (float64, bool) {
+	s := string(agg)
+	if len(s) < 2 || (s[0] != 'p' && s[0] != 'P') {
+		return 0, false
+	}
+	pct, err := strconv.ParseFloat(s[1:], 64)
+	if err != nil {
+		return 0, false
+	}
+	if pct <= 0 || pct >= 100 {
+		return 0, false
+	}
+	return pct / 100, true
+}
+
+// isPercentile reports whether agg is any percentile aggregation.
+//
+// isPercentile 判断聚合类型是否为任意百分位聚合。
+func isPercentile(agg Aggregation) bool {
+	_, ok := parsePercentile(agg)
+	return ok
+}
+
+// percentileFractionString renders the fraction for SQL percentile_cont, e.g.
+// "p99.9" -> "0.999". Trailing zeros are trimmed for stable SQL text.
+//
+// percentileFractionString 把百分位聚合转换为 SQL percentile_cont 需要的
+// 小数字符串，并去掉尾随零以保持 SQL 文本稳定。
+func percentileFractionString(agg Aggregation) (string, bool) {
+	f, ok := parsePercentile(agg)
+	if !ok {
+		return "", false
+	}
+	s := strconv.FormatFloat(f, 'f', -1, 64)
+	if !strings.Contains(s, ".") {
+		// f is in (0,1) so this should not happen, but guard anyway.
+		s = strconv.FormatFloat(f, 'f', 1, 64)
+	}
+	return s, true
+}

pkg/metric/postgres_integration_test.go

@@ -0,0 +1,184 @@
+package metric
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"strings"
+	"testing"
+	"time"
+)
+
+// TestPostgreSQLIntegration runs the PostgreSQL integration test when configured.
+//
+// TestPostgreSQLIntegration 在配置 DSN 后运行 PostgreSQL 集成测试。
+func TestPostgreSQLIntegration(t *testing.T) {
+	dsn := os.Getenv("METRIC_POSTGRES_DSN")
+	if dsn == "" {
+		t.Skip("METRIC_POSTGRES_DSN is not set")
+	}
+
+	runSQLIntegration(t, "postgres", PostgreSQL(dsn), true)
+}
+
+// TestMySQLIntegration runs the MySQL integration test when configured.
+//
+// TestMySQLIntegration 在配置 DSN 后运行 MySQL 集成测试。
+func TestMySQLIntegration(t *testing.T) {
+	dsn := os.Getenv("METRIC_MYSQL_DSN")
+	if dsn == "" {
+		t.Skip("METRIC_MYSQL_DSN is not set")
+	}
+
+	runSQLIntegration(t, "mysql", MySQL(dsn), false)
+}
+
+// runSQLIntegration exercises the SQL store against an external database.
+//
+// runSQLIntegration 在外部数据库上执行通用 SQL 集成测试流程。
+func runSQLIntegration(t *testing.T, name string, cfg Config, expectSQLPercentile bool) {
+	t.Helper()
+
+	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+	defer cancel()
+
+	prefix := fmt.Sprintf("it_%d_", time.Now().UnixNano())
+	cfg.TablePrefix = prefix
+	store, err := Open(ctx, cfg)
+	if err != nil {
+		t.Fatalf("open %s store: %v", name, err)
+	}
+	defer store.Close()
+	defer dropIntegrationTables(t, store, prefix)
+
+	if err := store.CreateMetric(ctx, Definition{
+		Name:        "http.latency",
+		Type:        TypeGauge,
+		Unit:        "ms",
+		Description: "HTTP latency",
+		Metadata:    map[string]string{"source": "integration"},
+	}); err != nil {
+		t.Fatalf("create metric: %v", err)
+	}
+	if err := store.CreateMetric(ctx, Definition{Name: "http.latency", Type: TypeGauge}); err == nil {
+		t.Fatalf("duplicate CreateMetric should fail")
+	}
+	if err := store.UpsertMetric(ctx, Definition{Name: "http.latency", Type: TypeGauge, Unit: "milliseconds"}); err != nil {
+		t.Fatalf("upsert metric: %v", err)
+	}
+	def, err := store.GetMetric(ctx, "http.latency")
+	if err != nil {
+		t.Fatalf("get metric: %v", err)
+	}
+	if def.Unit != "milliseconds" {
+		t.Fatalf("upsert did not update unit: %#v", def)
+	}
+
+	base := time.Date(2026, 6, 20, 1, 0, 0, 0, time.UTC)
+	points := []Point{
+		{MetricName: "http.latency", EntityID: "api-1", Timestamp: base, Value: 10, Tags: map[string]string{"region.zone": "ap-1", "route-name": "/v1/nodes"}},
+		{MetricName: "http.latency", EntityID: "api-1", Timestamp: base.Add(time.Minute), Value: 20, Tags: map[string]string{"region.zone": "ap-1", "route-name": "/v1/nodes"}},
+		{MetricName: "http.latency", EntityID: "api-1", Timestamp: base.Add(2 * time.Minute), Value: 30, Tags: map[string]string{"region.zone": "ap-1", "route-name": "/v1/nodes"}},
+		{MetricName: "http.latency", EntityID: "api-1", Timestamp: base.Add(3 * time.Minute), Value: 40, Tags: map[string]string{"region.zone": "eu-1", "route-name": "/v1/nodes"}},
+	}
+	if err := store.WriteBatch(ctx, points); err != nil {
+		t.Fatalf("write batch: %v", err)
+	}
+
+	got, err := store.Query(ctx, Query{
+		MetricName: "http.latency",
+		EntityID:   "api-1",
+		Start:      base.Add(-time.Second),
+		End:        base.Add(5 * time.Minute),
+		Tags:       map[string]string{"region.zone": "ap-1", "route-name": "/v1/nodes"},
+	})
+	if err != nil {
+		t.Fatalf("query with json tags: %v", err)
+	}
+	if len(got) != 3 {
+		t.Fatalf("expected 3 ap-1 points, got %d: %#v", len(got), got)
+	}
+
+	stats, err := store.Stats(ctx, Query{
+		MetricName: "http.latency",
+		EntityID:   "api-1",
+		Start:      base,
+		End:        base.Add(5 * time.Minute),
+	})
+	if err != nil {
+		t.Fatalf("stats: %v", err)
+	}
+	if stats.Count != 4 || stats.Avg != 25 {
+		t.Fatalf("unexpected stats: %#v", stats)
+	}
+
+	avgBuckets, err := store.Aggregate(ctx, AggregateQuery{
+		Query: Query{
+			MetricName: "http.latency",
+			EntityID:   "api-1",
+			Start:      base,
+			End:        base.Add(5 * time.Minute),
+		},
+		Aggregation:  AggAvg,
+		Interval:     2 * time.Minute,
+		BucketLimit:  1,
+		BucketOffset: 1,
+	})
+	if err != nil {
+		t.Fatalf("avg aggregate: %v", err)
+	}
+	if len(avgBuckets) != 1 || avgBuckets[0].Value != 35 || avgBuckets[0].Count != 2 {
+		t.Fatalf("unexpected bucket-paged avg aggregate: %#v", avgBuckets)
+	}
+
+	p95Buckets, err := store.Aggregate(ctx, AggregateQuery{
+		Query: Query{
+			MetricName: "http.latency",
+			EntityID:   "api-1",
+			Start:      base,
+			End:        base.Add(5 * time.Minute),
+		},
+		Aggregation: AggP95,
+		Interval:    10 * time.Minute,
+	})
+	if err != nil {
+		t.Fatalf("p95 aggregate: %v", err)
+	}
+	if len(p95Buckets) != 1 || p95Buckets[0].Count != 4 || p95Buckets[0].Value <= 38 {
+		t.Fatalf("unexpected p95 aggregate: %#v", p95Buckets)
+	}
+	if _, ok := sqlAggValueExpr(cfg.Driver, AggP95); ok != expectSQLPercentile {
+		t.Fatalf("%s p95 pushdown expectation mismatch: got %v want %v", name, ok, expectSQLPercentile)
+	}
+
+	latest, err := store.Latest(ctx, "http.latency", "api-1", 1)
+	if err != nil {
+		t.Fatalf("latest: %v", err)
+	}
+	if len(latest) != 1 || latest[0].Value != 40 {
+		t.Fatalf("unexpected latest point: %#v", latest)
+	}
+
+	deleted, err := store.DeleteBefore(ctx, "http.latency", base.Add(2*time.Minute))
+	if err != nil {
+		t.Fatalf("delete before: %v", err)
+	}
+	if deleted != 2 {
+		t.Fatalf("expected 2 deleted points, got %d", deleted)
+	}
+}
+
+// dropIntegrationTables drops integration-test tables.
+//
+// dropIntegrationTables 删除集成测试创建的表。
+func dropIntegrationTables(t *testing.T, store *Store, prefix string) {
+	t.Helper()
+	if strings.TrimSpace(prefix) == "" {
+		t.Fatal("refusing to drop tables with empty prefix")
+	}
+	for _, name := range []string{prefix + "rollups", prefix + "points", prefix + "definitions"} {
+		if _, err := store.db.Exec("DROP TABLE IF EXISTS " + name); err != nil {
+			t.Fatalf("drop integration table %s: %v", name, err)
+		}
+	}
+}

pkg/metric/rollup.go

@@ -0,0 +1,320 @@
+package metric
+
+import (
+	"fmt"
+	"math"
+	"time"
+)
+
+// RollupTier describes one downsampled resolution: raw points (or the next
+// finer tier) are aggregated into buckets Interval wide, and those buckets are
+// kept for Retention. A policy lists tiers from finest to coarsest, e.g.
+//
+//	1m kept 7d  ->  5m kept 30d  ->  1h kept 1y
+//
+// As data ages it falls off the finer tiers and survives only in coarser ones,
+// so storage shrinks with age — the "data gets sparser as it gets older"
+// behavior of a downsampling TSDB.
+//
+// RollupTier 描述一个降采样分辨率：原始点（或下一层更细的层级）会被聚合进
+// Interval 宽的桶，而这些桶会保留 Retention。策略会按从细到粗排列层级，例如：
+//
+//	1m 保留 7d  ->  5m 保留 30d  ->  1h 保留 1y
+//
+// 随着数据变旧，它会从更细层级中过期，只在更粗层级中保留下来，因此存储量会随
+// 数据年龄下降，也就是降采样 TSDB 中“数据越旧越稀疏”的行为。
+type RollupTier struct {
+	// Interval is the bucket width for this rollup tier.
+	//
+	// Interval 是该 rollup 层的桶宽。
+	Interval time.Duration `json:"interval"`
+	// Retention is how long buckets in this tier are kept.
+	//
+	// Retention 是该层桶的保留时间。
+	Retention time.Duration `json:"retention"`
+}
+
+// RollupPolicy is the full retention ladder for a store: how long raw points
+// live, then a chain of progressively coarser tiers. Compact materializes the
+// tiers and enforces every retention window.
+//
+// RollupPolicy 描述完整保留阶梯：原始点保留多久，以及后续逐级变粗的
+// rollup 层；Compact 会物化这些层并执行保留窗口。
+type RollupPolicy struct {
+	// RawRetention is how long raw points are kept before Compact deletes them
+	// (after they have been rolled into the finest tier). Zero means "never
+	// delete raw" — rollups are still built, but raw is retained.
+	//
+	// RawRetention 是 Compact 删除原始点之前保留它们的时间（在它们已进入最细
+	// rollup 层之后）。零值表示“永不删除原始点”；rollup 仍会构建，但原始点保留。
+	RawRetention time.Duration `json:"raw_retention"`
+	// Tiers are ordered finest-first. Each Interval must be a positive integer
+	// multiple of the previous tier's Interval (so a coarse bucket is composed
+	// of whole finer buckets), and each Retention must be >= the previous
+	// tier's Retention (coarse data outlives fine data).
+	//
+	// Tiers 按从细到粗排序。每个 Interval 必须是前一层 Interval 的正整数倍
+	// （这样粗桶由完整细桶组成），每个 Retention 必须 >= 前一层 Retention
+	// （粗数据比细数据活得更久）。
+	Tiers []RollupTier `json:"tiers"`
+	// Compression tunes the per-bucket t-digest (size vs. percentile accuracy).
+	// <=1 uses the default (100).
+	//
+	// Compression 调整每个桶的 t-digest（大小与百分位精度之间的取舍）。
+	// <=1 时使用默认值（100）。
+	Compression float64 `json:"compression"`
+}
+
+// Enabled reports whether the policy actually defines any rollup tiers.
+//
+// Enabled 表示策略是否实际定义了 rollup 层级。
+func (p RollupPolicy) Enabled() bool { return len(p.Tiers) > 0 }
+
+// compression returns the effective t-digest compression for a rollup policy.
+//
+// compression 返回 rollup 策略实际使用的 t-digest 压缩参数。
+func (p RollupPolicy) compression() float64 {
+	if p.Compression <= 1 {
+		return defaultTDigestCompression
+	}
+	return p.Compression
+}
+
+func (p RollupPolicy) rawCutoff(now time.Time) time.Time {
+	if p.RawRetention <= 0 || len(p.Tiers) == 0 {
+		return time.Time{}
+	}
+	cutoff := now.UTC().Add(-p.RawRetention)
+	interval := p.Tiers[0].Interval.Nanoseconds()
+	return time.Unix(0, floorDivNano(cutoff.UnixNano(), interval)).UTC()
+}
+
+// Validate enforces the structural rules that make cascading composition and
+// retention well-defined.
+//
+// Validate 检查策略结构是否满足级联合成和保留语义所需的约束。
+func (p RollupPolicy) Validate() error {
+	if len(p.Tiers) == 0 {
+		return nil // a store with no tiers simply does no rollup work
+	}
+	if p.RawRetention < 0 {
+		return fmt.Errorf("%w: raw retention cannot be negative", ErrInvalidArgument)
+	}
+	var prev RollupTier
+	for i, tr := range p.Tiers {
+		if tr.Interval <= 0 {
+			return fmt.Errorf("%w: tier %d interval must be positive", ErrInvalidArgument, i)
+		}
+		if tr.Retention <= 0 {
+			return fmt.Errorf("%w: tier %d retention must be positive", ErrInvalidArgument, i)
+		}
+		if i == 0 {
+			// The finest tier rolls up raw points. Require raw retention (when
+			// set) to cover at least two of its buckets so the in-progress and
+			// most-recent sealed bucket still have their raw backing when Compact
+			// runs and deletes old raw.
+			if p.RawRetention > 0 && p.RawRetention < 2*tr.Interval {
+				return fmt.Errorf("%w: raw retention must be >= 2x the finest tier interval", ErrInvalidArgument)
+			}
+		} else {
+			if tr.Interval <= prev.Interval {
+				return fmt.Errorf("%w: tier %d interval must be larger than tier %d", ErrInvalidArgument, i, i-1)
+			}
+			if tr.Interval%prev.Interval != 0 {
+				return fmt.Errorf("%w: tier %d interval must be a multiple of tier %d interval", ErrInvalidArgument, i, i-1)
+			}
+			if tr.Retention < prev.Retention {
+				return fmt.Errorf("%w: tier %d retention must be >= tier %d retention", ErrInvalidArgument, i, i-1)
+			}
+		}
+		prev = tr
+	}
+	return nil
+}
+
+// rollupBucket is the in-memory accumulator for one (metric, entity,
+// resolution, bucket) cell. It carries exactly the summaries that can be
+// re-aggregated losslessly when composing coarser tiers — count/sum/sumSq for
+// avg & population stddev, min/max, first/last by timestamp — plus a t-digest
+// so arbitrary percentiles survive downsampling with bounded error.
+//
+// rollupBucket 是单个（指标、实体、分辨率、桶）单元的内存累加器。它只携带在
+// 合成更粗层级时可无损重新聚合的摘要：用于平均值和总体标准差的 count/sum/sumSq，
+// min/max，按时间记录的 first/last，外加一个 t-digest，让任意百分位能以
+// 有界误差在降采样后保留下来。
+type rollupBucket struct {
+	// count is the total number of raw points represented.
+	//
+	// count 是该桶代表的原始点总数。
+	count int64
+	// sum is the sum of represented values.
+	//
+	// sum 是该桶代表值的总和。
+	sum float64
+	// sumSq is the sum of squared values for population stddev.
+	//
+	// sumSq 是用于总体标准差的平方和。
+	sumSq float64
+	// min is the minimum represented value.
+	//
+	// min 是该桶代表的最小值。
+	min float64
+	// max is the maximum represented value.
+	//
+	// max 是该桶代表的最大值。
+	max float64
+	// firstVal is the value with the earliest timestamp.
+	//
+	// firstVal 是时间戳最早的值。
+	firstVal float64
+	// firstTS is the earliest timestamp in nanoseconds.
+	//
+	// firstTS 是最早时间戳的纳秒值。
+	firstTS int64
+	// lastVal is the value with the latest timestamp.
+	//
+	// lastVal 是时间戳最晚的值。
+	lastVal float64
+	// lastTS is the latest timestamp in nanoseconds.
+	//
+	// lastTS 是最晚时间戳的纳秒值。
+	lastTS int64
+	// digest estimates percentiles for represented values.
+	//
+	// digest 用于估算该桶代表值的百分位。
+	digest *TDigest
+	// tagsHash carries the stable tag-set fingerprint for this rollup cell.
+	//
+	// tagsHash 携带该 rollup 单元的稳定标签集合指纹。
+	tagsHash string
+	// tagsJSON carries the canonical tag map written back to the tags column.
+	//
+	// tagsJSON 携带写回 tags 列的规范标签 map。
+	tagsJSON string
+}
+
+// newRollupBucket creates an empty rollup accumulator.
+//
+// newRollupBucket 创建一个空的 rollup 累加器。
+func newRollupBucket(compression float64) *rollupBucket {
+	return &rollupBucket{
+		min:     0,
+		max:     0,
+		firstTS: 0,
+		lastTS:  0,
+		digest:  NewTDigest(compression),
+	}
+}
+
+// addPoint folds a raw observation into the bucket.
+//
+// addPoint 将一个原始观测值合入当前桶。
+func (b *rollupBucket) addPoint(value float64, tsNano int64) {
+	if b.count == 0 {
+		b.min, b.max = value, value
+		b.firstVal, b.firstTS = value, tsNano
+		b.lastVal, b.lastTS = value, tsNano
+	} else {
+		if value < b.min {
+			b.min = value
+		}
+		if value > b.max {
+			b.max = value
+		}
+		if tsNano < b.firstTS {
+			b.firstVal, b.firstTS = value, tsNano
+		}
+		if tsNano > b.lastTS {
+			b.lastVal, b.lastTS = value, tsNano
+		}
+	}
+	b.count++
+	b.sum += value
+	b.sumSq += value * value
+	b.digest.Add(value, 1)
+}
+
+// mergeStored folds a finer rollup row (already-summarized) into this coarser
+// bucket. This is the cascade step: tier i+1 buckets are built by merging the
+// tier i rows they span.
+//
+// mergeStored 将一个已汇总的细层 rollup 行合入当前更粗桶。这是级联步骤：
+// tier i+1 的桶通过合并其覆盖的 tier i 行构建。
+func (b *rollupBucket) mergeStored(o *rollupBucket) {
+	if o.count == 0 {
+		return
+	}
+	if b.count == 0 {
+		b.min, b.max = o.min, o.max
+		b.firstVal, b.firstTS = o.firstVal, o.firstTS
+		b.lastVal, b.lastTS = o.lastVal, o.lastTS
+	} else {
+		if o.min < b.min {
+			b.min = o.min
+		}
+		if o.max > b.max {
+			b.max = o.max
+		}
+		if o.firstTS < b.firstTS {
+			b.firstVal, b.firstTS = o.firstVal, o.firstTS
+		}
+		if o.lastTS > b.lastTS {
+			b.lastVal, b.lastTS = o.lastVal, o.lastTS
+		}
+	}
+	b.count += o.count
+	b.sum += o.sum
+	b.sumSq += o.sumSq
+	if b.digest == nil {
+		b.digest = NewTDigest(defaultTDigestCompression)
+	}
+	b.digest.Merge(o.digest)
+}
+
+// value computes the requested aggregation from the bucket summaries. ok=false
+// means the aggregation is not derivable from a rollup (only AggRate, which
+// needs the ordered raw series).
+//
+// value 从桶摘要中计算请求的聚合值；ok=false 表示该聚合无法由 rollup 推导
+// （目前主要是需要有序原始序列的 AggRate）。
+func (b *rollupBucket) value(agg Aggregation) (float64, bool) {
+	switch agg {
+	case AggAvg:
+		if b.count == 0 {
+			return 0, true
+		}
+		return b.sum / float64(b.count), true
+	case AggMin:
+		return b.min, true
+	case AggMax:
+		return b.max, true
+	case AggSum:
+		return b.sum, true
+	case AggCount:
+		return float64(b.count), true
+	case AggFirst:
+		return b.firstVal, true
+	case AggLast:
+		return b.lastVal, true
+	case AggStdDev:
+		if b.count == 0 {
+			return 0, true
+		}
+		mean := b.sum / float64(b.count)
+		variance := b.sumSq/float64(b.count) - mean*mean
+		if variance < 0 {
+			variance = 0 // floating-point guard
+		}
+		return math.Sqrt(variance), true
+	case AggRate:
+		return 0, false // rate needs the ordered raw series; not in a rollup
+	default:
+		if frac, ok := parsePercentile(agg); ok {
+			if b.digest == nil || b.digest.Count() == 0 {
+				return 0, true
+			}
+			return b.digest.Quantile(frac), true
+		}
+		return 0, false
+	}
+}

pkg/metric/rollup_boundary2_test.go

@@ -0,0 +1,135 @@
+package metric
+
+import (
+	"context"
+	"math"
+	"testing"
+	"time"
+)
+
+// TestSeriesHybridKeepsUncompactedRawNearCutoff verifies the hybrid read does
+// not lose raw points that fall between the (finest-aligned) raw cutoff and the
+// next bucket boundary, including a point written AFTER the last compaction.
+//
+// Reproduction: now=12:00:30, RawRetention=30m so the cutoff is 11:30:30, which
+// the policy floors to 11:30:00 (finest = 1m). A point at 11:30:15 written
+// before compaction is kept (>= aligned cutoff) and NOT rolled up; a point at
+// 11:30:45 written after compaction is also raw. Both belong to the 11:00 1h
+// output bucket. The old code aligned the split UP to 11:31:00 and read raw from
+// there, dropping both boundary points (count=1, the stale rollup half only).
+// Serving raw from the aligned cutoff fixes it (count=2).
+//
+// TestSeriesHybridKeepsUncompactedRawNearCutoff 验证混合读取不会丢失落在
+// （最细对齐后的）原始 cutoff 与下一个桶边界之间的 raw 点，包括上次 compact 之后
+// 才写入的点。
+func TestSeriesHybridKeepsUncompactedRawNearCutoff(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 30 * time.Minute,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "nearcut", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	now := time.Date(2026, 6, 18, 12, 0, 30, 0, time.UTC)
+	// 11:30:15 is >= the finest-aligned cutoff (11:30:00) -> survives compaction
+	// as raw and is not rolled up.
+	before := time.Date(2026, 6, 18, 11, 30, 15, 0, time.UTC)
+	if err := s.Write(ctx, Point{MetricName: "nearcut", EntityID: "n1", Timestamp: before, Value: 10}); err != nil {
+		t.Fatalf("write before: %v", err)
+	}
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+	// Written after compaction: only ever lives in raw.
+	after := time.Date(2026, 6, 18, 11, 30, 45, 0, time.UTC)
+	if err := s.Write(ctx, Point{MetricName: "nearcut", EntityID: "n1", Timestamp: after, Value: 100}); err != nil {
+		t.Fatalf("write after: %v", err)
+	}
+
+	got, err := s.Series(ctx, AggregateQuery{
+		Query:       Query{MetricName: "nearcut", EntityID: "n1", Start: time.Date(2026, 6, 18, 11, 0, 0, 0, time.UTC), End: now},
+		Aggregation: AggAvg,
+		Interval:    time.Hour,
+	}, now)
+	if err != nil {
+		t.Fatalf("series: %v", err)
+	}
+	bucket := alignTime(time.Date(2026, 6, 18, 11, 0, 0, 0, time.UTC), time.Hour)
+	var found bool
+	for _, p := range got {
+		if p.Bucket.Equal(bucket) {
+			found = true
+			if p.Count != 2 {
+				t.Fatalf("near-cutoff bucket count: got %d want 2 (%#v)", p.Count, p)
+			}
+			if math.Abs(p.Value-55) > 1e-9 {
+				t.Fatalf("near-cutoff bucket avg: got %v want 55 (%#v)", p.Value, p)
+			}
+		}
+	}
+	if !found {
+		t.Fatalf("expected 11:00 bucket with both raw points: %#v", got)
+	}
+}
+
+// TestSeriesHybridEndEqualsCutoff verifies a query whose End lands exactly on the
+// raw cutoff still includes the raw point sitting at the cutoff, instead of
+// returning empty.
+//
+// Reproduction: now=12:00:00, RawRetention=30m -> cutoff 11:30:00. A point at
+// exactly 11:30:00 is kept by compaction (not < cutoff) and lives in raw. An old
+// point at 11:00:10 is rolled up and its raw deleted. Querying [11:00, 11:30:00]
+// (End == cutoff) must combine the rolled-up old point and the raw cutoff point.
+// The old entry gate required End strictly AFTER the cutoff, so this fell through
+// to the pure-rollup path and returned the cutoff point's bucket empty.
+//
+// TestSeriesHybridEndEqualsCutoff 验证 End 恰好等于原始 cutoff 的查询仍会包含
+// 落在 cutoff 上的 raw 点，而不是返回空。
+func TestSeriesHybridEndEqualsCutoff(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 30 * time.Minute,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "endcut", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	now := time.Date(2026, 6, 18, 12, 0, 0, 0, time.UTC)
+	cutoff := time.Date(2026, 6, 18, 11, 30, 0, 0, time.UTC)
+	old := time.Date(2026, 6, 18, 11, 0, 10, 0, time.UTC) // rolled up, raw deleted
+	if err := s.Write(ctx, Point{MetricName: "endcut", EntityID: "n1", Timestamp: old, Value: 10}); err != nil {
+		t.Fatalf("write old: %v", err)
+	}
+	// Point exactly at the cutoff: kept as raw (not < cutoff) after compaction.
+	if err := s.Write(ctx, Point{MetricName: "endcut", EntityID: "n1", Timestamp: cutoff, Value: 100}); err != nil {
+		t.Fatalf("write cutoff: %v", err)
+	}
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+
+	got, err := s.Series(ctx, AggregateQuery{
+		Query:       Query{MetricName: "endcut", EntityID: "n1", Start: time.Date(2026, 6, 18, 11, 0, 0, 0, time.UTC), End: cutoff},
+		Aggregation: AggSum,
+		Interval:    time.Hour,
+	}, now)
+	if err != nil {
+		t.Fatalf("series: %v", err)
+	}
+	bucket := alignTime(time.Date(2026, 6, 18, 11, 0, 0, 0, time.UTC), time.Hour)
+	var found bool
+	for _, p := range got {
+		if p.Bucket.Equal(bucket) {
+			found = true
+			if p.Count != 2 || math.Abs(p.Value-110) > 1e-9 {
+				t.Fatalf("End==cutoff bucket: got count=%d value=%v want count=2 value=110", p.Count, p.Value)
+			}
+		}
+	}
+	if !found {
+		t.Fatalf("expected 11:00 bucket including the cutoff raw point: %#v", got)
+	}
+}

pkg/metric/rollup_boundary_test.go

@@ -0,0 +1,347 @@
+package metric
+
+import (
+	"context"
+	"math"
+	"testing"
+	"time"
+)
+
+// TestAggregateRollupExcludesPartialBucket verifies the rollup query no longer
+// drags an out-of-window sample in through a partially-overlapping bucket.
+//
+// Two samples share one 1m rollup bucket [00:00:00, 00:01:00): 00:00:10=100 and
+// 00:00:50=10. A query window of [00:00:40, 00:01:30] only overlaps part of that
+// bucket. Because a rollup bucket is an indivisible summary, the old "any
+// overlap" rule folded the whole bucket in and counted 00:00:10 (value 100),
+// which lies OUTSIDE the window. Full containment excludes the straddling
+// bucket entirely, so the out-of-window sample can no longer leak in.
+//
+// TestAggregateRollupExcludesPartialBucket 验证 rollup 查询不再通过部分重叠的桶
+// 把窗口外的样本带进来。
+func TestAggregateRollupExcludesPartialBucket(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{Tiers: []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}}}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "win", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	batch := []Point{
+		{MetricName: "win", EntityID: "n1", Timestamp: base.Add(10 * time.Second), Value: 100}, // out of window
+		{MetricName: "win", EntityID: "n1", Timestamp: base.Add(50 * time.Second), Value: 10},  // in window
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(time.Hour)); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+
+	// Window [00:00:40, 00:01:30] partially overlaps the single 1m bucket.
+	res, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query:       Query{MetricName: "win", EntityID: "n1", Start: base.Add(40 * time.Second), End: base.Add(90 * time.Second)},
+		Aggregation: AggSum,
+		Interval:    time.Minute,
+	}, time.Minute)
+	if err != nil {
+		t.Fatalf("aggregate rollup: %v", err)
+	}
+	// The partially-overlapping bucket is excluded under full containment, so the
+	// out-of-window 00:00:10 sample (value 100) must not appear. Concretely: no
+	// returned bucket may carry the out-of-window value or a count of 2.
+	for _, p := range res {
+		if p.Count >= 2 || p.Value == 100 {
+			t.Fatalf("partial bucket leaked out-of-window sample: %#v", p)
+		}
+	}
+}
+
+// TestAggregateRollupEndBoundaryInclusive verifies the rollup query end boundary
+// is inclusive (matching the raw query's ts <= end) and that a window aligned to
+// resolution boundaries returns every whole bucket it covers.
+//
+// Buckets at 00:00 and 00:01 are both whole-contained in [00:00:00, 00:02:00]:
+// the second bucket's window is [00:01:00, 00:02:00), whose last covered nano is
+// < 00:02:00, so an inclusive end keeps it. The previous half-open (< end) rule
+// could drop a bucket sitting exactly on the end edge.
+//
+// TestAggregateRollupEndBoundaryInclusive 验证 rollup 查询 end 边界为闭区间
+// （与 raw 的 ts <= end 一致），且对齐到分辨率边界的窗口会返回它覆盖的每个完整桶。
+func TestAggregateRollupEndBoundaryInclusive(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{Tiers: []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}}}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "edge", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	batch := []Point{
+		{MetricName: "edge", EntityID: "n1", Timestamp: base.Add(10 * time.Second), Value: 1},             // bucket 00:00
+		{MetricName: "edge", EntityID: "n1", Timestamp: base.Add(time.Minute + 10*time.Second), Value: 2}, // bucket 00:01
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(time.Hour)); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+
+	res, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query:       Query{MetricName: "edge", EntityID: "n1", Start: base, End: base.Add(2 * time.Minute)},
+		Aggregation: AggSum,
+		Interval:    time.Minute,
+	}, time.Minute)
+	if err != nil {
+		t.Fatalf("aggregate rollup: %v", err)
+	}
+	if len(res) != 2 {
+		t.Fatalf("aligned window should return both whole buckets, got %d: %#v", len(res), res)
+	}
+	if res[0].Value != 1 || res[1].Value != 2 {
+		t.Fatalf("unexpected bucket values: %#v", res)
+	}
+}
+
+// TestSeriesHybridMergesStraddlingBucket verifies a query output bucket that
+// straddles the raw-retention boundary aggregates BOTH its rollup half and its
+// raw half, instead of letting raw replace the rollup half.
+//
+// With a 30m raw retention and now=12:00 the boundary is 11:30. A 1h output
+// bucket [11:00, 12:00) therefore straddles it. An 11:05 sample (value 10, older
+// than the boundary -> rolled up, raw deleted) and an 11:35 sample (value 100,
+// within retention -> still raw) both belong to that 11:00 bucket. The correct
+// answer folds both: count=2, avg=(10+100)/2=55. The old merge-by-bucket-time
+// path kept only the raw half and reported count=1, avg=100.
+//
+// TestSeriesHybridMergesStraddlingBucket 验证跨越原始保留期边界的输出桶会同时
+// 聚合 rollup 半边和 raw 半边，而不是让 raw 覆盖 rollup 半边。
+func TestSeriesHybridMergesStraddlingBucket(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 30 * time.Minute,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "straddle", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	now := time.Date(2026, 6, 18, 12, 0, 0, 0, time.UTC)
+	// 11:05 is older than the 11:30 cutoff (rolled up, raw deleted on compact);
+	// 11:35 is within retention (kept raw).
+	old := now.Add(-55 * time.Minute)    // 11:05
+	recent := now.Add(-25 * time.Minute) // 11:35
+	batch := []Point{
+		{MetricName: "straddle", EntityID: "n1", Timestamp: old, Value: 10},
+		{MetricName: "straddle", EntityID: "n1", Timestamp: recent, Value: 100},
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+
+	// Confirm the old raw point was deleted and the recent one survives, so the
+	// two halves genuinely come from different sources.
+	oldRaw, err := s.Query(ctx, Query{MetricName: "straddle", EntityID: "n1", Start: old.Add(-time.Minute), End: old.Add(time.Minute)})
+	if err != nil {
+		t.Fatalf("query old raw: %v", err)
+	}
+	if len(oldRaw) != 0 {
+		t.Fatalf("expected old raw point deleted by retention, got %d", len(oldRaw))
+	}
+
+	got, err := s.Series(ctx, AggregateQuery{
+		Query:       Query{MetricName: "straddle", EntityID: "n1", Start: now.Add(-time.Hour), End: now},
+		Aggregation: AggAvg,
+		Interval:    time.Hour,
+	}, now)
+	if err != nil {
+		t.Fatalf("series: %v", err)
+	}
+
+	bucket := alignTime(now.Add(-time.Hour), time.Hour) // 11:00
+	var found bool
+	for _, p := range got {
+		if p.Bucket.Equal(bucket) {
+			found = true
+			if p.Count != 2 {
+				t.Fatalf("straddling bucket count: got %d want 2 (%#v)", p.Count, p)
+			}
+			if math.Abs(p.Value-55) > 1e-9 {
+				t.Fatalf("straddling bucket avg: got %v want 55 (%#v)", p.Value, p)
+			}
+		}
+	}
+	if !found {
+		t.Fatalf("expected a merged 11:00 bucket in result: %#v", got)
+	}
+}
+
+// TestSeriesHybridSumAcrossBoundary is the AggSum analogue: the straddling 1h
+// bucket must sum both halves (10 + 100 = 110), proving the merge is at the
+// summary level rather than a bucket-time replacement.
+//
+// TestSeriesHybridSumAcrossBoundary 是 AggSum 版本：跨边界的 1h 桶必须把两半求和
+// （10 + 100 = 110），证明合并发生在摘要层面，而不是按桶时间替换。
+func TestSeriesHybridSumAcrossBoundary(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 30 * time.Minute,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "sumstraddle", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	now := time.Date(2026, 6, 18, 12, 0, 0, 0, time.UTC)
+	batch := []Point{
+		{MetricName: "sumstraddle", EntityID: "n1", Timestamp: now.Add(-55 * time.Minute), Value: 10},
+		{MetricName: "sumstraddle", EntityID: "n1", Timestamp: now.Add(-25 * time.Minute), Value: 100},
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+	got, err := s.Series(ctx, AggregateQuery{
+		Query:       Query{MetricName: "sumstraddle", EntityID: "n1", Start: now.Add(-time.Hour), End: now},
+		Aggregation: AggSum,
+		Interval:    time.Hour,
+	}, now)
+	if err != nil {
+		t.Fatalf("series: %v", err)
+	}
+	bucket := alignTime(now.Add(-time.Hour), time.Hour)
+	var found bool
+	for _, p := range got {
+		if p.Bucket.Equal(bucket) {
+			found = true
+			if p.Count != 2 || math.Abs(p.Value-110) > 1e-9 {
+				t.Fatalf("straddling sum bucket: got count=%d value=%v want count=2 value=110", p.Count, p.Value)
+			}
+		}
+	}
+	if !found {
+		t.Fatalf("expected merged sum bucket: %#v", got)
+	}
+}
+
+func TestSeriesHybridIncludesRawBetweenCutoffAndNextBucket(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 30 * time.Minute,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "cutgap", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+
+	now := time.Date(2026, 6, 18, 12, 0, 30, 0, time.UTC)
+	old := time.Date(2026, 6, 18, 11, 30, 10, 0, time.UTC)
+	if err := s.Write(ctx, Point{MetricName: "cutgap", EntityID: "n1", Timestamp: old, Value: 10}); err != nil {
+		t.Fatalf("write old: %v", err)
+	}
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+	recentUncompacted := time.Date(2026, 6, 18, 11, 30, 45, 0, time.UTC)
+	if err := s.Write(ctx, Point{MetricName: "cutgap", EntityID: "n1", Timestamp: recentUncompacted, Value: 100}); err != nil {
+		t.Fatalf("write recent: %v", err)
+	}
+
+	got, err := s.Series(ctx, AggregateQuery{
+		Query:       Query{MetricName: "cutgap", EntityID: "n1", Start: time.Date(2026, 6, 18, 11, 0, 0, 0, time.UTC), End: now},
+		Aggregation: AggCount,
+		Interval:    time.Hour,
+	}, now)
+	if err != nil {
+		t.Fatalf("series: %v", err)
+	}
+	if len(got) != 1 || got[0].Count != 2 {
+		t.Fatalf("series should include rollup plus raw at the aligned cutoff, got %#v", got)
+	}
+}
+
+func TestSeriesHybridEndAtCutoffIncludesRawBoundary(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 30 * time.Minute,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "cutedge", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+
+	now := time.Date(2026, 6, 18, 12, 0, 30, 0, time.UTC)
+	cutoff := policy.rawCutoff(now)
+	if err := s.Write(ctx, Point{MetricName: "cutedge", EntityID: "n1", Timestamp: cutoff, Value: 7}); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+
+	got, err := s.Series(ctx, AggregateQuery{
+		Query:       Query{MetricName: "cutedge", EntityID: "n1", Start: cutoff.Add(-time.Hour), End: cutoff},
+		Aggregation: AggCount,
+		Interval:    time.Hour,
+	}, now)
+	if err != nil {
+		t.Fatalf("series: %v", err)
+	}
+	var total int
+	for _, p := range got {
+		total += p.Count
+	}
+	if total != 1 {
+		t.Fatalf("series should include raw point exactly at cutoff/end, got %#v", got)
+	}
+}
+
+func TestCompactDoesNotDoubleCountRawBucketAfterCutoffAdvances(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 30 * time.Minute,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "nodup", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+
+	now := time.Date(2026, 6, 18, 12, 0, 30, 0, time.UTC)
+	if err := s.Write(ctx, Point{MetricName: "nodup", EntityID: "n1", Timestamp: time.Date(2026, 6, 18, 11, 30, 10, 0, time.UTC), Value: 10}); err != nil {
+		t.Fatalf("write old: %v", err)
+	}
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact 1: %v", err)
+	}
+	if err := s.Write(ctx, Point{MetricName: "nodup", EntityID: "n1", Timestamp: time.Date(2026, 6, 18, 11, 30, 45, 0, time.UTC), Value: 100}); err != nil {
+		t.Fatalf("write recent: %v", err)
+	}
+	now = now.Add(time.Minute)
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact 2: %v", err)
+	}
+
+	got, err := s.Series(ctx, AggregateQuery{
+		Query:       Query{MetricName: "nodup", EntityID: "n1", Start: time.Date(2026, 6, 18, 11, 0, 0, 0, time.UTC), End: now},
+		Aggregation: AggCount,
+		Interval:    time.Hour,
+	}, now)
+	if err != nil {
+		t.Fatalf("series: %v", err)
+	}
+	var total int
+	for _, p := range got {
+		total += p.Count
+	}
+	if total != 2 {
+		t.Fatalf("compaction should not double-count a bucket after cutoff advances, got %#v", got)
+	}
+}

pkg/metric/rollup_query.go

@@ -0,0 +1,503 @@
+package metric
+
+import (
+	"context"
+	"database/sql"
+	"fmt"
+	"sort"
+	"strings"
+	"time"
+)
+
+// AggregateRollup answers an AggregateQuery from a stored rollup tier instead of
+// raw points. resolution names which tier to read (it must match a tier
+// Interval that Compact has materialized). The query Interval must be a positive
+// integer multiple of resolution, so each output bucket is composed of whole
+// rollup buckets.
+//
+// query.Tags is honored: only rollup series whose stored tag set matches the
+// filter are folded in, so a tag filter selects the same data it would over raw
+// points (each distinct tag combination is its own rollup series). With no tag
+// filter, all tag series within a bucket merge together, matching the raw path's
+// cross-tag aggregation.
+//
+// Every aggregation works except AggRate, which needs the ordered raw series and
+// is therefore raw-only. Percentiles (p50, p95, p99, and arbitrary pXX) are
+// answered by merging the per-bucket t-digests, so they survive downsampling
+// with bounded error.
+//
+// AggregateRollup 从已存储的 rollup 层回答 AggregateQuery，而不是读取原始点。
+// resolution 指定要读取的层级（必须匹配 Compact 已物化的某个层级 Interval）。
+// 查询的 Interval 必须是 resolution 的正整数倍，因此每个输出桶都由完整的
+// rollup 桶组成。
+//
+// query.Tags 会被遵守：只有存储标签集合匹配过滤条件的 rollup 序列会被合入，
+// 因此标签过滤会选中与原始点查询相同的数据（每个不同标签组合都是自己的
+// rollup 序列）。没有标签过滤时，同一桶内的所有标签序列会合并在一起，
+// 与原始路径的跨标签聚合一致。
+//
+// 除 AggRate 外，每种聚合都可用；AggRate 需要有序原始序列，因此只能基于原始点。
+// 百分位（p50、p95、p99 和任意 pXX）通过合并每桶 t-digest 回答，因此能在
+// 降采样后以有界误差保留下来。
+func (s *Store) AggregateRollup(ctx context.Context, query AggregateQuery, resolution time.Duration) ([]AggregatePoint, error) {
+	if err := s.ensureOpen(); err != nil {
+		return nil, err
+	}
+	if err := query.Validate(); err != nil {
+		return nil, err
+	}
+	if resolution <= 0 {
+		return nil, fmt.Errorf("%w: rollup resolution must be positive", ErrInvalidArgument)
+	}
+	if query.Interval < resolution || query.Interval%resolution != 0 {
+		return nil, fmt.Errorf("%w: query interval must be a positive multiple of the rollup resolution", ErrInvalidArgument)
+	}
+	if query.Aggregation == AggRate {
+		return nil, fmt.Errorf("%w: rate is not derivable from rollups (raw only)", ErrInvalidArgument)
+	}
+
+	q := query.Query.normalized()
+	comp := s.cfg.RollupPolicy.compression()
+
+	// Read rollup buckets at this resolution that are FULLY contained in the
+	// inclusive window [Start, End] (entity and tag filters pushed into SQL),
+	// then fold them into query.Interval-wide output buckets. Full containment
+	// (rather than mere overlap) is what keeps a partially-overlapping bucket's
+	// out-of-window samples from leaking into the result: a rollup bucket is an
+	// indivisible summary, so a bucket straddling a window edge cannot be
+	// trimmed to the window and is therefore excluded. Callers that need every
+	// sample in a sub-bucket window must align the window to resolution
+	// boundaries (or query raw points).
+	rows, err := s.scanRollupRowsContained(ctx, q.MetricName, q.EntityID, q.Tags, resolution, q.Start, q.End)
+	if err != nil {
+		return nil, err
+	}
+	groups := foldRollupRows(nil, rows, query.Interval, comp)
+
+	out, err := rollupGroupsToPoints(groups, query)
+	if err != nil {
+		return nil, err
+	}
+	return pageBuckets(out, query.BucketLimit, query.BucketOffset), nil
+}
+
+// rollupGroupsToPoints turns the merged output buckets into ordered
+// AggregatePoints, computing the requested aggregation from each bucket's
+// summaries/digest. FillEmpty emits zero-count buckets for gaps, mirroring the
+// raw AggregatePoints behavior.
+//
+// rollupGroupsToPoints 将合并后的输出桶转换为有序 AggregatePoint，并根据
+// 每个桶的摘要或 digest 计算请求的聚合。FillEmpty 会为空洞输出零计数桶，
+// 与原始 AggregatePoints 行为一致。
+func rollupGroupsToPoints(groups map[int64]*rollupBucket, query AggregateQuery) ([]AggregatePoint, error) {
+	var bucketStarts []int64
+	if query.FillEmpty {
+		for t := alignTime(query.Start, query.Interval); !t.After(query.End); t = t.Add(query.Interval) {
+			bucketStarts = append(bucketStarts, t.UnixNano())
+		}
+	} else {
+		for k := range groups {
+			bucketStarts = append(bucketStarts, k)
+		}
+		sort.Slice(bucketStarts, func(i, j int) bool { return bucketStarts[i] < bucketStarts[j] })
+	}
+
+	out := make([]AggregatePoint, 0, len(bucketStarts))
+	for _, start := range bucketStarts {
+		b := groups[start]
+		if b == nil {
+			out = append(out, AggregatePoint{
+				MetricName: query.MetricName,
+				EntityID:   query.EntityID,
+				Bucket:     time.Unix(0, start).UTC(),
+			})
+			continue
+		}
+		v, ok := b.value(query.Aggregation)
+		if !ok {
+			return nil, fmt.Errorf("%w: aggregation %q not supported over rollups", ErrInvalidArgument, query.Aggregation)
+		}
+		out = append(out, AggregatePoint{
+			MetricName: query.MetricName,
+			EntityID:   query.EntityID,
+			Bucket:     time.Unix(0, start).UTC(),
+			Value:      v,
+			Count:      int(b.count),
+		})
+	}
+	return out, nil
+}
+
+// scanRollupRowsContained loads rollup rows for one resolution whose whole
+// bucket window [bucket, bucket+resolution) lies inside the inclusive query
+// window [start, end], with optional entity and tag filters pushed into SQL.
+//
+// Full containment (not mere overlap) is the boundary rule that keeps a rollup
+// query from over-counting: a rollup bucket is an indivisible summary, so a
+// bucket that only partially overlaps the window would drag its out-of-window
+// samples in if it were included. The end boundary is inclusive to match the
+// raw query semantics (raw uses ts <= end); a bucket whose last covered nano is
+// exactly end is therefore still contained. A degenerate zero-width window
+// (start == end) contains no whole bucket and yields an empty result — callers
+// that need sub-bucket precision must query raw points.
+//
+// scanRollupRowsContained 读取某分辨率下整桶窗口 [bucket, bucket+resolution)
+// 完整落在闭区间 [start, end] 内的 rollup 行，并可把实体和标签过滤下推到 SQL。
+//
+// 采用“完整包含”（而非仅重叠）作为边界规则，避免 rollup 查询过度计数：rollup
+// 桶是不可分割的摘要，只与窗口部分重叠的桶若被纳入，会把窗口外的样本一起带入。
+// end 边界为闭区间，以匹配原始查询语义（raw 使用 ts <= end）；因此最后覆盖纳秒
+// 恰为 end 的桶仍算被包含。零宽窗口（start == end）不包含任何完整桶，返回空结果，
+// 需要亚桶精度的调用方应查询原始点。
+func (s *Store) scanRollupRowsContained(ctx context.Context, metricName, entityID string, tags map[string]string, resolution time.Duration, start, end time.Time) ([]storedRollup, error) {
+	resNano := resolution.Nanoseconds()
+	startNano := start.UTC().UnixNano()
+	endNano := end.UTC().UnixNano()
+	// A fully-contained bucket has start >= startNano and end (inclusive,
+	// bucket+resNano-1) <= endNano, i.e. bucket in [startNano, endNano-resNano+1].
+	// Push that exact closed range into SQL; no post-filter is then required.
+	lower := startNano
+	upper := endNano - resNano + 1 // inclusive upper bound for bucket_nano
+	if upper < lower {
+		return nil, nil // window narrower than one bucket: nothing is contained
+	}
+	rows, err := s.scanRollupRowsBetween(ctx, metricName, entityID, tags, resNano, lower, upper)
+	if err != nil {
+		return nil, err
+	}
+	return rows, nil
+}
+
+// scanRollupRowsBetween loads rollup rows for one resolution whose bucket start
+// falls in the inclusive nano range [lowerBucket, upperBucket], with optional
+// entity and tag filters pushed into SQL. It is the shared SQL primitive behind
+// the containment and hybrid scans; the bucket-window semantics are imposed by
+// the caller through the bounds it passes.
+//
+// scanRollupRowsBetween 读取某分辨率下桶起点落在闭区间
+// [lowerBucket, upperBucket] 内的 rollup 行，并可把实体和标签过滤下推到 SQL。
+// 它是包含扫描和混合扫描共用的 SQL 原语；桶窗口语义由调用方通过传入的边界决定。
+func (s *Store) scanRollupRowsBetween(ctx context.Context, metricName, entityID string, tags map[string]string, resNano, lowerBucket, upperBucket int64) ([]storedRollup, error) {
+	args := []any{metricName, resNano, lowerBucket, upperBucket}
+	parts := []string{
+		"metric_name = " + s.dialect.placeholder(1),
+		"resolution_nano = " + s.dialect.placeholder(2),
+		"bucket_nano >= " + s.dialect.placeholder(3),
+		"bucket_nano <= " + s.dialect.placeholder(4),
+	}
+	if strings.TrimSpace(entityID) != "" {
+		args = append(args, entityID)
+		parts = append(parts, "entity_id = "+s.dialect.placeholder(len(args)))
+	}
+	for _, k := range sortedKeys(tags) {
+		args = append(args, tags[k])
+		parts = append(parts, s.dialect.jsonExtractEquals("tags", k, s.dialect.placeholder(len(args))))
+	}
+	sqlText := fmt.Sprintf(
+		`SELECT entity_id, tags_hash, tags, bucket_nano, count, sum, sum_sq, min_val, max_val, first_val, first_ts, last_val, last_ts, digest
+		 FROM %s WHERE %s ORDER BY bucket_nano ASC`,
+		s.tables.rollups, strings.Join(parts, " AND "),
+	)
+	rows, err := s.reader().QueryContext(ctx, sqlText, args...)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+	return scanStoredRollups(rows)
+}
+
+// foldRollupRows folds stored rollup rows into interval-wide output buckets,
+// merging summaries (including the t-digest) per bucket. groups may be nil, in
+// which case a fresh map is allocated; passing an existing map lets a caller
+// accumulate rollup and raw contributions into the same output buckets.
+//
+// foldRollupRows 将存储的 rollup 行折叠进 interval 宽的输出桶，按桶合并摘要
+// （包括 t-digest）。groups 可为 nil，此时会分配新 map；传入已有 map 可让调用方
+// 把 rollup 和 raw 的贡献累加到同一批输出桶中。
+func foldRollupRows(groups map[int64]*rollupBucket, rows []storedRollup, interval time.Duration, comp float64) map[int64]*rollupBucket {
+	if groups == nil {
+		groups = make(map[int64]*rollupBucket)
+	}
+	size := interval.Nanoseconds()
+	for _, r := range rows {
+		bkt := floorDivNano(r.bucket, size)
+		b := groups[bkt]
+		if b == nil {
+			b = newRollupBucket(comp)
+			groups[bkt] = b
+		}
+		b.mergeStored(r.bucketData)
+	}
+	return groups
+}
+
+// foldRawPoints folds raw points into interval-wide output buckets, adding each
+// observation into the matching bucket's accumulator. It shares the bucket map
+// with foldRollupRows so a hybrid query can combine an old rollup half and a
+// recent raw half into the same output buckets and aggregate them together
+// (correct count/avg/percentile across the boundary).
+//
+// foldRawPoints 将原始点折叠进 interval 宽的输出桶，把每个观测值加入对应桶的
+// 累加器。它与 foldRollupRows 共用桶 map，因此混合查询可以把旧 rollup 半边和
+// 近期 raw 半边合并到同一批输出桶中并一起聚合（跨边界的 count/avg/百分位正确）。
+func foldRawPoints(groups map[int64]*rollupBucket, points []Point, interval time.Duration, comp float64) map[int64]*rollupBucket {
+	if groups == nil {
+		groups = make(map[int64]*rollupBucket)
+	}
+	size := interval.Nanoseconds()
+	for _, p := range points {
+		ts := p.Timestamp.UTC().UnixNano()
+		bkt := floorDivNano(ts, size)
+		b := groups[bkt]
+		if b == nil {
+			b = newRollupBucket(comp)
+			groups[bkt] = b
+		}
+		b.addPoint(p.Value, ts)
+	}
+	return groups
+}
+
+// scanStoredRollups reconstructs storedRollup rows (including tag identity and
+// the decoded t-digest) from a result set whose columns are, in order:
+// entity_id, tags_hash, tags, bucket_nano, count, sum, sum_sq, min_val, max_val,
+// first_val, first_ts, last_val, last_ts, digest.
+//
+// scanStoredRollups 从结果集中还原 storedRollup 行（包括标签身份和解码后的
+// t-digest）。结果集的列顺序必须是：entity_id、tags_hash、tags、bucket_nano、
+// count、sum、sum_sq、min_val、max_val、first_val、first_ts、last_val、
+// last_ts、digest。
+func scanStoredRollups(rows *sql.Rows) ([]storedRollup, error) {
+	var out []storedRollup
+	for rows.Next() {
+		var (
+			eid                                   string
+			tagsHash                              string
+			rawTags                               any
+			bucketNano                            int64
+			count                                 int64
+			sum, sumSq, minV, maxV, firstV, lastV float64
+			firstTS, lastTS                       int64
+			digestBlob                            []byte
+		)
+		if err := rows.Scan(&eid, &tagsHash, &rawTags, &bucketNano, &count, &sum, &sumSq, &minV, &maxV, &firstV, &firstTS, &lastV, &lastTS, &digestBlob); err != nil {
+			return nil, err
+		}
+		td, err := DecodeTDigest(digestBlob)
+		if err != nil {
+			return nil, err
+		}
+		tagsJSON, err := rawTagsToJSON(rawTags)
+		if err != nil {
+			return nil, err
+		}
+		out = append(out, storedRollup{
+			entityID: eid,
+			bucket:   bucketNano,
+			bucketData: &rollupBucket{
+				count: count, sum: sum, sumSq: sumSq,
+				min: minV, max: maxV,
+				firstVal: firstV, firstTS: firstTS,
+				lastVal: lastV, lastTS: lastTS,
+				digest:   td,
+				tagsHash: tagsHash,
+				tagsJSON: tagsJSON,
+			},
+		})
+	}
+	return out, rows.Err()
+}
+
+// rawTagsToJSON normalizes a scanned tags column (string or []byte) into the
+// canonical JSON string used when the bucket is re-written by a coarser tier.
+//
+// rawTagsToJSON 将扫描出的 tags 列（string 或 []byte）规范化为 JSON 字符串，
+// 供更粗层级重写桶时复用。
+func rawTagsToJSON(v any) (string, error) {
+	switch x := v.(type) {
+	case nil:
+		return "{}", nil
+	case string:
+		if x == "" {
+			return "{}", nil
+		}
+		return x, nil
+	case []byte:
+		if len(x) == 0 {
+			return "{}", nil
+		}
+		return string(x), nil
+	default:
+		return "", fmt.Errorf("unsupported tags column type %T", v)
+	}
+}
+
+// Series answers an AggregateQuery by transparently choosing the best data
+// source for the requested window, given `now`:
+//
+//   - If rollups are disabled, or the whole window still lies within raw
+//     retention, it reads raw points (Aggregate) for full fidelity.
+//   - Otherwise it picks the FINEST rollup tier that both (a) has an Interval
+//     dividing query.Interval and (b) whose retention reaches back to the start
+//     of the window, and serves the query from that tier.
+//   - If the query spans the raw retention boundary, it uses a hybrid approach:
+//     read rollups for the old part and raw for the recent part, then merge
+//     the results to avoid losing uncompacted recent data.
+//   - If no tier qualifies, it falls back to raw (which may be incomplete for
+//     data already aged out) so the call still returns its best answer.
+//
+// query.Tags is honored on both branches: the raw path already filters by tag,
+// and the rollup path filters by the stored tag set, so a tag filter selects the
+// same series regardless of which source answers. This is the "downsampling
+// TSDB" read path: recent ranges answer from raw at full resolution, older
+// ranges answer from progressively coarser rollups.
+//
+// Series 会在给定 `now` 的情况下，通过透明选择最佳数据源来回答 AggregateQuery：
+//
+//   - 如果 rollup 已禁用，或整个窗口仍在原始保留期内，它会读取原始点
+//     （Aggregate）以获得完整保真度。
+//   - 否则它会选择最细的 rollup 层，该层必须同时满足 (a) Interval 能整除
+//     query.Interval，且 (b) 保留时间能覆盖到窗口起点，然后从该层服务查询。
+//   - 如果查询跨越原始数据保留期边界，它使用混合方式：读取旧部分的 rollup
+//     和最近部分的原始点，然后合并结果，避免丢掉未 compact 的最近数据。
+//   - 如果没有层级符合条件，它会回退到原始点（对于已经过期的数据可能不完整），
+//     让调用仍返回当前能给出的最佳答案。
+//
+// query.Tags 在两条分支上都会被遵守：原始路径已按标签过滤，rollup 路径会按
+// 存储的标签集合过滤，因此无论由哪个数据源回答，标签过滤都会选择相同序列。
+// 这是“降采样 TSDB”的读取路径：近期范围以完整分辨率从原始点回答，旧范围从
+// 逐级更粗的 rollup 回答。
+func (s *Store) Series(ctx context.Context, query AggregateQuery, now time.Time) ([]AggregatePoint, error) {
+	if err := s.ensureOpen(); err != nil {
+		return nil, err
+	}
+	if err := query.Validate(); err != nil {
+		return nil, err
+	}
+	policy := s.cfg.RollupPolicy
+	if !policy.Enabled() {
+		return s.Aggregate(ctx, query)
+	}
+	q := query.Query.normalized()
+	now = now.UTC()
+
+	rawCutoff := policy.rawCutoff(now)
+
+	// Whole window inside raw retention (or raw kept forever) -> raw.
+	if rawCutoff.IsZero() || !q.Start.Before(rawCutoff) {
+		return s.Aggregate(ctx, query)
+	}
+	// Rate is raw-only; the caller asked for something rollups can't provide, so
+	// answer from raw regardless of age.
+	if query.Aggregation == AggRate {
+		return s.Aggregate(ctx, query)
+	}
+
+	if q.Start.Before(rawCutoff) && !q.End.Before(rawCutoff) {
+		for _, tier := range policy.Tiers { // finest-first
+			if query.Interval < tier.Interval || query.Interval%tier.Interval != 0 {
+				continue
+			}
+			if now.Add(-tier.Retention).After(q.Start) {
+				continue
+			}
+			return s.seriesHybrid(ctx, query, rawCutoff, &tier)
+		}
+	}
+
+	// Find the finest tier that can serve the entire window from the start.
+	var servicingTier *RollupTier
+	for _, tier := range policy.Tiers { // finest-first
+		if query.Interval < tier.Interval || query.Interval%tier.Interval != 0 {
+			continue
+		}
+		if now.Add(-tier.Retention).After(q.Start) {
+			continue // tier doesn't reach back to the window start
+		}
+		servicingTier = &tier
+		break
+	}
+
+	if servicingTier == nil {
+		return s.Aggregate(ctx, query)
+	}
+
+	return s.AggregateRollup(ctx, query, servicingTier.Interval)
+}
+
+// seriesHybrid answers a query that spans the raw-retention boundary by reading
+// rollups for the old part and raw for the recent part and folding BOTH into the
+// same query.Interval-wide output buckets before reducing them.
+//
+// The old lossy approach reduced each half to AggregatePoints independently and
+// then deduplicated by bucket time, letting raw fully replace a rollup point in
+// any output bucket they shared. That dropped the rollup half of a bucket that
+// straddles the boundary: e.g. with a 1h output bucket and the boundary mid-hour,
+// an 11:05 rollup sample and an 11:35 raw sample both belong to the 11:00 bucket,
+// yet the result kept only the raw sample (count/avg wrong). Folding both halves
+// into one rollupBucket per output bucket fixes that: count, sum/avg, min/max and
+// percentiles are computed over the union of the two halves.
+//
+// To avoid double counting or gaps, the caller passes a boundary already aligned
+// to a rollup-resolution bucket. The rollup half supplies whole buckets strictly
+// before the boundary and the raw half supplies points at or after it.
+//
+// seriesHybrid 回答跨越原始保留期边界的查询：读取旧部分的 rollup 和近期部分的
+// 原始点，并在归约前把两者折叠进同一批 query.Interval 宽的输出桶。
+//
+// 旧的有损做法会把两半各自归约成 AggregatePoint 再按桶时间去重，让 raw 在二者
+// 共有的输出桶里完全覆盖 rollup 点，从而丢掉跨边界桶的 rollup 半边（例如 1h 输出
+// 桶且边界落在小时中间时，11:05 的 rollup 样本和 11:35 的 raw 样本都属于 11:00 桶，
+// 结果却只剩 raw 样本，count/avg 错误）。把两半折叠进同一个 rollupBucket 即可修复：
+// count、sum/avg、min/max 和百分位都在两半的并集上计算。
+//
+// 为避免重复计数或留空，调用方传入的边界已经对齐到 rollup 分辨率桶。rollup 半边
+// 提供严格早于该边界的完整桶，raw 半边提供该边界及之后的点。
+func (s *Store) seriesHybrid(ctx context.Context, query AggregateQuery, boundary time.Time, tier *RollupTier) ([]AggregatePoint, error) {
+	q := query.Query.normalized()
+	comp := s.cfg.RollupPolicy.compression()
+	resNano := tier.Interval.Nanoseconds()
+	startNano := q.Start.UnixNano()
+	endNano := q.End.UnixNano()
+	splitAt := boundary.UTC().UnixNano()
+
+	groups := make(map[int64]*rollupBucket)
+
+	// Old half: rollup buckets fully contained in [Start, min(splitAt, End]].
+	// "Fully before splitAt" is bucket <= splitAt-resNano (splitAt is aligned);
+	// also clamp to the query's own inclusive end so a bucket past End is never
+	// pulled in when splitAt overshoots a short window.
+	upperBucket := splitAt - resNano
+	if clamp := endNano - resNano + 1; clamp < upperBucket {
+		upperBucket = clamp
+	}
+	if upperBucket >= startNano {
+		rows, err := s.scanRollupRowsBetween(ctx, q.MetricName, q.EntityID, q.Tags, resNano, startNano, upperBucket)
+		if err != nil {
+			return nil, err
+		}
+		foldRollupRows(groups, rows, query.Interval, comp)
+	}
+
+	// Recent half: raw points in [splitAt, End]. Reuse the raw Query path so the
+	// entity and tag filters match the rollup half exactly, then fold the points
+	// into the SAME output buckets so a bucket straddling splitAt aggregates both
+	// halves. Skip when splitAt is past the window end (no recent raw to add).
+	if splitAt <= endNano {
+		rawQuery := q
+		rawQuery.Start = time.Unix(0, splitAt).UTC()
+		rawQuery.Limit = 0
+		rawQuery.Offset = 0
+		points, err := s.Query(ctx, rawQuery)
+		if err != nil {
+			return nil, err
+		}
+		foldRawPoints(groups, points, query.Interval, comp)
+	}
+
+	out, err := rollupGroupsToPoints(groups, query)
+	if err != nil {
+		return nil, err
+	}
+	return pageBuckets(out, query.BucketLimit, query.BucketOffset), nil
+}

pkg/metric/rollup_store.go

@@ -0,0 +1,740 @@
+package metric
+
+import (
+	"context"
+	"database/sql"
+	"fmt"
+	"sort"
+	"strings"
+	"time"
+)
+
+// Compact runs the store's RollupPolicy over every metric: it rebuilds the
+// rollup tiers up to `now` and enforces every retention window (raw points and
+// each tier). It is the explicit maintenance entry point — call it from a cron
+// job or scheduler. It is safe to call repeatedly and is idempotent for
+// unchanged windows.
+//
+// Returns the number of rollup buckets written across all tiers and metrics.
+//
+// Compact 会对所有指标执行 Store 的 RollupPolicy：它会重建截至 `now` 的
+// rollup 层，并执行每个保留窗口（原始点和各层级）。这是显式维护入口，
+// 应从 cron 任务或调度器调用。它可以重复安全调用，未变化窗口保持幂等。
+//
+// 返回所有指标、所有层级中写入的 rollup 桶数量。
+func (s *Store) Compact(ctx context.Context, now time.Time) (int, error) {
+	if err := s.ensureOpen(); err != nil {
+		return 0, err
+	}
+	if !s.cfg.RollupPolicy.Enabled() {
+		return 0, nil
+	}
+	defs, err := s.ListMetrics(ctx)
+	if err != nil {
+		return 0, err
+	}
+	total := 0
+	for _, def := range defs {
+		n, err := s.CompactMetric(ctx, def.Name, now)
+		if err != nil {
+			return total, fmt.Errorf("compact metric %q: %w", def.Name, err)
+		}
+		total += n
+	}
+	return total, nil
+}
+
+// CompactMetric compacts a single metric. It builds the finest tier from raw
+// points, then each coarser tier from the tier below it, upserts the resulting
+// buckets, and finally deletes data that has aged out of each retention window
+// (raw points first, then each tier).
+//
+// The entire compaction is performed within a SERIALIZABLE transaction to
+// guarantee that the raw scan and the raw deletion observe the same snapshot.
+// Without this, the default isolation level on PostgreSQL/MySQL lets a point
+// inserted between the scan and the delete be deleted by the cutoff DELETE
+// without ever being folded into a rollup: the scan never saw it (it ran on an
+// earlier read), yet the DELETE (a fresh read) does. SERIALIZABLE makes the
+// scan acquire predicate/range protection so a concurrent write that would
+// otherwise slip into that gap either is excluded from the delete or forces a
+// serialization failure that we retry on a fresh snapshot. SQLite serializes
+// writes on a single connection, so its default isolation already provides this
+// guarantee and needs no escalation.
+//
+// Rollups are keyed by tag set as well as entity, so each distinct tag
+// combination (e.g. a GPU device_index) is summarized into its own series and
+// can be queried independently after the raw points are gone.
+//
+// CompactMetric 会压缩单个指标。它先由原始点构建最细层，再由下一层之下的
+// 层级逐层合成更粗层，upsert 生成的桶，最后删除每个保留窗口中过期的数据
+// （先删除原始点，再删除各层级）。
+//
+// 整个 compaction 在一个 SERIALIZABLE 事务内执行，以保证 raw 扫描和 raw 删除
+// 看到同一个快照。否则在 PostgreSQL/MySQL 的默认隔离级别下，扫描之后、删除
+// 之前写入的旧时间点仍可能被 cutoff 删除却没有进入 rollup：扫描（较早的读）
+// 没看到它，而删除（一次新的读）看到了。SERIALIZABLE 让扫描获得谓词/范围
+// 保护，使这种并发写要么被删除排除在外，要么触发可重试的序列化失败，由我们
+// 在新快照上重试。SQLite 在单连接上串行化写入，其默认隔离已提供该保证，
+// 无需提升隔离级别。
+func (s *Store) CompactMetric(ctx context.Context, metricName string, now time.Time) (int, error) {
+	if err := s.ensureOpen(); err != nil {
+		return 0, err
+	}
+	policy := s.cfg.RollupPolicy
+	if !policy.Enabled() {
+		return 0, nil
+	}
+	now = now.UTC()
+
+	// Retry the whole compaction on a transient serialization/deadlock failure.
+	// Under SERIALIZABLE, a concurrent writer touching the same raw range can
+	// abort our transaction; re-running on a fresh snapshot is the correct and
+	// expected recovery, and Compact is idempotent so a retry is safe.
+	const maxAttempts = 5
+	var lastErr error
+	for attempt := 0; attempt < maxAttempts; attempt++ {
+		written, err := s.compactMetricOnce(ctx, metricName, now)
+		if err == nil {
+			return written, nil
+		}
+		if !isRetryableSerializationError(err) {
+			return written, err
+		}
+		lastErr = err
+	}
+	return 0, fmt.Errorf("compact metric %q: exhausted retries after serialization failures: %w", metricName, lastErr)
+}
+
+// compactMetricOnce runs a single compaction attempt inside one transaction.
+//
+// compactMetricOnce 在单个事务内执行一次 compaction 尝试。
+func (s *Store) compactMetricOnce(ctx context.Context, metricName string, now time.Time) (int, error) {
+	// Use a transaction to ensure consistency between raw scan, rollup write, and
+	// raw deletion. The isolation level is backend-specific (SERIALIZABLE on
+	// PostgreSQL/MySQL, default on SQLite) so late-arriving points cannot be
+	// deleted without first being rolled up.
+	tx, err := s.db.BeginTx(ctx, s.dialect.compactTxOptions())
+	if err != nil {
+		return 0, err
+	}
+	defer func() { _ = tx.Rollback() }()
+
+	written, err := s.compactMetricWithinTx(ctx, metricName, now, tx)
+	if err != nil {
+		return written, err
+	}
+
+	if err := tx.Commit(); err != nil {
+		return written, err
+	}
+	return written, nil
+}
+
+// isRetryableSerializationError reports whether err is a transient
+// serialization failure or deadlock that should be retried on a fresh
+// transaction. It matches on portable SQLSTATE codes and driver error text so
+// the package stays free of driver-specific error type imports:
+//
+//   - PostgreSQL: 40001 serialization_failure, 40P01 deadlock_detected.
+//   - MySQL: 1213 deadlock, 1205 lock wait timeout.
+//   - SQLite: SQLITE_BUSY / database is locked (only relevant with >1 conn).
+//
+// isRetryableSerializationError 判断 err 是否为应在新事务上重试的瞬时序列化
+// 失败或死锁。它通过可移植的 SQLSTATE 码和驱动错误文本匹配，使包不需要导入
+// 驱动专用错误类型。
+func isRetryableSerializationError(err error) bool {
+	if err == nil {
+		return false
+	}
+	msg := strings.ToLower(err.Error())
+	switch {
+	case strings.Contains(msg, "40001"): // postgres serialization_failure
+		return true
+	case strings.Contains(msg, "40p01"): // postgres deadlock_detected
+		return true
+	case strings.Contains(msg, "serialization failure"):
+		return true
+	case strings.Contains(msg, "could not serialize"):
+		return true
+	case strings.Contains(msg, "deadlock"): // mysql 1213, postgres text
+		return true
+	case strings.Contains(msg, "lock wait timeout"): // mysql 1205
+		return true
+	case strings.Contains(msg, "database is locked"): // sqlite busy
+		return true
+	case strings.Contains(msg, "database table is locked"): // sqlite busy
+		return true
+	default:
+		return false
+	}
+}
+
+// compactMetricWithinTx is the actual compaction logic, meant to be called
+// within a transaction. The tx parameter is passed explicitly to all operations
+// that need transactional consistency.
+func (s *Store) compactMetricWithinTx(ctx context.Context, metricName string, now time.Time, tx *sql.Tx) (int, error) {
+	policy := s.cfg.RollupPolicy
+	if !policy.Enabled() {
+		return 0, nil
+	}
+	comp := policy.compression()
+
+	written := 0
+	var prevInterval time.Duration
+	var prevDelta map[rollupKey]*rollupBucket
+	rawCutoff := policy.rawCutoff(now)
+
+	for tierIdx, tier := range policy.Tiers {
+		var buckets map[rollupKey]*rollupBucket
+		var currentDelta map[rollupKey]*rollupBucket
+		var err error
+		// Reads must run on the transaction itself, not s.reader(): with a
+		// single-connection pool (SQLite MaxOpenConns=1) the tx already holds
+		// the only connection, so a pool read here would deadlock.
+		if tierIdx == 0 {
+			buckets, err = s.buildFinestTier(ctx, tx, metricName, tier.Interval, comp)
+			if err == nil && !rawCutoff.IsZero() {
+				cutoffNano := rawCutoff.UnixNano()
+				for k := range buckets {
+					if k.bucket >= cutoffNano {
+						delete(buckets, k)
+					}
+				}
+			}
+			if !rawCutoff.IsZero() {
+				currentDelta = buckets
+			}
+		} else {
+			buckets, err = s.buildCoarserTier(ctx, tx, metricName, prevInterval, tier.Interval, comp)
+			if prevDelta != nil {
+				currentDelta = buildCoarserBucketsFromDelta(prevDelta, tier.Interval, comp)
+			}
+		}
+
+		if err != nil {
+			return written, err
+		}
+		// For the finest tier only, merge buckets older than the raw cutoff:
+		// those buckets may already have a retained rollup while only late raw
+		// samples remain in the raw table.
+		var n int
+		if tierIdx == 0 && !rawCutoff.IsZero() {
+			n, err = s.writeRollupBucketsWithMergePointTx(ctx, metricName, tier.Interval, buckets, rawCutoff, time.Time{}, nil, tx)
+		} else if tierIdx > 0 {
+			deltaCutoff := alignRollupRetentionCutoff(now.Add(-policy.Tiers[tierIdx-1].Retention), tier.Interval)
+			n, err = s.writeRollupBucketsWithMergePointTx(ctx, metricName, tier.Interval, buckets, time.Time{}, deltaCutoff, currentDelta, tx)
+		} else {
+			n, err = s.writeRollupBucketsTx(ctx, metricName, tier.Interval, buckets, tx)
+		}
+		if err != nil {
+			return written, err
+		}
+		written += n
+		prevInterval = tier.Interval
+		prevDelta = currentDelta
+	}
+
+	// Enforce retention windows after all tiers are materialized, so a coarser
+	// tier is always built before the finer source it depends on is trimmed.
+	if policy.RawRetention > 0 {
+		cutoff := policy.rawCutoff(now)
+		if _, err := s.DeleteBeforeTx(ctx, metricName, cutoff, tx); err != nil {
+			return written, err
+		}
+	}
+	for i, tier := range policy.Tiers {
+		cutoff := now.Add(-tier.Retention)
+		if i+1 < len(policy.Tiers) {
+			cutoff = alignRollupRetentionCutoff(cutoff, policy.Tiers[i+1].Interval)
+		}
+		if err := s.deleteRollupsBeforeTx(ctx, metricName, tier.Interval, cutoff, tx); err != nil {
+			return written, err
+		}
+	}
+	return written, nil
+}
+
+func alignRollupRetentionCutoff(cutoff time.Time, nextInterval time.Duration) time.Time {
+	if nextInterval <= 0 {
+		return cutoff.UTC()
+	}
+	return time.Unix(0, floorDivNano(cutoff.UTC().UnixNano(), nextInterval.Nanoseconds())).UTC()
+}
+
+// rollupKey identifies one rollup cell. The tag dimension (tagsHash) is part of
+// the key so points carrying different tags never collapse into the same bucket.
+//
+// rollupKey 标识一个 rollup 单元；tagsHash 是 key 的一部分，确保不同标签
+// 的点不会落入同一个桶。
+type rollupKey struct {
+	// entityID is the entity dimension of the rollup cell.
+	//
+	// entityID 是 rollup 单元的实体维度。
+	entityID string
+	// tagsHash is the stable fingerprint of the tag set.
+	//
+	// tagsHash 是标签集合的稳定指纹。
+	tagsHash string
+	// bucket is the bucket start timestamp in nanoseconds.
+	//
+	// bucket 是桶起始时间的纳秒时间戳。
+	bucket int64
+}
+
+// buildFinestTier scans raw points for the metric and groups them into buckets
+// of the given interval, keyed by (entity, tag set, bucket-start). Each point's
+// tag map determines which series it belongs to.
+//
+// buildFinestTier 扫描某指标的原始点，并按给定 interval 分桶，key 为
+// （实体、标签集合、桶起点）。每个点的标签 map 决定它属于哪条序列。
+func (s *Store) buildFinestTier(ctx context.Context, q querier, metricName string, interval time.Duration, comp float64) (map[rollupKey]*rollupBucket, error) {
+	size := interval.Nanoseconds()
+	out := make(map[rollupKey]*rollupBucket)
+	sqlText := fmt.Sprintf(
+		`SELECT entity_id, ts_nano, value, tags FROM %s WHERE metric_name = %s ORDER BY ts_nano ASC`,
+		s.tables.points, s.dialect.placeholder(1),
+	)
+	rows, err := q.QueryContext(ctx, sqlText, metricName)
+
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+	for rows.Next() {
+		var entityID string
+		var ts int64
+		var value float64
+		var rawTags any
+		if err := rows.Scan(&entityID, &ts, &value, &rawTags); err != nil {
+			return nil, err
+		}
+		tags, err := decodeMap(rawTags)
+		if err != nil {
+			return nil, err
+		}
+		hash, canonical, err := tagsFingerprint(tags)
+		if err != nil {
+			return nil, err
+		}
+		bucket := floorDivNano(ts, size)
+		k := rollupKey{entityID: entityID, tagsHash: hash, bucket: bucket}
+		b := out[k]
+		if b == nil {
+			b = newRollupBucket(comp)
+			b.tagsHash = hash
+			b.tagsJSON = canonical
+			out[k] = b
+		}
+		b.addPoint(value, ts)
+	}
+	return out, rows.Err()
+}
+
+// buildCoarserTier composes a coarser tier from the already-stored finer tier:
+// it reads the finer rollup rows and merges every finer bucket into the coarse
+// bucket that shares its (entity, tag set). Tag identity is preserved end to
+// end, so a coarse series only ever merges finer buckets of the same tag set.
+//
+// buildCoarserTier 基于已存储的细层 rollup 合成更粗层：它读取细层 rollup 行，
+// 并把每个细桶合并进共享相同（实体、标签集合）的粗桶。标签身份会端到端保留，
+// 因此一条粗粒度序列只会合并相同标签集合的细桶。
+func (s *Store) buildCoarserTier(ctx context.Context, q querier, metricName string, fineInterval, coarseInterval time.Duration, comp float64) (map[rollupKey]*rollupBucket, error) {
+	coarseSize := coarseInterval.Nanoseconds()
+	out := make(map[rollupKey]*rollupBucket)
+	fineRows, err := s.scanRollupRows(ctx, q, metricName, fineInterval)
+
+	if err != nil {
+		return nil, err
+	}
+	for _, fr := range fineRows {
+		bucket := floorDivNano(fr.bucket, coarseSize)
+		k := rollupKey{entityID: fr.entityID, tagsHash: fr.bucketData.tagsHash, bucket: bucket}
+		b := out[k]
+		if b == nil {
+			b = newRollupBucket(comp)
+			b.tagsHash = fr.bucketData.tagsHash
+			b.tagsJSON = fr.bucketData.tagsJSON
+			out[k] = b
+		}
+		b.mergeStored(fr.bucketData)
+	}
+	return out, nil
+}
+
+func buildCoarserBucketsFromDelta(delta map[rollupKey]*rollupBucket, coarseInterval time.Duration, comp float64) map[rollupKey]*rollupBucket {
+	out := make(map[rollupKey]*rollupBucket)
+	if len(delta) == 0 {
+		return out
+	}
+	coarseSize := coarseInterval.Nanoseconds()
+	for k, src := range delta {
+		bucket := floorDivNano(k.bucket, coarseSize)
+		ck := rollupKey{entityID: k.entityID, tagsHash: k.tagsHash, bucket: bucket}
+		b := out[ck]
+		if b == nil {
+			b = newRollupBucket(comp)
+			b.tagsHash = src.tagsHash
+			b.tagsJSON = src.tagsJSON
+			out[ck] = b
+		}
+		b.mergeStored(src)
+	}
+	return out
+}
+
+// storedRollup represents a rollup row reconstructed from storage.
+//
+// storedRollup 表示从存储中还原的一行 rollup 数据。
+type storedRollup struct {
+	// entityID is the entity stored on the rollup row.
+	//
+	// entityID 是 rollup 行中保存的实体。
+	entityID string
+	// bucket is the stored bucket start timestamp in nanoseconds.
+	//
+	// bucket 是存储的桶起始纳秒时间戳。
+	bucket int64
+	// bucketData is the reconstructed in-memory accumulator for the row.
+	//
+	// bucketData 是该行还原出的内存累加器。
+	bucketData *rollupBucket
+}
+
+// scanRollupRows loads all rollup rows for a metric at a given resolution and
+// reconstructs their in-memory accumulators (including tag identity and the
+// decoded t-digest).
+//
+// scanRollupRows 读取某指标在给定分辨率下的所有 rollup 行，并还原它们的
+// 内存累加器（包括标签身份和解码后的 t-digest）。
+func (s *Store) scanRollupRows(ctx context.Context, q querier, metricName string, interval time.Duration) ([]storedRollup, error) {
+	sqlText := fmt.Sprintf(
+		`SELECT entity_id, tags_hash, tags, bucket_nano, count, sum, sum_sq, min_val, max_val, first_val, first_ts, last_val, last_ts, digest
+		 FROM %s WHERE metric_name = %s AND resolution_nano = %s ORDER BY bucket_nano ASC`,
+		s.tables.rollups, s.dialect.placeholder(1), s.dialect.placeholder(2),
+	)
+	rows, err := q.QueryContext(ctx, sqlText, metricName, interval.Nanoseconds())
+
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+	return scanStoredRollups(rows)
+}
+
+// writeRollupBuckets upserts a set of computed buckets for one resolution.
+//
+// writeRollupBuckets 将某个分辨率下计算出的 rollup 桶批量 upsert 到数据库。
+func (s *Store) writeRollupBuckets(ctx context.Context, metricName string, interval time.Duration, buckets map[rollupKey]*rollupBucket) (int, error) {
+	return s.writeRollupBucketsWithMergePoint(ctx, metricName, interval, buckets, time.Time{})
+}
+
+// writeRollupBucketsTx upserts buckets within an existing transaction.
+func (s *Store) writeRollupBucketsTx(ctx context.Context, metricName string, interval time.Duration, buckets map[rollupKey]*rollupBucket, tx *sql.Tx) (int, error) {
+	return s.writeRollupBucketsWithMergePointTx(ctx, metricName, interval, buckets, time.Time{}, time.Time{}, nil, tx)
+}
+
+// writeRollupBucketsWithMergePoint is the internal implementation that optionally
+// merges buckets older than a cutoff point instead of replacing them.
+func (s *Store) writeRollupBucketsWithMergePoint(ctx context.Context, metricName string, interval time.Duration, buckets map[rollupKey]*rollupBucket, mergeCutoff time.Time) (int, error) {
+	if len(buckets) == 0 {
+		return 0, nil
+	}
+	keys := make([]rollupKey, 0, len(buckets))
+	for k := range buckets {
+		keys = append(keys, k)
+	}
+	sort.Slice(keys, func(i, j int) bool {
+		if keys[i].entityID != keys[j].entityID {
+			return keys[i].entityID < keys[j].entityID
+		}
+		if keys[i].tagsHash != keys[j].tagsHash {
+			return keys[i].tagsHash < keys[j].tagsHash
+		}
+		return keys[i].bucket < keys[j].bucket
+	})
+
+	stmt := s.dialect.upsertRollupSQL(s.tables)
+	resNano := interval.Nanoseconds()
+	now := time.Now().UTC().UnixNano()
+	mergeCutoffNano := mergeCutoff.UnixNano()
+
+	run := func(ex execer) error {
+		for _, k := range keys {
+			b := buckets[k]
+			tagsJSON := b.tagsJSON
+			if tagsJSON == "" {
+				tagsJSON = "{}"
+			}
+
+			// If mergeCutoff is set and this bucket is older than the cutoff,
+			// merge with existing data instead of replacing it.
+			if !mergeCutoff.IsZero() && k.bucket < mergeCutoffNano {
+				// Read existing row if it exists
+				existing, err := s.readRollupBucket(ctx, metricName, k.entityID, k.tagsHash, interval, k.bucket)
+				if err != nil {
+					return err
+				}
+				if existing != nil {
+					// Merge the new bucket into the existing one
+					existing.mergeStored(b)
+					b = existing
+				}
+			}
+
+			// Column order must match rollupColumns in dialect_rollup.go:
+			// metric_name, entity_id, tags_hash, tags, resolution_nano, bucket_nano,
+			// count, sum, sum_sq, min_val, max_val, first_val, first_ts, last_val,
+			// last_ts, digest, created_at.
+			_, err := ex.ExecContext(ctx, stmt,
+				metricName, k.entityID, k.tagsHash, tagsJSON, resNano, k.bucket,
+				b.count, b.sum, b.sumSq, b.min, b.max,
+				b.firstVal, b.firstTS, b.lastVal, b.lastTS,
+				b.digest.Encode(), now,
+			)
+			if err != nil {
+				return err
+			}
+		}
+		return nil
+	}
+
+	if len(keys) == 1 {
+		if err := run(s.db); err != nil {
+			return 0, err
+		}
+		return len(keys), nil
+	}
+	tx, err := s.db.BeginTx(ctx, nil)
+	if err != nil {
+		return 0, err
+	}
+	defer func() { _ = tx.Rollback() }()
+	if err := run(tx); err != nil {
+		return 0, err
+	}
+	if err := tx.Commit(); err != nil {
+		return 0, err
+	}
+	return len(keys), nil
+}
+
+// writeRollupBucketsWithMergePointTx is the transactional version that executes
+// all operations within an existing transaction. deltaCutoff marks coarser
+// buckets whose source tier may already have aged out; a smaller recomputed
+// bucket in that region is treated as late-arriving delta data and merged into
+// the retained coarse bucket instead of replacing it.
+func (s *Store) writeRollupBucketsWithMergePointTx(ctx context.Context, metricName string, interval time.Duration, buckets map[rollupKey]*rollupBucket, mergeCutoff, protectCutoff time.Time, deltaBuckets map[rollupKey]*rollupBucket, tx *sql.Tx) (int, error) {
+	if len(buckets) == 0 {
+		return 0, nil
+	}
+	keys := make([]rollupKey, 0, len(buckets))
+	for k := range buckets {
+		keys = append(keys, k)
+	}
+	sort.Slice(keys, func(i, j int) bool {
+		if keys[i].entityID != keys[j].entityID {
+			return keys[i].entityID < keys[j].entityID
+		}
+		if keys[i].tagsHash != keys[j].tagsHash {
+			return keys[i].tagsHash < keys[j].tagsHash
+		}
+		return keys[i].bucket < keys[j].bucket
+	})
+
+	stmt := s.dialect.upsertRollupSQL(s.tables)
+	resNano := interval.Nanoseconds()
+	now := time.Now().UTC().UnixNano()
+	mergeCutoffNano := mergeCutoff.UnixNano()
+	deltaCutoffNano := protectCutoff.UnixNano()
+	written := 0
+
+	for _, k := range keys {
+		b := buckets[k]
+		tagsJSON := b.tagsJSON
+		if tagsJSON == "" {
+			tagsJSON = "{}"
+		}
+
+		// If mergeCutoff is set and this bucket is older than the cutoff,
+		// merge with existing data instead of replacing it.
+		if !mergeCutoff.IsZero() && k.bucket < mergeCutoffNano {
+			// Read existing row if it exists
+			existing, err := s.readRollupBucketTx(ctx, metricName, k.entityID, k.tagsHash, interval, k.bucket, tx)
+			if err != nil {
+				return 0, err
+			}
+			if existing != nil {
+				// Merge the new bucket into the existing one
+				existing.mergeStored(b)
+				b = existing
+			}
+		}
+		if !protectCutoff.IsZero() && k.bucket < deltaCutoffNano {
+			existing, err := s.readRollupBucketTx(ctx, metricName, k.entityID, k.tagsHash, interval, k.bucket, tx)
+			if err != nil {
+				return 0, err
+			}
+			if existing != nil {
+				if deltaBuckets == nil {
+					// Without delta tracking (raw is retained forever), the
+					// source tier is rebuilt from full raw data and can replace.
+				} else if delta := deltaBuckets[k]; delta != nil && delta.count > 0 {
+					existing.mergeStored(delta)
+					b = existing
+				} else {
+					continue
+				}
+			}
+		}
+
+		// Column order must match rollupColumns in dialect_rollup.go
+		_, err := tx.ExecContext(ctx, stmt,
+			metricName, k.entityID, k.tagsHash, tagsJSON, resNano, k.bucket,
+			b.count, b.sum, b.sumSq, b.min, b.max,
+			b.firstVal, b.firstTS, b.lastVal, b.lastTS,
+			b.digest.Encode(), now,
+		)
+		if err != nil {
+			return 0, err
+		}
+		written++
+	}
+	return written, nil
+}
+
+// readRollupBucket reads a single rollup bucket from storage, returning nil if
+// the bucket doesn't exist. Used to merge late-arriving data with existing rollups.
+func (s *Store) readRollupBucket(ctx context.Context, metricName, entityID, tagsHash string, interval time.Duration, bucketNano int64) (*rollupBucket, error) {
+	sqlText := fmt.Sprintf(
+		`SELECT count, sum, sum_sq, min_val, max_val, first_val, first_ts, last_val, last_ts, digest, tags
+		 FROM %s WHERE metric_name = %s AND resolution_nano = %s AND entity_id = %s
+		 AND tags_hash = %s AND bucket_nano = %s`,
+		s.tables.rollups,
+		s.dialect.placeholder(1), s.dialect.placeholder(2), s.dialect.placeholder(3),
+		s.dialect.placeholder(4), s.dialect.placeholder(5),
+	)
+	row := s.reader().QueryRowContext(ctx, sqlText, metricName, interval.Nanoseconds(), entityID, tagsHash, bucketNano)
+
+	var count int64
+	var sum, sumSq, minV, maxV, firstV, lastV float64
+	var firstTS, lastTS int64
+	var digestBlob []byte
+	var rawTags any
+
+	err := row.Scan(&count, &sum, &sumSq, &minV, &maxV, &firstV, &firstTS, &lastV, &lastTS, &digestBlob, &rawTags)
+	if err == sql.ErrNoRows {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, err
+	}
+
+	td, err := DecodeTDigest(digestBlob)
+	if err != nil {
+		return nil, err
+	}
+	tagsJSON, err := rawTagsToJSON(rawTags)
+	if err != nil {
+		return nil, err
+	}
+
+	return &rollupBucket{
+		count: count, sum: sum, sumSq: sumSq,
+		min: minV, max: maxV,
+		firstVal: firstV, firstTS: firstTS,
+		lastVal: lastV, lastTS: lastTS,
+		digest:   td,
+		tagsHash: tagsHash,
+		tagsJSON: tagsJSON,
+	}, nil
+}
+
+// deleteRollupsBefore deletes stored rollup rows older than a cutoff.
+func (s *Store) deleteRollupsBefore(ctx context.Context, metricName string, interval time.Duration, before time.Time) error {
+	sqlText := fmt.Sprintf(
+		`DELETE FROM %s WHERE metric_name = %s AND resolution_nano = %s AND bucket_nano < %s`,
+		s.tables.rollups, s.dialect.placeholder(1), s.dialect.placeholder(2), s.dialect.placeholder(3),
+	)
+	_, err := s.db.ExecContext(ctx, sqlText, metricName, interval.Nanoseconds(), before.UTC().UnixNano())
+	return err
+}
+
+// deleteRollupsBeforeTx deletes stored rollup rows within a transaction.
+func (s *Store) deleteRollupsBeforeTx(ctx context.Context, metricName string, interval time.Duration, before time.Time, tx *sql.Tx) error {
+	sqlText := fmt.Sprintf(
+		`DELETE FROM %s WHERE metric_name = %s AND resolution_nano = %s AND bucket_nano < %s`,
+		s.tables.rollups, s.dialect.placeholder(1), s.dialect.placeholder(2), s.dialect.placeholder(3),
+	)
+	_, err := tx.ExecContext(ctx, sqlText, metricName, interval.Nanoseconds(), before.UTC().UnixNano())
+	return err
+}
+
+// DeleteBeforeTx deletes raw points before a cutoff within a transaction.
+func (s *Store) DeleteBeforeTx(ctx context.Context, metricName string, before time.Time, tx *sql.Tx) (int64, error) {
+	sqlText := fmt.Sprintf(
+		`DELETE FROM %s WHERE metric_name = %s AND ts_nano < %s`,
+		s.tables.points, s.dialect.placeholder(1), s.dialect.placeholder(2),
+	)
+	result, err := tx.ExecContext(ctx, sqlText, metricName, before.UTC().UnixNano())
+	if err != nil {
+		return 0, err
+	}
+	return result.RowsAffected()
+}
+
+// floorDivNano floors ts to the start of its size-wide bucket, handling
+// negative timestamps (pre-epoch) toward negative infinity so buckets align
+// consistently. Mirrors alignTime but operates on raw nanos.
+//
+// floorDivNano 将 ts 向下对齐到 size 宽桶的起点，并把负时间戳（Unix epoch 前）
+// 朝负无穷取整，让桶保持一致对齐。它与 alignTime 逻辑一致，但直接操作纳秒值。
+func floorDivNano(ts, size int64) int64 {
+	rem := ((ts % size) + size) % size
+	return ts - rem
+}
+
+// readRollupBucketTx reads a single rollup bucket from a transaction.
+func (s *Store) readRollupBucketTx(ctx context.Context, metricName, entityID, tagsHash string, interval time.Duration, bucketNano int64, tx *sql.Tx) (*rollupBucket, error) {
+	sqlText := fmt.Sprintf(
+		`SELECT count, sum, sum_sq, min_val, max_val, first_val, first_ts, last_val, last_ts, digest, tags
+		 FROM %s WHERE metric_name = %s AND resolution_nano = %s AND entity_id = %s
+		 AND tags_hash = %s AND bucket_nano = %s`,
+		s.tables.rollups,
+		s.dialect.placeholder(1), s.dialect.placeholder(2), s.dialect.placeholder(3),
+		s.dialect.placeholder(4), s.dialect.placeholder(5),
+	)
+	row := tx.QueryRowContext(ctx, sqlText, metricName, interval.Nanoseconds(), entityID, tagsHash, bucketNano)
+
+	var count int64
+	var sum, sumSq, minV, maxV, firstV, lastV float64
+	var firstTS, lastTS int64
+	var digestBlob []byte
+	var rawTags any
+
+	err := row.Scan(&count, &sum, &sumSq, &minV, &maxV, &firstV, &firstTS, &lastV, &lastTS, &digestBlob, &rawTags)
+	if err == sql.ErrNoRows {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, err
+	}
+
+	td, err := DecodeTDigest(digestBlob)
+	if err != nil {
+		return nil, err
+	}
+	tagsJSON, err := rawTagsToJSON(rawTags)
+	if err != nil {
+		return nil, err
+	}
+
+	return &rollupBucket{
+		count: count, sum: sum, sumSq: sumSq,
+		min: minV, max: maxV,
+		firstVal: firstV, firstTS: firstTS,
+		lastVal: lastV, lastTS: lastTS,
+		digest:   td,
+		tagsHash: tagsHash,
+		tagsJSON: tagsJSON,
+	}, nil
+}

pkg/metric/rollup_tags_test.go

@@ -0,0 +1,349 @@
+package metric
+
+import (
+	"context"
+	"math"
+	"testing"
+	"time"
+)
+
+// TestRollupKeepsTagSeriesSeparate is the regression test for the bug where the
+// rollup dropped the tag dimension: two points in the SAME metric/entity/bucket
+// but with different tags must become two independent rollup series, and a tag
+// filter must return that tag's own value rather than a cross-tag merge.
+//
+// TestRollupKeepsTagSeriesSeparate 是针对 rollup 曾丢失标签维度这一问题的
+// 回归测试：同一个指标、实体和桶内带有不同标签的两个点，必须成为两条独立的
+// rollup 序列；标签过滤必须返回该标签自己的值，而不是跨标签合并结果。
+func TestRollupKeepsTagSeriesSeparate(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{Tiers: []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}}}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "util", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	// Same entity, same 1-minute bucket; two GPUs distinguished only by tag.
+	// device 0 carries values 10..19, device 1 carries 80..89.
+	var batch []Point
+	for i := 0; i < 10; i++ {
+		batch = append(batch,
+			Point{MetricName: "util", EntityID: "host-1", Timestamp: base.Add(time.Duration(i) * time.Second), Value: float64(10 + i), Tags: map[string]string{"device_index": "0"}},
+			Point{MetricName: "util", EntityID: "host-1", Timestamp: base.Add(time.Duration(i) * time.Second), Value: float64(80 + i), Tags: map[string]string{"device_index": "1"}},
+		)
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	written, err := s.Compact(ctx, base.Add(10*time.Minute))
+	if err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+	// Two distinct tag series in the one bucket => two rollup rows.
+	if written != 2 {
+		t.Fatalf("expected 2 rollup series (one per device_index), got %d", written)
+	}
+
+	q := func(dev string) AggregateQuery {
+		return AggregateQuery{
+			Query: Query{
+				MetricName: "util", EntityID: "host-1",
+				Start: base, End: base.Add(10 * time.Minute),
+				Tags: map[string]string{"device_index": dev},
+			},
+			Aggregation: AggAvg,
+			Interval:    time.Minute,
+		}
+	}
+	// device 0: avg of 10..19 = 14.5
+	res0, err := s.AggregateRollup(ctx, q("0"), time.Minute)
+	if err != nil {
+		t.Fatalf("rollup dev0: %v", err)
+	}
+	if len(res0) != 1 || res0[0].Count != 10 || math.Abs(res0[0].Value-14.5) > 1e-9 {
+		t.Fatalf("device 0 should be its own series avg=14.5 count=10, got %#v", res0)
+	}
+	// device 1: avg of 80..89 = 84.5
+	res1, err := s.AggregateRollup(ctx, q("1"), time.Minute)
+	if err != nil {
+		t.Fatalf("rollup dev1: %v", err)
+	}
+	if len(res1) != 1 || res1[0].Count != 10 || math.Abs(res1[0].Value-84.5) > 1e-9 {
+		t.Fatalf("device 1 should be its own series avg=84.5 count=10, got %#v", res1)
+	}
+
+	// No tag filter => both series merge (count 20, avg of all 20 values).
+	all, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query:       Query{MetricName: "util", EntityID: "host-1", Start: base, End: base.Add(10 * time.Minute)},
+		Aggregation: AggAvg,
+		Interval:    time.Minute,
+	}, time.Minute)
+	if err != nil {
+		t.Fatalf("rollup all: %v", err)
+	}
+	wantAvg := (145.0 + 845.0) / 20.0 // sum(10..19)=145, sum(80..89)=845
+	if len(all) != 1 || all[0].Count != 20 || math.Abs(all[0].Value-wantAvg) > 1e-9 {
+		t.Fatalf("no-filter should merge both tags count=20 avg=%.2f, got %#v", wantAvg, all)
+	}
+}
+
+// TestRollupTagPercentileSurvivesRetention proves the headline TSDB property
+// still holds per tag: after raw is deleted, a tag-filtered percentile is still
+// answerable from that tag's surviving rollup series (and not contaminated by
+// the other tag).
+//
+// TestRollupTagPercentileSurvivesRetention 验证核心 TSDB 特性在每个标签上
+// 仍然成立：原始点删除后，按标签过滤的百分位仍能从该标签保留下来的 rollup 序列
+// 回答，并且不会被另一个标签污染。
+func TestRollupTagPercentileSurvivesRetention(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 2 * time.Minute,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 365 * 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "lat", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	// One 1m bucket. region=ap: values 0..99; region=eu: values 1000..1099.
+	var batch []Point
+	for i := 0; i < 100; i++ {
+		ts := base.Add(time.Duration(i) * time.Millisecond * 500)
+		batch = append(batch,
+			Point{MetricName: "lat", EntityID: "n1", Timestamp: ts, Value: float64(i), Tags: map[string]string{"region": "ap"}},
+			Point{MetricName: "lat", EntityID: "n1", Timestamp: ts, Value: float64(1000 + i), Tags: map[string]string{"region": "eu"}},
+		)
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(48*time.Hour)); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+	// Raw gone.
+	raw, err := s.Query(ctx, Query{MetricName: "lat", EntityID: "n1", Start: base, End: base.Add(time.Hour)})
+	if err != nil {
+		t.Fatalf("query raw: %v", err)
+	}
+	if len(raw) != 0 {
+		t.Fatalf("expected raw deleted, got %d", len(raw))
+	}
+	// p90 of region=ap must be near the ap-only exact (≈89), NOT pulled up toward
+	// the eu values (which would happen if tags were merged).
+	res, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query: Query{
+			MetricName: "lat", EntityID: "n1",
+			Start: base, End: base.Add(time.Hour),
+			Tags: map[string]string{"region": "ap"},
+		},
+		Aggregation: Pxx(90),
+		Interval:    time.Minute,
+	}, time.Minute)
+	if err != nil {
+		t.Fatalf("rollup ap p90: %v", err)
+	}
+	if len(res) != 1 {
+		t.Fatalf("expected 1 ap bucket, got %d", len(res))
+	}
+	exactAP := percentileSortedRange(0, 99, 0.90)
+	if math.Abs(res[0].Value-exactAP)/exactAP > 0.05 {
+		t.Fatalf("ap p90 %v contaminated or wrong (exact ap ≈ %v)", res[0].Value, exactAP)
+	}
+	if res[0].Value > 500 {
+		t.Fatalf("ap p90 %v is contaminated by eu series (>500)", res[0].Value)
+	}
+}
+
+// TestSeriesTagFilterRoutesPerTag checks the auto-routing read path honors tags
+// on both the raw branch (recent) and the rollup branch (old).
+//
+// TestSeriesTagFilterRoutesPerTag 检查自动路由读取路径在原始分支（近期数据）和
+// rollup 分支（旧数据）上都遵守标签。
+func TestSeriesTagFilterRoutesPerTag(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: time.Hour,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 30 * 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "m", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	now := time.Date(2026, 6, 18, 12, 0, 0, 0, time.UTC)
+	old := now.Add(-48 * time.Hour)
+	// Old bucket: env=prod values 0..9, env=stage values 500..509.
+	var batch []Point
+	for i := 0; i < 10; i++ {
+		ts := old.Add(time.Duration(i) * time.Second)
+		batch = append(batch,
+			Point{MetricName: "m", EntityID: "n1", Timestamp: ts, Value: float64(i), Tags: map[string]string{"env": "prod"}},
+			Point{MetricName: "m", EntityID: "n1", Timestamp: ts, Value: float64(500 + i), Tags: map[string]string{"env": "stage"}},
+		)
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+	// Old window, raw aged out -> Series must use rollup, filtered to env=prod
+	// (avg of 0..9 = 4.5), not the merged 252.
+	got, err := s.Series(ctx, AggregateQuery{
+		Query: Query{
+			MetricName: "m", EntityID: "n1",
+			Start: old.Add(-time.Minute), End: old.Add(2 * time.Minute),
+			Tags: map[string]string{"env": "prod"},
+		},
+		Aggregation: AggAvg,
+		Interval:    time.Minute,
+	}, now)
+	if err != nil {
+		t.Fatalf("series old prod: %v", err)
+	}
+	var found bool
+	for _, p := range got {
+		if p.Count == 10 && math.Abs(p.Value-4.5) < 1e-9 {
+			found = true
+		}
+		if p.Count > 10 {
+			t.Fatalf("env=prod bucket contaminated by stage: %#v", p)
+		}
+	}
+	if !found {
+		t.Fatalf("env=prod old series not served correctly from rollup: %#v", got)
+	}
+}
+
+// TestDeleteSeriesRemovesTaggedTaskAcrossAgents verifies a task_id tag can be
+// deleted across every agent, including stored rollups.
+//
+// TestDeleteSeriesRemovesTaggedTaskAcrossAgents 验证可以跨所有 agent 删除某个
+// task_id 标签，包括已存储的 rollup。
+func TestDeleteSeriesRemovesTaggedTaskAcrossAgents(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 2 * time.Minute,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "ping.latency_ms", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var batch []Point
+	for _, agent := range []string{"agent-001", "agent-002"} {
+		for i := 0; i < 60; i++ {
+			ts := base.Add(time.Duration(i) * time.Second)
+			batch = append(batch,
+				Point{MetricName: "ping.latency_ms", EntityID: agent, Timestamp: ts, Value: float64(20 + i), Tags: map[string]string{"task_id": "pingtask1"}},
+				Point{MetricName: "ping.latency_ms", EntityID: agent, Timestamp: ts, Value: float64(60 + i), Tags: map[string]string{"task_id": "pingtask2"}},
+			)
+		}
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(time.Hour)); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+	if _, err := s.DeleteSeries(ctx, Query{
+		MetricName: "ping.latency_ms",
+		Tags:       map[string]string{"task_id": "pingtask2"},
+	}); err != nil {
+		t.Fatalf("delete series: %v", err)
+	}
+
+	task2, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query: Query{
+			MetricName: "ping.latency_ms",
+			Start:      base, End: base.Add(time.Minute),
+			Tags: map[string]string{"task_id": "pingtask2"},
+		},
+		Aggregation: AggAvg,
+		Interval:    time.Minute,
+	}, time.Minute)
+	if err != nil {
+		t.Fatalf("rollup task2: %v", err)
+	}
+	if len(task2) != 0 {
+		t.Fatalf("deleted task still has rollup buckets: %#v", task2)
+	}
+	task1, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query: Query{
+			MetricName: "ping.latency_ms",
+			Start:      base, End: base.Add(time.Minute),
+			Tags: map[string]string{"task_id": "pingtask1"},
+		},
+		Aggregation: AggAvg,
+		Interval:    time.Minute,
+	}, time.Minute)
+	if err != nil {
+		t.Fatalf("rollup task1: %v", err)
+	}
+	if len(task1) != 1 || task1[0].Count != 120 {
+		t.Fatalf("remaining task should keep both agents, got %#v", task1)
+	}
+}
+
+// TestDeleteMetricRemovesRollups verifies metric deletion clears rollup rows.
+//
+// TestDeleteMetricRemovesRollups 验证删除指标会清理 rollup 行。
+func TestDeleteMetricRemovesRollups(t *testing.T) {
+	ctx := context.Background()
+	s := newRollupStore(t, RollupPolicy{
+		Tiers: []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	})
+	if err := s.CreateMetric(ctx, Definition{Name: "gone", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	if err := s.Write(ctx, Point{MetricName: "gone", EntityID: "agent-001", Timestamp: base, Value: 1}); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(time.Hour)); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+	if err := s.DeleteMetric(ctx, "gone"); err != nil {
+		t.Fatalf("delete metric: %v", err)
+	}
+	rows, err := s.scanRollupRows(ctx, s.reader(), "gone", time.Minute)
+
+	if err != nil {
+		t.Fatalf("scan rollups: %v", err)
+	}
+	if len(rows) != 0 {
+		t.Fatalf("deleted metric still has rollups: %#v", rows)
+	}
+}
+
+// TestTagsFingerprintStable verifies equal tag maps fingerprint identically
+// regardless of construction order, and different maps differ.
+//
+// TestTagsFingerprintStable 验证相同标签 map 无论构造顺序如何都会生成相同指纹，
+// 而不同 map 会得到不同指纹。
+func TestTagsFingerprintStable(t *testing.T) {
+	a := map[string]string{"region": "ap", "device_index": "3"}
+	b := map[string]string{"device_index": "3", "region": "ap"}
+	ha, _, err := tagsFingerprint(a)
+	if err != nil {
+		t.Fatalf("fp a: %v", err)
+	}
+	hb, _, err := tagsFingerprint(b)
+	if err != nil {
+		t.Fatalf("fp b: %v", err)
+	}
+	if ha != hb {
+		t.Fatalf("equal tag maps must fingerprint identically: %s vs %s", ha, hb)
+	}
+	hc, _, _ := tagsFingerprint(map[string]string{"region": "eu", "device_index": "3"})
+	if ha == hc {
+		t.Fatalf("different tag maps must fingerprint differently")
+	}
+	// nil and empty map agree.
+	hn, _, _ := tagsFingerprint(nil)
+	he, _, _ := tagsFingerprint(map[string]string{})
+	if hn != he {
+		t.Fatalf("nil and empty tag maps must agree: %s vs %s", hn, he)
+	}
+}

pkg/metric/rollup_test.go

@@ -0,0 +1,676 @@
+package metric
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"math"
+	"testing"
+	"time"
+)
+
+// newRollupStore opens an in-memory SQLite store with the given policy. A unique
+// DSN per test keeps the shared-cache in-memory databases isolated.
+//
+// newRollupStore 使用给定策略打开内存 SQLite store；每个测试使用唯一 DSN，
+// 避免 shared-cache 内存数据库相互影响。
+func newRollupStore(t *testing.T, policy RollupPolicy) *Store {
+	t.Helper()
+	dsn := fmt.Sprintf("file:rollup-%d?mode=memory&cache=shared", time.Now().UnixNano())
+	store, err := Open(context.Background(), SQLite(dsn, WithRollupPolicy(policy)))
+	if err != nil {
+		t.Fatalf("open rollup store: %v", err)
+	}
+	t.Cleanup(func() { store.Close() })
+	return store
+}
+
+// TestArbitraryPercentileOverRaw verifies arbitrary percentile (pxx)
+// aggregation end-to-end over raw points.
+//
+// TestArbitraryPercentileOverRaw 验证任意百分位（pxx）聚合在原始点路径上的
+// 端到端行为。
+func TestArbitraryPercentileOverRaw(t *testing.T) {
+	ctx := context.Background()
+	s := newMemStore(t)
+	if err := s.CreateMetric(ctx, Definition{Name: "lat", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var batch []Point
+	for i := 1; i <= 100; i++ { // values 1..100 in one bucket
+		batch = append(batch, Point{MetricName: "lat", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Second), Value: float64(i)})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	for _, tc := range []struct {
+		agg  Aggregation
+		want float64
+	}{
+		{Pxx(50), percentileSortedRange(1, 100, 0.50)},
+		{Pxx(90), percentileSortedRange(1, 100, 0.90)},
+		{Pxx(99), percentileSortedRange(1, 100, 0.99)},
+		{Pxx(99.9), percentileSortedRange(1, 100, 0.999)},
+		{Pxx(75), percentileSortedRange(1, 100, 0.75)},
+	} {
+		res, err := s.Aggregate(ctx, AggregateQuery{
+			Query:       Query{MetricName: "lat", EntityID: "n1", Start: base, End: base.Add(time.Hour)},
+			Aggregation: tc.agg,
+			Interval:    time.Hour,
+		})
+		if err != nil {
+			t.Fatalf("aggregate %s: %v", tc.agg, err)
+		}
+		if len(res) != 1 {
+			t.Fatalf("%s: expected 1 bucket, got %d", tc.agg, len(res))
+		}
+		if math.Abs(res[0].Value-tc.want) > 1e-9 {
+			t.Fatalf("%s: got %v want %v", tc.agg, res[0].Value, tc.want)
+		}
+	}
+}
+
+// percentileSortedRange computes an exact percentile for an integer range.
+//
+// percentileSortedRange 为整数范围计算精确百分位。
+func percentileSortedRange(lo, hi int, q float64) float64 {
+	vals := make([]float64, 0, hi-lo+1)
+	for i := lo; i <= hi; i++ {
+		vals = append(vals, float64(i))
+	}
+	return percentileSorted(vals, q)
+}
+
+// TestPxxStringForms verifies arbitrary percentile aggregation names.
+//
+// TestPxxStringForms 验证任意百分位聚合名的字符串形式。
+func TestPxxStringForms(t *testing.T) {
+	if Pxx(95) != AggP95 {
+		t.Fatalf("Pxx(95)=%q, want %q", Pxx(95), AggP95)
+	}
+	if Pxx(50) != AggP50 {
+		t.Fatalf("Pxx(50)=%q, want %q", Pxx(50), AggP50)
+	}
+	if Pxx(99.9) != Aggregation("p99.9") {
+		t.Fatalf("Pxx(99.9)=%q", Pxx(99.9))
+	}
+	// Out-of-range percentiles must be rejected by validation.
+	q := AggregateQuery{Query: Query{MetricName: "m", Start: time.Now(), End: time.Now().Add(time.Hour)}, Aggregation: Pxx(100), Interval: time.Minute}
+	if err := q.Validate(); !errors.Is(err, ErrInvalidArgument) {
+		t.Fatalf("p100 should be invalid, got %v", err)
+	}
+}
+
+// TestCompactBuildsFinestTier verifies Compact builds correct finest-tier
+// rollup statistics from raw points.
+//
+// TestCompactBuildsFinestTier 验证 Compact 能从原始点构建正确的最细层
+// rollup 统计。
+func TestCompactBuildsFinestTier(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		Tiers: []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "m", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	// Two 1-minute buckets: bucket A has values 0..59, bucket B has 60..119.
+	var batch []Point
+	for i := 0; i < 120; i++ {
+		batch = append(batch, Point{MetricName: "m", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Second), Value: float64(i)})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	now := base.Add(10 * time.Minute)
+	written, err := s.Compact(ctx, now)
+	if err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+	if written != 2 {
+		t.Fatalf("expected 2 rollup buckets written, got %d", written)
+	}
+	// AggregateRollup at the 1m resolution should reproduce per-bucket stats.
+	res, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query:       Query{MetricName: "m", EntityID: "n1", Start: base, End: base.Add(10 * time.Minute)},
+		Aggregation: AggAvg,
+		Interval:    time.Minute,
+	}, time.Minute)
+	if err != nil {
+		t.Fatalf("aggregate rollup: %v", err)
+	}
+	if len(res) != 2 {
+		t.Fatalf("expected 2 buckets, got %d", len(res))
+	}
+	if res[0].Value != 29.5 || res[0].Count != 60 {
+		t.Fatalf("bucket A avg/count wrong: %#v", res[0])
+	}
+	if res[1].Value != 89.5 || res[1].Count != 60 {
+		t.Fatalf("bucket B avg/count wrong: %#v", res[1])
+	}
+	// Sum/min/max derived from rollup must match raw.
+	for _, tc := range []struct {
+		agg   Aggregation
+		wantA float64
+		wantB float64
+	}{
+		{AggSum, 1770, 5370}, // sum 0..59 and 60..119
+		{AggMin, 0, 60},
+		{AggMax, 59, 119},
+	} {
+		r, err := s.AggregateRollup(ctx, AggregateQuery{
+			Query:       Query{MetricName: "m", EntityID: "n1", Start: base, End: base.Add(10 * time.Minute)},
+			Aggregation: tc.agg,
+			Interval:    time.Minute,
+		}, time.Minute)
+		if err != nil {
+			t.Fatalf("rollup %s: %v", tc.agg, err)
+		}
+		if r[0].Value != tc.wantA || r[1].Value != tc.wantB {
+			t.Fatalf("%s: got %v,%v want %v,%v", tc.agg, r[0].Value, r[1].Value, tc.wantA, tc.wantB)
+		}
+	}
+}
+
+// TestCompactCascadeFineToCoarse verifies a coarse tier composed from a fine
+// tier matches a direct raw rollup.
+//
+// TestCompactCascadeFineToCoarse 验证由细层合成的粗层与直接从原始点计算的
+// rollup 一致。
+func TestCompactCascadeFineToCoarse(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		Tiers: []RollupTier{
+			{Interval: time.Minute, Retention: time.Hour},
+			{Interval: 5 * time.Minute, Retention: 24 * time.Hour},
+		},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "c", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	// 5 minutes of data, one point per second, value = second index 0..299.
+	var batch []Point
+	for i := 0; i < 300; i++ {
+		batch = append(batch, Point{MetricName: "c", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Second), Value: float64(i)})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(time.Hour)); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+	// The single 5m coarse bucket should cover all 300 points.
+	res, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query:       Query{MetricName: "c", EntityID: "n1", Start: base, End: base.Add(5 * time.Minute)},
+		Aggregation: AggAvg,
+		Interval:    5 * time.Minute,
+	}, 5*time.Minute)
+	if err != nil {
+		t.Fatalf("rollup: %v", err)
+	}
+	if len(res) != 1 || res[0].Count != 300 {
+		t.Fatalf("expected 1 coarse bucket covering 300 points, got %#v", res)
+	}
+	if math.Abs(res[0].Value-149.5) > 1e-9 { // avg of 0..299
+		t.Fatalf("coarse avg wrong: %v", res[0].Value)
+	}
+	// Percentile from the cascaded coarse digest should be close to the exact.
+	pres, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query:       Query{MetricName: "c", EntityID: "n1", Start: base, End: base.Add(5 * time.Minute)},
+		Aggregation: Pxx(95),
+		Interval:    5 * time.Minute,
+	}, 5*time.Minute)
+	if err != nil {
+		t.Fatalf("rollup p95: %v", err)
+	}
+	exact := percentileSortedRange(0, 299, 0.95)
+	if math.Abs(pres[0].Value-exact)/exact > 0.02 {
+		t.Fatalf("cascaded p95 %v too far from exact %v", pres[0].Value, exact)
+	}
+}
+
+func TestCompactDoesNotOverwriteCoarseRollupWithPartialFineRows(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 10 * time.Minute,
+		Tiers: []RollupTier{
+			{Interval: time.Minute, Retention: 11 * time.Minute},
+			{Interval: 5 * time.Minute, Retention: time.Hour},
+		},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "partial", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var batch []Point
+	for i := 0; i < 300; i++ {
+		batch = append(batch, Point{MetricName: "partial", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Second), Value: 1})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(15*time.Minute)); err != nil {
+		t.Fatalf("compact initial: %v", err)
+	}
+	query := AggregateQuery{
+		Query:       Query{MetricName: "partial", EntityID: "n1", Start: base, End: base.Add(5*time.Minute - time.Nanosecond)},
+		Aggregation: AggCount,
+		Interval:    5 * time.Minute,
+	}
+	before, err := s.AggregateRollup(ctx, query, 5*time.Minute)
+	if err != nil {
+		t.Fatalf("rollup before: %v", err)
+	}
+	if len(before) != 1 || before[0].Count != 300 {
+		t.Fatalf("expected initial complete coarse bucket, got %#v", before)
+	}
+	if _, err := s.Compact(ctx, base.Add(18*time.Minute)); err != nil {
+		t.Fatalf("compact after fine retention moves: %v", err)
+	}
+	after, err := s.AggregateRollup(ctx, query, 5*time.Minute)
+	if err != nil {
+		t.Fatalf("rollup after: %v", err)
+	}
+	if len(after) != 1 || after[0].Count != 300 {
+		t.Fatalf("coarse bucket was overwritten from partial fine rows: %#v", after)
+	}
+	if err := s.Write(ctx, Point{MetricName: "partial", EntityID: "n1", Timestamp: base.Add(4*time.Minute + 30*time.Second), Value: 1}); err != nil {
+		t.Fatalf("write late: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(19*time.Minute)); err != nil {
+		t.Fatalf("compact late: %v", err)
+	}
+	late, err := s.AggregateRollup(ctx, query, 5*time.Minute)
+	if err != nil {
+		t.Fatalf("rollup after late: %v", err)
+	}
+	if len(late) != 1 || late[0].Count != 301 {
+		t.Fatalf("late fine delta should merge into retained coarse bucket: %#v", late)
+	}
+	if _, err := s.Compact(ctx, base.Add(20*time.Minute)); err != nil {
+		t.Fatalf("compact late again: %v", err)
+	}
+	again, err := s.AggregateRollup(ctx, query, 5*time.Minute)
+	if err != nil {
+		t.Fatalf("rollup after repeated compact: %v", err)
+	}
+	if len(again) != 1 || again[0].Count != 301 {
+		t.Fatalf("repeated compact should not merge the same late delta twice: %#v", again)
+	}
+}
+
+func TestCompactMergesLateFineDeltaLargerThanCoarseBucket(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 10 * time.Minute,
+		Tiers: []RollupTier{
+			{Interval: time.Minute, Retention: 11 * time.Minute},
+			{Interval: 5 * time.Minute, Retention: time.Hour},
+		},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "latebig", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	if err := s.Write(ctx, Point{MetricName: "latebig", EntityID: "n1", Timestamp: base.Add(10 * time.Second), Value: 1}); err != nil {
+		t.Fatalf("write original: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(15*time.Minute)); err != nil {
+		t.Fatalf("compact initial: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(18*time.Minute)); err != nil {
+		t.Fatalf("compact expire fine: %v", err)
+	}
+	for _, ts := range []time.Time{base.Add(20 * time.Second), base.Add(30 * time.Second)} {
+		if err := s.Write(ctx, Point{MetricName: "latebig", EntityID: "n1", Timestamp: ts, Value: 1}); err != nil {
+			t.Fatalf("write late: %v", err)
+		}
+	}
+	if _, err := s.Compact(ctx, base.Add(19*time.Minute)); err != nil {
+		t.Fatalf("compact late: %v", err)
+	}
+	got, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query:       Query{MetricName: "latebig", EntityID: "n1", Start: base, End: base.Add(5*time.Minute - time.Nanosecond)},
+		Aggregation: AggCount,
+		Interval:    5 * time.Minute,
+	}, 5*time.Minute)
+	if err != nil {
+		t.Fatalf("rollup: %v", err)
+	}
+	if len(got) != 1 || got[0].Count != 3 {
+		t.Fatalf("late delta should merge even when larger than existing bucket, got %#v", got)
+	}
+}
+
+// TestRetentionDropsRawButPercentileSurvives verifies the TSDB property that
+// raw data can age out while percentiles remain answerable from rollups.
+//
+// TestRetentionDropsRawButPercentileSurvives 验证 TSDB 特性：原始数据可以过期，
+// 但百分位仍能从 rollup 中回答。
+func TestRetentionDropsRawButPercentileSurvives(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 2 * time.Minute,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 365 * 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "old", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var batch []Point
+	for i := 0; i < 60; i++ { // one minute bucket, values 0..59
+		batch = append(batch, Point{MetricName: "old", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Second), Value: float64(i)})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	// Compact far in the future: raw (older than 2m) is deleted, rollup kept.
+	now := base.Add(48 * time.Hour)
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+	// Raw must be gone.
+	raw, err := s.Query(ctx, Query{MetricName: "old", EntityID: "n1", Start: base, End: base.Add(time.Hour)})
+	if err != nil {
+		t.Fatalf("query raw: %v", err)
+	}
+	if len(raw) != 0 {
+		t.Fatalf("expected raw points deleted by retention, got %d", len(raw))
+	}
+	// Percentile must still be answerable from the surviving rollup.
+	res, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query:       Query{MetricName: "old", EntityID: "n1", Start: base, End: base.Add(time.Hour)},
+		Aggregation: Pxx(90),
+		Interval:    time.Minute,
+	}, time.Minute)
+	if err != nil {
+		t.Fatalf("rollup p90 after retention: %v", err)
+	}
+	if len(res) != 1 {
+		t.Fatalf("expected 1 surviving rollup bucket, got %d", len(res))
+	}
+	exact := percentileSortedRange(0, 59, 0.90)
+	if math.Abs(res[0].Value-exact)/exact > 0.05 {
+		t.Fatalf("surviving p90 %v too far from exact %v", res[0].Value, exact)
+	}
+}
+
+// TestCompactMergesLateRawIntoExpiredRollup verifies a late-arriving raw point
+// for an already-retained rollup bucket is folded into the stored rollup instead
+// of replacing the bucket with only the late sample.
+//
+// TestCompactMergesLateRawIntoExpiredRollup 验证已过原始保留期的 rollup 桶收到
+// 迟到 raw 点时，会把迟到样本合入已有 rollup，而不是用迟到样本覆盖整桶。
+func TestCompactMergesLateRawIntoExpiredRollup(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: 2 * time.Minute,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "late", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var batch []Point
+	for i := 0; i < 60; i++ {
+		batch = append(batch, Point{MetricName: "late", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Second), Value: float64(i)})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(time.Hour)); err != nil {
+		t.Fatalf("compact initial: %v", err)
+	}
+	if err := s.Write(ctx, Point{
+		MetricName: "late",
+		EntityID:   "n1",
+		Timestamp:  base.Add(30*time.Second + 500*time.Millisecond),
+		Value:      1000,
+	}); err != nil {
+		t.Fatalf("write late: %v", err)
+	}
+	if _, err := s.Compact(ctx, base.Add(2*time.Hour)); err != nil {
+		t.Fatalf("compact late: %v", err)
+	}
+	res, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query:       Query{MetricName: "late", EntityID: "n1", Start: base, End: base.Add(time.Minute)},
+		Aggregation: AggSum,
+		Interval:    time.Minute,
+	}, time.Minute)
+	if err != nil {
+		t.Fatalf("rollup sum: %v", err)
+	}
+	if len(res) != 1 {
+		t.Fatalf("expected 1 rollup bucket, got %d", len(res))
+	}
+	if res[0].Count != 61 || res[0].Value != 2770 {
+		t.Fatalf("late point should merge into existing bucket count=61 sum=2770, got %#v", res[0])
+	}
+}
+
+// TestSeriesRoutesByAge verifies Series auto-routes between raw and rollup data
+// based on the query window's age.
+//
+// TestSeriesRoutesByAge 验证 Series 会根据查询窗口的新旧程度在原始数据和
+// rollup 数据之间自动路由。
+func TestSeriesRoutesByAge(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: time.Hour,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 30 * 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "sr", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	now := time.Date(2026, 6, 18, 12, 0, 0, 0, time.UTC)
+	// Recent window (10 min ago) stays raw; old window (2 days ago) uses rollup.
+	recent := now.Add(-10 * time.Minute)
+	old := now.Add(-48 * time.Hour)
+	var batch []Point
+	for i := 0; i < 60; i++ {
+		batch = append(batch, Point{MetricName: "sr", EntityID: "n1", Timestamp: recent.Add(time.Duration(i) * time.Second), Value: float64(i)})
+		batch = append(batch, Point{MetricName: "sr", EntityID: "n1", Timestamp: old.Add(time.Duration(i) * time.Second), Value: float64(100 + i)})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact: %v", err)
+	}
+
+	// Recent window: Series should match raw Aggregate exactly (full fidelity).
+	recentQ := AggregateQuery{
+		Query:       Query{MetricName: "sr", EntityID: "n1", Start: recent.Add(-time.Minute), End: recent.Add(2 * time.Minute)},
+		Aggregation: AggAvg, Interval: time.Minute,
+	}
+	got, err := s.Series(ctx, recentQ, now)
+	if err != nil {
+		t.Fatalf("series recent: %v", err)
+	}
+	want, err := s.Aggregate(ctx, recentQ)
+	if err != nil {
+		t.Fatalf("aggregate recent: %v", err)
+	}
+	if len(got) != len(want) {
+		t.Fatalf("recent: series=%d raw=%d buckets", len(got), len(want))
+	}
+	for i := range got {
+		if !got[i].Bucket.Equal(want[i].Bucket) || got[i].Value != want[i].Value {
+			t.Fatalf("recent bucket %d differs: series=%#v raw=%#v", i, got[i], want[i])
+		}
+	}
+
+	// Old window: raw is gone, so Series must answer from the rollup tier.
+	oldRaw, err := s.Query(ctx, Query{MetricName: "sr", EntityID: "n1", Start: old, End: old.Add(time.Hour)})
+	if err != nil {
+		t.Fatalf("query old raw: %v", err)
+	}
+	if len(oldRaw) != 0 {
+		t.Fatalf("expected old raw aged out, got %d", len(oldRaw))
+	}
+	oldQ := AggregateQuery{
+		Query:       Query{MetricName: "sr", EntityID: "n1", Start: old.Add(-time.Minute), End: old.Add(2 * time.Minute)},
+		Aggregation: AggAvg, Interval: time.Minute,
+	}
+	oldGot, err := s.Series(ctx, oldQ, now)
+	if err != nil {
+		t.Fatalf("series old: %v", err)
+	}
+	var found bool
+	for _, p := range oldGot {
+		if p.Count == 60 && math.Abs(p.Value-129.5) < 1e-9 { // avg of 100..159
+			found = true
+		}
+	}
+	if !found {
+		t.Fatalf("old window not served from rollup correctly: %#v", oldGot)
+	}
+}
+
+// TestSeriesAcrossRetentionIncludesUncompactedRecentRaw verifies a query that
+// spans the raw-retention boundary uses rollups for old buckets and raw data for
+// recent buckets that may not have been compacted yet.
+//
+// TestSeriesAcrossRetentionIncludesUncompactedRecentRaw 验证跨原始保留期边界的查询
+// 会用 rollup 回答旧桶，并包含可能尚未 compact 的近期 raw 桶。
+func TestSeriesAcrossRetentionIncludesUncompactedRecentRaw(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{
+		RawRetention: time.Hour,
+		Tiers:        []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}},
+	}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "hybrid", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	now := time.Date(2026, 6, 18, 12, 0, 0, 0, time.UTC)
+	old := now.Add(-2 * time.Hour)
+	var batch []Point
+	for i := 0; i < 60; i++ {
+		batch = append(batch, Point{MetricName: "hybrid", EntityID: "n1", Timestamp: old.Add(time.Duration(i) * time.Second), Value: float64(i)})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write old: %v", err)
+	}
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact old: %v", err)
+	}
+
+	recent := now.Add(-10 * time.Minute)
+	if err := s.Write(ctx, Point{MetricName: "hybrid", EntityID: "n1", Timestamp: recent, Value: 500}); err != nil {
+		t.Fatalf("write recent: %v", err)
+	}
+
+	got, err := s.Series(ctx, AggregateQuery{
+		Query: Query{
+			MetricName: "hybrid",
+			EntityID:   "n1",
+			Start:      old.Add(-time.Minute),
+			End:        recent.Add(time.Minute),
+		},
+		Aggregation: AggCount,
+		Interval:    time.Minute,
+	}, now)
+	if err != nil {
+		t.Fatalf("series: %v", err)
+	}
+	var oldFound, recentFound bool
+	for _, p := range got {
+		if p.Bucket.Equal(alignTime(old, time.Minute)) && p.Count == 60 {
+			oldFound = true
+		}
+		if p.Bucket.Equal(alignTime(recent, time.Minute)) && p.Count == 1 {
+			recentFound = true
+		}
+	}
+	if !oldFound || !recentFound {
+		t.Fatalf("series should include old rollup and recent raw buckets (old=%v recent=%v): %#v", oldFound, recentFound, got)
+	}
+}
+
+// TestRollupPolicyValidate verifies rollup policy validation rules.
+//
+// TestRollupPolicyValidate 验证 rollup 策略校验规则。
+func TestRollupPolicyValidate(t *testing.T) {
+	ok := RollupPolicy{
+		RawRetention: 10 * time.Minute,
+		Tiers: []RollupTier{
+			{Interval: time.Minute, Retention: time.Hour},
+			{Interval: 5 * time.Minute, Retention: 24 * time.Hour},
+		},
+	}
+	if err := ok.Validate(); err != nil {
+		t.Fatalf("valid policy rejected: %v", err)
+	}
+	bad := []RollupPolicy{
+		{Tiers: []RollupTier{{Interval: 0, Retention: time.Hour}}},                                                                  // zero interval
+		{Tiers: []RollupTier{{Interval: time.Minute, Retention: 0}}},                                                                // zero retention
+		{Tiers: []RollupTier{{Interval: time.Minute, Retention: time.Hour}, {Interval: 90 * time.Second, Retention: time.Hour}}},    // not a multiple
+		{Tiers: []RollupTier{{Interval: time.Minute, Retention: 2 * time.Hour}, {Interval: 5 * time.Minute, Retention: time.Hour}}}, // coarse retention < fine
+		{RawRetention: 30 * time.Second, Tiers: []RollupTier{{Interval: time.Minute, Retention: time.Hour}}},                        // raw retention < 2x finest
+	}
+	for i, p := range bad {
+		if err := p.Validate(); !errors.Is(err, ErrInvalidArgument) {
+			t.Fatalf("bad policy %d should be invalid, got %v", i, err)
+		}
+	}
+}
+
+// TestCompactIdempotent verifies Compact is idempotent for unchanged windows.
+//
+// TestCompactIdempotent 验证 Compact 对未变化窗口保持幂等。
+func TestCompactIdempotent(t *testing.T) {
+	ctx := context.Background()
+	policy := RollupPolicy{Tiers: []RollupTier{{Interval: time.Minute, Retention: 24 * time.Hour}}}
+	s := newRollupStore(t, policy)
+	if err := s.CreateMetric(ctx, Definition{Name: "idem", Type: TypeGauge}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+	base := time.Date(2026, 6, 18, 0, 0, 0, 0, time.UTC)
+	var batch []Point
+	for i := 0; i < 120; i++ {
+		batch = append(batch, Point{MetricName: "idem", EntityID: "n1", Timestamp: base.Add(time.Duration(i) * time.Second), Value: float64(i)})
+	}
+	if err := s.WriteBatch(ctx, batch); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	now := base.Add(10 * time.Minute)
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact 1: %v", err)
+	}
+	r1, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query:       Query{MetricName: "idem", EntityID: "n1", Start: base, End: base.Add(10 * time.Minute)},
+		Aggregation: AggAvg, Interval: time.Minute,
+	}, time.Minute)
+	if err != nil {
+		t.Fatalf("rollup 1: %v", err)
+	}
+	if _, err := s.Compact(ctx, now); err != nil {
+		t.Fatalf("compact 2: %v", err)
+	}
+	r2, err := s.AggregateRollup(ctx, AggregateQuery{
+		Query:       Query{MetricName: "idem", EntityID: "n1", Start: base, End: base.Add(10 * time.Minute)},
+		Aggregation: AggAvg, Interval: time.Minute,
+	}, time.Minute)
+	if err != nil {
+		t.Fatalf("rollup 2: %v", err)
+	}
+	if len(r1) != len(r2) {
+		t.Fatalf("idempotency broken: %d vs %d buckets", len(r1), len(r2))
+	}
+	for i := range r1 {
+		if r1[i].Value != r2[i].Value || r1[i].Count != r2[i].Count {
+			t.Fatalf("idempotency broken at %d: %#v vs %#v", i, r1[i], r2[i])
+		}
+	}
+}

pkg/metric/store_test.go

@@ -0,0 +1,146 @@
+package metric
+
+import (
+	"context"
+	"errors"
+	"math"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+)
+
+// TestSQLiteStoreWriteQueryAggregate verifies SQLite write, query, aggregate, and stats.
+//
+// TestSQLiteStoreWriteQueryAggregate 验证 SQLite 写入、查询、聚合和统计。
+func TestSQLiteStoreWriteQueryAggregate(t *testing.T) {
+	ctx := context.Background()
+	store, err := Open(ctx, SQLite("file:test-metric?mode=memory&cache=shared"))
+	if err != nil {
+		t.Fatalf("open sqlite store: %v", err)
+	}
+	defer store.Close()
+
+	if err := store.CreateMetric(ctx, Definition{Name: "cpu.usage", Type: TypeGauge, Unit: "%"}); err != nil {
+		t.Fatalf("create metric: %v", err)
+	}
+
+	base := time.Date(2026, 6, 18, 10, 0, 0, 0, time.UTC)
+	points := []Point{
+		{MetricName: "cpu.usage", EntityID: "server-1", Timestamp: base, Value: 10, Tags: map[string]string{"region": "ap"}},
+		{MetricName: "cpu.usage", EntityID: "server-1", Timestamp: base.Add(time.Minute), Value: 20, Tags: map[string]string{"region": "ap"}},
+		{MetricName: "cpu.usage", EntityID: "server-1", Timestamp: base.Add(2 * time.Minute), Value: 30, Tags: map[string]string{"region": "ap"}},
+		{MetricName: "cpu.usage", EntityID: "server-2", Timestamp: base.Add(time.Minute), Value: 99, Tags: map[string]string{"region": "eu"}},
+	}
+	if err := store.WriteBatch(ctx, points); err != nil {
+		t.Fatalf("write batch: %v", err)
+	}
+
+	got, err := store.Query(ctx, Query{
+		MetricName: "cpu.usage",
+		EntityID:   "server-1",
+		Start:      base.Add(-time.Second),
+		End:        base.Add(3 * time.Minute),
+		Tags:       map[string]string{"region": "ap"},
+	})
+	if err != nil {
+		t.Fatalf("query: %v", err)
+	}
+	if len(got) != 3 {
+		t.Fatalf("expected 3 points, got %d", len(got))
+	}
+	if got[0].Value != 10 || got[2].Value != 30 {
+		t.Fatalf("unexpected ordered values: %#v", got)
+	}
+
+	agg, err := store.Aggregate(ctx, AggregateQuery{
+		Query: Query{
+			MetricName: "cpu.usage",
+			EntityID:   "server-1",
+			Start:      base,
+			End:        base.Add(3 * time.Minute),
+		},
+		Aggregation: AggAvg,
+		Interval:    2 * time.Minute,
+	})
+	if err != nil {
+		t.Fatalf("aggregate: %v", err)
+	}
+	if len(agg) != 2 {
+		t.Fatalf("expected 2 aggregate buckets, got %d", len(agg))
+	}
+	if agg[0].Value != 15 || agg[0].Count != 2 {
+		t.Fatalf("unexpected first aggregate: %#v", agg[0])
+	}
+
+	stats, err := store.Stats(ctx, Query{
+		MetricName: "cpu.usage",
+		EntityID:   "server-1",
+		Start:      base,
+		End:        base.Add(3 * time.Minute),
+	})
+	if err != nil {
+		t.Fatalf("stats: %v", err)
+	}
+	if stats.Count != 3 || stats.Avg != 20 || stats.P95 != 29 {
+		t.Fatalf("unexpected stats: %#v", stats)
+	}
+}
+
+func TestWriteRejectsNonFiniteValues(t *testing.T) {
+	ctx := context.Background()
+	store, err := Open(ctx, SQLite("file:test-non-finite?mode=memory&cache=shared"))
+	if err != nil {
+		t.Fatalf("open sqlite store: %v", err)
+	}
+	defer store.Close()
+
+	for _, value := range []float64{math.NaN(), math.Inf(1), math.Inf(-1)} {
+		err := store.Write(ctx, Point{
+			MetricName: "bad",
+			EntityID:   "server-1",
+			Timestamp:  time.Now(),
+			Value:      value,
+		})
+		if !errors.Is(err, ErrInvalidArgument) {
+			t.Fatalf("expected ErrInvalidArgument for %v, got %v", value, err)
+		}
+	}
+}
+
+// TestSQLiteInDirCreatesDirectoryAndAppliesPragmas verifies SQLite file setup and PRAGMAs.
+//
+// TestSQLiteInDirCreatesDirectoryAndAppliesPragmas 验证 SQLite 文件初始化和 PRAGMA 设置。
+func TestSQLiteInDirCreatesDirectoryAndAppliesPragmas(t *testing.T) {
+	ctx := context.Background()
+	dir := filepath.Join(t.TempDir(), "metrics")
+	store, err := Open(ctx, SQLiteInDir(
+		dir,
+		WithSQLiteProfile(SQLiteProfilePerformance),
+		WithSQLiteCacheSizeKB(32*1024),
+	))
+	if err != nil {
+		t.Fatalf("open sqlite dir store: %v", err)
+	}
+	defer store.Close()
+
+	if _, err := os.Stat(filepath.Join(dir, "metrics.db")); err != nil {
+		t.Fatalf("expected sqlite database file to be created: %v", err)
+	}
+
+	var journalMode string
+	if err := store.db.QueryRowContext(ctx, "PRAGMA journal_mode").Scan(&journalMode); err != nil {
+		t.Fatalf("query journal mode: %v", err)
+	}
+	if journalMode != "wal" {
+		t.Fatalf("expected WAL journal mode, got %q", journalMode)
+	}
+
+	var synchronous int
+	if err := store.db.QueryRowContext(ctx, "PRAGMA synchronous").Scan(&synchronous); err != nil {
+		t.Fatalf("query synchronous: %v", err)
+	}
+	if synchronous != 0 {
+		t.Fatalf("expected performance profile synchronous=OFF(0), got %d", synchronous)
+	}
+}

pkg/metric/tdigest.go

@@ -0,0 +1,331 @@
+package metric
+
+import (
+	"encoding/binary"
+	"errors"
+	"math"
+	"sort"
+)
+
+// TDigest is a mergeable sketch for estimating arbitrary quantiles of a stream
+// of float64 values. It is the piece that makes downsampling lossless *enough*
+// for percentiles: count/sum/min/max can be re-aggregated exactly when rolling
+// finer buckets into coarser ones, but a percentile cannot be recovered from
+// those scalars. Storing a small t-digest per rollup bucket lets a coarse
+// bucket's pXX be computed by merging the digests of the finer buckets it
+// covers, with bounded error and bounded size.
+//
+// This is the "merging" variant of Dunning & Ertl's t-digest. Centroids near
+// the median are allowed to absorb more weight (lower resolution where the CDF
+// is flat) while centroids in the tails stay small (high resolution where
+// accuracy matters), giving relative error that is small at the extremes — the
+// regime that matters for p95/p99/p99.9 latency work.
+//
+// TDigest 是用于估算 float64 数据流任意分位数的可合并 sketch。它让降采样对
+// 百分位来说“足够无损”：count/sum/min/max 可以在把细桶合入粗桶时精确重聚合，
+// 但仅靠这些标量无法恢复百分位。每个 rollup 桶保存一个小型 t-digest 后，
+// 粗桶的 pXX 就可以通过合并其覆盖的细桶 digest 来计算，并保持有界误差和有界大小。
+//
+// 这是 Dunning 与 Ertl 的 t-digest 的“merging”变体。靠近中位数的质心允许吸收
+// 更多权重（CDF 平坦处使用较低分辨率），尾部质心则保持较小（精度重要处使用
+// 较高分辨率），从而在极端位置取得较小相对误差，这正是 p95/p99/p99.9 延迟分析
+// 最关心的区间。
+type TDigest struct {
+	// compression controls the size/accuracy tradeoff.
+	//
+	// compression 控制大小与精度之间的取舍。
+	compression float64
+	// centroids stores the digest's weighted clusters.
+	//
+	// centroids 保存 digest 的加权簇。
+	centroids []centroid
+	// count is the total observed weight.
+	//
+	// count 是已观测的总权重。
+	count float64
+	// min is the minimum observed value.
+	//
+	// min 是已观测的最小值。
+	min float64
+	// max is the maximum observed value.
+	//
+	// max 是已观测的最大值。
+	max float64
+	// processed reports whether centroids is sorted+merged. Add() appends
+	// unprocessed singletons and flips this false; process() restores it.
+	//
+	// processed 表示 centroids 是否已排序并合并。Add() 会追加未处理的单点并把它
+	// 置为 false；process() 会恢复它。
+	processed bool
+}
+
+// centroid stores one t-digest centroid.
+//
+// centroid 保存一个 t-digest 质心。
+type centroid struct {
+	// mean is the centroid's weighted mean.
+	//
+	// mean 是质心的加权均值。
+	mean float64
+	// weight is the total weight represented by the centroid.
+	//
+	// weight 是该质心代表的总权重。
+	weight float64
+}
+
+const (
+	// defaultTDigestCompression is used when a caller supplies no useful value.
+	//
+	// defaultTDigestCompression 在调用方未提供有效值时使用。
+	defaultTDigestCompression = 100.0
+	// tdigestMagic0 is the first magic byte in the binary format.
+	//
+	// tdigestMagic0 是二进制格式的第一个 magic 字节。
+	tdigestMagic0 = 'T'
+	// tdigestMagic1 is the second magic byte in the binary format.
+	//
+	// tdigestMagic1 是二进制格式的第二个 magic 字节。
+	tdigestMagic1 = 'D'
+	// tdigestVersion is the current binary encoding version.
+	//
+	// tdigestVersion 是当前二进制编码版本。
+	tdigestVersion = 1
+)
+
+// NewTDigest returns an empty digest. compression trades size for accuracy;
+// higher keeps more centroids. Values <= 1 fall back to the default (100),
+// which keeps each digest to a few KB while holding tail error to well under
+// 1% for typical distributions.
+//
+// NewTDigest 返回一个空 digest。compression 在大小和精度之间取舍；值越高会保留
+// 更多质心。小于等于 1 的值会回退到默认值（100），这通常能把每个 digest 控制在
+// 几 KB，同时让典型分布的尾部误差远低于 1%。
+func NewTDigest(compression float64) *TDigest {
+	if compression <= 1 {
+		compression = defaultTDigestCompression
+	}
+	return &TDigest{
+		compression: compression,
+		min:         math.Inf(1),
+		max:         math.Inf(-1),
+		processed:   true,
+	}
+}
+
+// Add folds a single observation with weight w (w must be > 0) into the digest.
+//
+// Add 将一个观测值及其权重合入摘要；权重必须大于 0。
+func (t *TDigest) Add(x, w float64) {
+	if w <= 0 || math.IsNaN(x) || math.IsInf(x, 0) {
+		return
+	}
+	t.centroids = append(t.centroids, centroid{mean: x, weight: w})
+	t.count += w
+	if x < t.min {
+		t.min = x
+	}
+	if x > t.max {
+		t.max = x
+	}
+	t.processed = false
+	// Bound the unprocessed buffer so a long stream cannot grow memory without
+	// limit; process() collapses it back to ~compression centroids.
+	if len(t.centroids) > int(8*t.compression)+16 {
+		t.process()
+	}
+}
+
+// Merge folds every centroid of other into t. This is the operation rollup
+// composition relies on: a coarse bucket merges the digests of the finer
+// buckets it spans.
+//
+// Merge 将 other 的每个质心合入 t。rollup 合成依赖这个操作：粗桶会合并它所覆盖的
+// 细桶 digest。
+func (t *TDigest) Merge(other *TDigest) {
+	if other == nil || other.count == 0 {
+		return
+	}
+	other.process()
+	for _, c := range other.centroids {
+		t.centroids = append(t.centroids, c)
+	}
+	t.count += other.count
+	if other.min < t.min {
+		t.min = other.min
+	}
+	if other.max > t.max {
+		t.max = other.max
+	}
+	t.processed = false
+	t.process()
+}
+
+// Count returns the total weight observed.
+//
+// Count 返回已观测样本的总权重。
+func (t *TDigest) Count() float64 { return t.count }
+
+// process sorts the buffered centroids by mean and merges adjacent ones while
+// the merged weight stays under the quantile-dependent size limit
+// 4*N*q*(1-q)/compression. That limit is generous near q=0.5 and tightens to
+// near zero in the tails, which is exactly the t-digest accuracy profile.
+//
+// process 将缓冲质心按均值排序，并在合并后权重仍低于分位数相关大小限制
+// 4*N*q*(1-q)/compression 时合并相邻质心。这个限制在 q=0.5 附近较宽松，
+// 到尾部会收紧到接近零，这正是 t-digest 的精度分布特征。
+func (t *TDigest) process() {
+	if t.processed {
+		return
+	}
+	if len(t.centroids) == 0 {
+		t.processed = true
+		return
+	}
+	sort.Slice(t.centroids, func(i, j int) bool {
+		return t.centroids[i].mean < t.centroids[j].mean
+	})
+	total := t.count
+	merged := t.centroids[:0:0] // fresh backing array; don't alias input mid-merge
+	cur := t.centroids[0]
+	weightBefore := 0.0
+	for i := 1; i < len(t.centroids); i++ {
+		next := t.centroids[i]
+		proposed := cur.weight + next.weight
+		// Quantile at the center of the proposed combined centroid.
+		q := (weightBefore + proposed/2) / total
+		limit := 4 * total * q * (1 - q) / t.compression
+		if proposed <= limit || limit < 1 && proposed <= 1 {
+			// Weighted-mean update keeps the centroid's mean exact.
+			cur.mean += next.weight * (next.mean - cur.mean) / proposed
+			cur.weight = proposed
+		} else {
+			merged = append(merged, cur)
+			weightBefore += cur.weight
+			cur = next
+		}
+	}
+	merged = append(merged, cur)
+	t.centroids = merged
+	t.processed = true
+}
+
+// Quantile estimates the value at q in [0,1] using linear interpolation between
+// centroid centers, with the extreme tails anchored to the observed min/max.
+//
+// Quantile 用质心中心之间的线性插值估算 [0,1] 分位点，极端尾部锚定到
+// 已观测的最小值和最大值。
+func (t *TDigest) Quantile(q float64) float64 {
+	t.process()
+	n := len(t.centroids)
+	if n == 0 {
+		return math.NaN()
+	}
+	if q <= 0 {
+		return t.min
+	}
+	if q >= 1 {
+		return t.max
+	}
+	if n == 1 {
+		return t.centroids[0].mean
+	}
+	index := q * t.count
+
+	// Head: between the observed min and the first centroid's center.
+	c0 := t.centroids[0]
+	if index < c0.weight/2 {
+		z := index / (c0.weight / 2)
+		return t.min + (c0.mean-t.min)*z
+	}
+	weightSoFar := c0.weight / 2
+	for i := 0; i < n-1; i++ {
+		c := t.centroids[i]
+		next := t.centroids[i+1]
+		dw := (c.weight + next.weight) / 2
+		if index < weightSoFar+dw {
+			z := (index - weightSoFar) / dw
+			return c.mean*(1-z) + next.mean*z
+		}
+		weightSoFar += dw
+	}
+	// Tail: between the last centroid's center and the observed max.
+	cl := t.centroids[n-1]
+	z := (index - weightSoFar) / (cl.weight / 2)
+	if z > 1 {
+		z = 1
+	}
+	return cl.mean + (t.max-cl.mean)*z
+}
+
+// Encode serializes the (processed) digest to a compact little-endian blob:
+// magic[2] version[1] compression[8] min[8] max[8] count[8] nCentroids[4]
+// then nCentroids * (mean[8] weight[8]).
+//
+// Encode 将处理后的 digest 序列化为紧凑的小端二进制 blob：
+// magic[2] version[1] compression[8] min[8] max[8] count[8] nCentroids[4]，
+// 后接 nCentroids * (mean[8] weight[8])。
+func (t *TDigest) Encode() []byte {
+	t.process()
+	n := len(t.centroids)
+	buf := make([]byte, 0, 3+8*4+4+n*16)
+	buf = append(buf, tdigestMagic0, tdigestMagic1, tdigestVersion)
+	var tmp [8]byte
+	putF := func(f float64) {
+		binary.LittleEndian.PutUint64(tmp[:], math.Float64bits(f))
+		buf = append(buf, tmp[:]...)
+	}
+	putF(t.compression)
+	putF(t.min)
+	putF(t.max)
+	putF(t.count)
+	var u32 [4]byte
+	binary.LittleEndian.PutUint32(u32[:], uint32(n))
+	buf = append(buf, u32[:]...)
+	for _, c := range t.centroids {
+		putF(c.mean)
+		putF(c.weight)
+	}
+	return buf
+}
+
+// DecodeTDigest reconstructs a digest produced by Encode. A nil/empty blob
+// yields an empty digest so callers can treat "no sketch stored" uniformly.
+//
+// DecodeTDigest 还原 Encode 生成的摘要；nil 或空 blob 会返回空摘要，
+// 方便调用方统一处理“没有保存 sketch”的情况。
+func DecodeTDigest(b []byte) (*TDigest, error) {
+	if len(b) == 0 {
+		return NewTDigest(defaultTDigestCompression), nil
+	}
+	if len(b) < 3+8*4+4 || b[0] != tdigestMagic0 || b[1] != tdigestMagic1 {
+		return nil, errors.New("metric: invalid t-digest blob")
+	}
+	if b[2] != tdigestVersion {
+		return nil, errors.New("metric: unsupported t-digest version")
+	}
+	off := 3
+	getF := func() float64 {
+		v := math.Float64frombits(binary.LittleEndian.Uint64(b[off : off+8]))
+		off += 8
+		return v
+	}
+	t := &TDigest{processed: true}
+	t.compression = getF()
+	t.min = getF()
+	t.max = getF()
+	t.count = getF()
+	n := int(binary.LittleEndian.Uint32(b[off : off+4]))
+	off += 4
+	if n < 0 || off+n*16 > len(b) {
+		return nil, errors.New("metric: truncated t-digest blob")
+	}
+	t.centroids = make([]centroid, n)
+	for i := 0; i < n; i++ {
+		t.centroids[i].mean = getF()
+		t.centroids[i].weight = getF()
+	}
+	if t.compression <= 1 {
+		t.compression = defaultTDigestCompression
+	}
+	return t, nil
+}

pkg/metric/tdigest_test.go

@@ -0,0 +1,101 @@
+package metric
+
+import (
+	"math"
+	"math/rand"
+	"sort"
+	"testing"
+)
+
+// exactQuantile computes an exact quantile for a sample slice.
+//
+// exactQuantile 为样本切片计算精确分位数。
+func exactQuantile(xs []float64, q float64) float64 {
+	s := append([]float64(nil), xs...)
+	sort.Float64s(s)
+	return percentileSorted(s, q)
+}
+
+// TestTDigestAccuracySmoke checks basic t-digest accuracy.
+//
+// TestTDigestAccuracySmoke 检查 t-digest 的基本精度。
+func TestTDigestAccuracySmoke(t *testing.T) {
+	rng := rand.New(rand.NewSource(42))
+	var xs []float64
+	td := NewTDigest(100)
+	for i := 0; i < 100000; i++ {
+		// Exponential-ish: skewed, heavy right tail (latency-like).
+		x := math.Abs(rng.NormFloat64())*100 + rng.ExpFloat64()*50
+		xs = append(xs, x)
+		td.Add(x, 1)
+	}
+	for _, q := range []float64{0.5, 0.9, 0.95, 0.99, 0.999} {
+		exact := exactQuantile(xs, q)
+		est := td.Quantile(q)
+		relErr := math.Abs(est-exact) / math.Abs(exact)
+		t.Logf("q=%.3f exact=%.3f est=%.3f relErr=%.4f", q, exact, est, relErr)
+		if relErr > 0.02 {
+			t.Errorf("q=%.3f relErr %.4f exceeds 2%%", q, relErr)
+		}
+	}
+}
+
+// Rollup composition: many finer buckets, each a digest over points drawn from
+// the SAME distribution, merged into one coarse digest. Its quantiles must
+// track the quantiles of all the raw points combined.
+//
+// 该测试模拟 rollup 合成：多个细桶的 digest 合并成一个粗桶 digest，合并后
+// 的分位数应接近所有原始点合在一起计算出的精确分位数。
+func TestTDigestMergeMatchesCombined(t *testing.T) {
+	rng := rand.New(rand.NewSource(7))
+	var all []float64
+	coarse := NewTDigest(100)
+	for b := 0; b < 24; b++ { // 24 fine buckets -> 1 coarse
+		fine := NewTDigest(100)
+		for i := 0; i < 5000; i++ {
+			x := math.Abs(rng.NormFloat64())*80 + 100
+			all = append(all, x)
+			fine.Add(x, 1)
+		}
+		coarse.Merge(fine)
+	}
+	if coarse.Count() != float64(len(all)) {
+		t.Fatalf("merged weight %v != %d points", coarse.Count(), len(all))
+	}
+	for _, q := range []float64{0.5, 0.9, 0.95, 0.99, 0.999} {
+		exact := exactQuantile(all, q)
+		est := coarse.Quantile(q)
+		relErr := math.Abs(est-exact) / math.Abs(exact)
+		t.Logf("merged q=%.3f exact=%.3f est=%.3f relErr=%.4f", q, exact, est, relErr)
+		if relErr > 0.02 {
+			t.Errorf("merged q=%.3f relErr %.4f exceeds 2%%", q, relErr)
+		}
+	}
+}
+
+// TestTDigestEncodeRoundTrip verifies t-digest serialization.
+//
+// TestTDigestEncodeRoundTrip 验证 t-digest 编码和解码往返。
+func TestTDigestEncodeRoundTrip(t *testing.T) {
+	td := NewTDigest(50)
+	for i := 0; i < 2000; i++ {
+		td.Add(float64(i%137)+0.5, 1)
+	}
+	blob := td.Encode()
+	back, err := DecodeTDigest(blob)
+	if err != nil {
+		t.Fatalf("decode: %v", err)
+	}
+	if back.Count() != td.Count() {
+		t.Fatalf("count mismatch: %v vs %v", back.Count(), td.Count())
+	}
+	for _, q := range []float64{0.1, 0.5, 0.9, 0.99} {
+		if math.Abs(back.Quantile(q)-td.Quantile(q)) > 1e-9 {
+			t.Fatalf("q=%v mismatch after round-trip: %v vs %v", q, back.Quantile(q), td.Quantile(q))
+		}
+	}
+	// Empty blob -> empty digest, no error.
+	if _, err := DecodeTDigest(nil); err != nil {
+		t.Fatalf("decode nil: %v", err)
+	}
+}

pkg/metric/types.go

@@ -0,0 +1,481 @@
+package metric
+
+import (
+	"fmt"
+	"math"
+	"strings"
+	"time"
+)
+
+// Driver names a supported database backend.
+//
+// Driver 表示 metric store 使用的数据库后端类型。
+type Driver string
+
+const (
+	// DriverSQLite selects the SQLite backend.
+	//
+	// DriverSQLite 选择 SQLite 后端。
+	DriverSQLite Driver = "sqlite"
+	// DriverMySQL selects the MySQL backend.
+	//
+	// DriverMySQL 选择 MySQL 后端。
+	DriverMySQL Driver = "mysql"
+	// DriverPostgreSQL selects the PostgreSQL backend.
+	//
+	// DriverPostgreSQL 选择 PostgreSQL 后端。
+	DriverPostgreSQL Driver = "postgresql"
+)
+
+// MetricType describes the semantic type of a metric.
+//
+// MetricType 表示指标的语义类型。
+type MetricType string
+
+const (
+	// TypeGauge represents a point-in-time value.
+	//
+	// TypeGauge 表示某一时刻的数值。
+	TypeGauge MetricType = "gauge"
+	// TypeCounter represents a monotonically increasing counter.
+	//
+	// TypeCounter 表示单调递增计数器。
+	TypeCounter MetricType = "counter"
+	// TypeHistogram represents histogram-style measurements.
+	//
+	// TypeHistogram 表示直方图类度量。
+	TypeHistogram MetricType = "histogram"
+	// TypeSummary represents pre-summarized measurements.
+	//
+	// TypeSummary 表示预汇总类度量。
+	TypeSummary MetricType = "summary"
+)
+
+// Aggregation names a supported aggregation operation.
+//
+// Aggregation 表示聚合方式，例如 avg、p95 或 rate。
+type Aggregation string
+
+const (
+	// AggAvg computes the arithmetic mean.
+	//
+	// AggAvg 计算算术平均值。
+	AggAvg Aggregation = "avg"
+	// AggMin computes the minimum value.
+	//
+	// AggMin 计算最小值。
+	AggMin Aggregation = "min"
+	// AggMax computes the maximum value.
+	//
+	// AggMax 计算最大值。
+	AggMax Aggregation = "max"
+	// AggSum computes the sum of values.
+	//
+	// AggSum 计算值的总和。
+	AggSum Aggregation = "sum"
+	// AggCount counts the number of points.
+	//
+	// AggCount 计算点数量。
+	AggCount Aggregation = "count"
+	// AggP50 computes the 50th percentile.
+	//
+	// AggP50 计算第 50 百分位。
+	AggP50 Aggregation = "p50"
+	// AggP95 computes the 95th percentile.
+	//
+	// AggP95 计算第 95 百分位。
+	AggP95 Aggregation = "p95"
+	// AggP99 computes the 99th percentile.
+	//
+	// AggP99 计算第 99 百分位。
+	AggP99 Aggregation = "p99"
+	// AggFirst returns the first value in time order.
+	//
+	// AggFirst 返回时间顺序上的第一个值。
+	AggFirst Aggregation = "first"
+	// AggLast returns the last value in time order.
+	//
+	// AggLast 返回时间顺序上的最后一个值。
+	AggLast Aggregation = "last"
+	// AggRate computes the reset-aware per-second rate.
+	//
+	// AggRate 计算可处理重置的每秒速率。
+	AggRate Aggregation = "rate"
+	// AggStdDev computes the population standard deviation.
+	//
+	// AggStdDev 计算总体标准差。
+	AggStdDev Aggregation = "stddev"
+)
+
+// Order controls chronological query ordering.
+//
+// Order 表示查询结果按时间升序或降序排列。
+type Order string
+
+const (
+	// OrderAsc orders points from oldest to newest.
+	//
+	// OrderAsc 按从旧到新的顺序排列点。
+	OrderAsc Order = "asc"
+	// OrderDesc orders points from newest to oldest.
+	//
+	// OrderDesc 按从新到旧的顺序排列点。
+	OrderDesc Order = "desc"
+)
+
+// Definition describes a metric and its metadata.
+//
+// Definition 描述一个指标的元数据和保留策略。
+type Definition struct {
+	// Name is the unique metric name.
+	//
+	// Name 是唯一指标名称。
+	Name string `json:"name"`
+	// Description is optional human-readable metric text.
+	//
+	// Description 是可选的人类可读指标说明。
+	Description string `json:"description,omitempty"`
+	// Type describes the metric's semantic type.
+	//
+	// Type 描述指标的语义类型。
+	Type MetricType `json:"type"`
+	// Unit names the value unit, such as bytes or percent.
+	//
+	// Unit 表示数值单位，例如 bytes 或 percent。
+	Unit string `json:"unit,omitempty"`
+	// RetentionDays controls raw-point retention for this metric.
+	//
+	// RetentionDays 控制该指标原始点的保留天数。
+	RetentionDays int `json:"retention_days,omitempty"`
+	// Metadata stores caller-defined metric metadata.
+	//
+	// Metadata 保存调用方定义的指标元数据。
+	Metadata map[string]string `json:"metadata,omitempty"`
+	// CreatedAt records when the metric definition was created.
+	//
+	// CreatedAt 记录指标定义创建时间。
+	CreatedAt time.Time `json:"created_at,omitempty"`
+	// UpdatedAt records when the metric definition was last updated.
+	//
+	// UpdatedAt 记录指标定义最后更新时间。
+	UpdatedAt time.Time `json:"updated_at,omitempty"`
+}
+
+// withDefaults fills default values on a metric definition.
+//
+// withDefaults 为指标定义填充默认类型和默认保留天数。
+func (d Definition) withDefaults(defaultRetentionDays int) Definition {
+	if d.Type == "" {
+		d.Type = TypeGauge
+	}
+	if d.RetentionDays == 0 {
+		d.RetentionDays = defaultRetentionDays
+	}
+	return d
+}
+
+// Validate checks whether the value is well formed.
+//
+// Validate 检查指标定义是否合法。
+func (d Definition) Validate() error {
+	if strings.TrimSpace(d.Name) == "" {
+		return fmt.Errorf("%w: metric name is required", ErrInvalidArgument)
+	}
+	switch d.Type {
+	case "", TypeGauge, TypeCounter, TypeHistogram, TypeSummary:
+	default:
+		return fmt.Errorf("%w: unsupported metric type %q", ErrInvalidArgument, d.Type)
+	}
+	if d.RetentionDays < 0 {
+		return fmt.Errorf("%w: retention days cannot be negative", ErrInvalidArgument)
+	}
+	return nil
+}
+
+// Point stores one metric sample.
+//
+// Point 表示某个实体在某一时刻的一条指标采样。
+type Point struct {
+	// MetricName names the metric this sample belongs to.
+	//
+	// MetricName 表示该采样所属的指标名称。
+	MetricName string `json:"metric_name"`
+	// EntityID identifies the entity that emitted the sample.
+	//
+	// EntityID 标识产生该采样的实体。
+	EntityID string `json:"entity_id"`
+	// Timestamp is the sample time.
+	//
+	// Timestamp 是采样时间。
+	Timestamp time.Time `json:"timestamp"`
+	// Value is the numeric sample value.
+	//
+	// Value 是采样的数值。
+	Value float64 `json:"value"`
+	// Tags identify the logical series within a metric and entity.
+	//
+	// Tags 标识同一指标和实体下的逻辑序列。
+	Tags map[string]string `json:"tags,omitempty"`
+	// Labels carry extra metadata that does not define series identity.
+	//
+	// Labels 携带不参与序列身份判定的额外元数据。
+	Labels map[string]string `json:"labels,omitempty"`
+}
+
+// Validate checks whether the value is well formed.
+//
+// Validate 检查采样点是否包含必要字段。
+func (p Point) Validate() error {
+	if strings.TrimSpace(p.MetricName) == "" {
+		return fmt.Errorf("%w: metric name is required", ErrInvalidArgument)
+	}
+	if strings.TrimSpace(p.EntityID) == "" {
+		return fmt.Errorf("%w: entity id is required", ErrInvalidArgument)
+	}
+	if p.Timestamp.IsZero() {
+		return fmt.Errorf("%w: timestamp is required", ErrInvalidArgument)
+	}
+	if math.IsNaN(p.Value) || math.IsInf(p.Value, 0) {
+		return fmt.Errorf("%w: value must be finite", ErrInvalidArgument)
+	}
+	return nil
+}
+
+// normalized returns a canonical form of the value.
+//
+// normalized 将采样点时间规范化为 UTC，并补齐空标签和标签说明 map。
+func (p Point) normalized() Point {
+	p.Timestamp = p.Timestamp.UTC()
+	if p.Tags == nil {
+		p.Tags = map[string]string{}
+	}
+	if p.Labels == nil {
+		p.Labels = map[string]string{}
+	}
+	return p
+}
+
+// Query loads raw metric points matching a query.
+//
+// Query 描述原始点查询条件，包括指标、实体、时间范围、标签和分页。
+type Query struct {
+	// MetricName restricts the query to one metric.
+	//
+	// MetricName 将查询限定到单个指标。
+	MetricName string `json:"metric_name"`
+	// EntityID optionally restricts the query to one entity.
+	//
+	// EntityID 可选地将查询限定到单个实体。
+	EntityID string `json:"entity_id,omitempty"`
+	// Start is the inclusive query start time.
+	//
+	// Start 是包含边界的查询起始时间。
+	Start time.Time `json:"start"`
+	// End is the inclusive query end time.
+	//
+	// End 是包含边界的查询结束时间。
+	End time.Time `json:"end"`
+	// Tags filters points by exact tag key/value matches.
+	//
+	// Tags 按标签键值精确匹配过滤采样点。
+	Tags map[string]string `json:"tags,omitempty"`
+	// Limit limits the number of raw points returned.
+	//
+	// Limit 限制返回的原始点数量。
+	Limit int `json:"limit,omitempty"`
+	// Offset skips this many raw points before returning results.
+	//
+	// Offset 在返回结果前跳过指定数量的原始点。
+	Offset int `json:"offset,omitempty"`
+	// Order controls chronological result ordering.
+	//
+	// Order 控制结果的时间顺序。
+	Order Order `json:"order,omitempty"`
+}
+
+// Validate checks whether the value is well formed.
+//
+// Validate 检查原始查询条件是否合法。
+func (q Query) Validate() error {
+	if strings.TrimSpace(q.MetricName) == "" {
+		return fmt.Errorf("%w: metric name is required", ErrInvalidArgument)
+	}
+	if q.Start.IsZero() || q.End.IsZero() {
+		return fmt.Errorf("%w: start and end time are required", ErrInvalidArgument)
+	}
+	if q.End.Before(q.Start) {
+		return fmt.Errorf("%w: end time cannot be before start time", ErrInvalidArgument)
+	}
+	if q.Limit < 0 || q.Offset < 0 {
+		return fmt.Errorf("%w: limit and offset cannot be negative", ErrInvalidArgument)
+	}
+	switch q.Order {
+	case "", OrderAsc, OrderDesc:
+	default:
+		return fmt.Errorf("%w: unsupported order %q", ErrInvalidArgument, q.Order)
+	}
+	return nil
+}
+
+// normalized returns a canonical form of the value.
+//
+// normalized 将查询时间规范化为 UTC，并设置默认排序。
+func (q Query) normalized() Query {
+	q.Start = q.Start.UTC()
+	q.End = q.End.UTC()
+	if q.Order == "" {
+		q.Order = OrderAsc
+	}
+	return q
+}
+
+// AggregateQuery describes a bucketed aggregate query.
+//
+// AggregateQuery 描述按固定时间间隔分桶后的聚合查询。
+type AggregateQuery struct {
+	// Query supplies the raw series filter and time window.
+	//
+	// Query 提供原始序列过滤条件和时间窗口。
+	Query
+	// Aggregation selects the bucket aggregation to compute.
+	//
+	// Aggregation 选择要计算的桶聚合方式。
+	Aggregation Aggregation `json:"aggregation"`
+	// Interval is the width of each aggregate bucket.
+	//
+	// Interval 是每个聚合桶的宽度。
+	Interval time.Duration `json:"interval"`
+	// FillEmpty emits zero-count buckets for empty intervals.
+	//
+	// FillEmpty 会为空时间段输出零计数桶。
+	FillEmpty bool `json:"fill_empty,omitempty"`
+	// BucketLimit and BucketOffset page over the produced aggregate buckets, not
+	// the underlying raw points. They are applied consistently across every
+	// backend and aggregation type. The embedded Query.Limit/Query.Offset are
+	// ignored for aggregation (they describe raw-point paging, which would mean
+	// something different depending on whether the aggregation is pushed down to
+	// SQL or computed in memory).
+	//
+	// BucketLimit 和 BucketOffset 对生成的聚合桶分页，而不是对底层原始点分页。
+	// 它们会在所有后端和聚合类型上保持一致。嵌入的 Query.Limit/Query.Offset
+	// 在聚合中会被忽略，因为它们描述的是原始点分页；根据聚合是下推到 SQL 还是在
+	// 内存计算，这会产生不同语义。
+	BucketLimit int `json:"bucket_limit,omitempty"`
+	// BucketOffset skips this many aggregate buckets before returning results.
+	//
+	// BucketOffset 在返回结果前跳过指定数量的聚合桶。
+	BucketOffset int `json:"bucket_offset,omitempty"`
+}
+
+// Validate checks whether the value is well formed.
+//
+// Validate 检查聚合查询的时间间隔、分页和聚合类型是否合法。
+func (q AggregateQuery) Validate() error {
+	if err := q.Query.Validate(); err != nil {
+		return err
+	}
+	if q.Interval <= 0 {
+		return fmt.Errorf("%w: aggregate interval must be positive", ErrInvalidArgument)
+	}
+	if q.BucketLimit < 0 || q.BucketOffset < 0 {
+		return fmt.Errorf("%w: bucket limit and offset cannot be negative", ErrInvalidArgument)
+	}
+	switch q.Aggregation {
+	case AggAvg, AggMin, AggMax, AggSum, AggCount, AggFirst, AggLast, AggRate, AggStdDev:
+	default:
+		// Any percentile (p50, p95, p99, p99.9, ...) is also valid. The fixed
+		// AggP50/AggP95/AggP99 constants fall through to here as well.
+		if !isPercentile(q.Aggregation) {
+			return fmt.Errorf("%w: unsupported aggregation %q", ErrInvalidArgument, q.Aggregation)
+		}
+	}
+	return nil
+}
+
+// AggregatePoint stores one aggregate bucket result.
+//
+// AggregatePoint 表示一个聚合桶的结果。
+type AggregatePoint struct {
+	// MetricName is the metric represented by the bucket.
+	//
+	// MetricName 是该桶代表的指标名称。
+	MetricName string `json:"metric_name"`
+	// EntityID is the entity represented by the bucket when one was requested.
+	//
+	// EntityID 是请求限定实体时该桶代表的实体。
+	EntityID string `json:"entity_id,omitempty"`
+	// Bucket is the bucket start time.
+	//
+	// Bucket 是桶起始时间。
+	Bucket time.Time `json:"bucket"`
+	// Value is the computed aggregate value.
+	//
+	// Value 是计算出的聚合值。
+	Value float64 `json:"value"`
+	// Count is the number of points represented by the bucket.
+	//
+	// Count 是该桶代表的点数量。
+	Count int `json:"count"`
+}
+
+// Stats stores or computes summary statistics for a point series.
+//
+// Stats 表示一段原始点序列的统计摘要。
+type Stats struct {
+	// Count is the number of points in the series.
+	//
+	// Count 是序列中的点数量。
+	Count int `json:"count"`
+	// Min is the minimum value.
+	//
+	// Min 是最小值。
+	Min float64 `json:"min"`
+	// Max is the maximum value.
+	//
+	// Max 是最大值。
+	Max float64 `json:"max"`
+	// Avg is the arithmetic mean.
+	//
+	// Avg 是算术平均值。
+	Avg float64 `json:"avg"`
+	// Sum is the sum of all values.
+	//
+	// Sum 是所有值的总和。
+	Sum float64 `json:"sum"`
+	// P50 is the 50th percentile.
+	//
+	// P50 是第 50 百分位。
+	P50 float64 `json:"p50"`
+	// P95 is the 95th percentile.
+	//
+	// P95 是第 95 百分位。
+	P95 float64 `json:"p95"`
+	// P99 is the 99th percentile.
+	//
+	// P99 是第 99 百分位。
+	P99 float64 `json:"p99"`
+	// First is the first value in time order.
+	//
+	// First 是时间顺序上的第一个值。
+	First float64 `json:"first"`
+	// Last is the last value in time order.
+	//
+	// Last 是时间顺序上的最后一个值。
+	Last float64 `json:"last"`
+	// Rate is the reset-aware per-second rate.
+	//
+	// Rate 是可处理重置的每秒速率。
+	Rate float64 `json:"rate"`
+	// Start is the first point timestamp.
+	//
+	// Start 是第一个点的时间戳。
+	Start time.Time `json:"start"`
+	// End is the last point timestamp.
+	//
+	// End 是最后一个点的时间戳。
+	End time.Time `json:"end"`
+	// StdDev is the population standard deviation.
+	//
+	// StdDev 是总体标准差。
+	StdDev float64 `json:"std_dev"`
+}

pkg/migrations/migrations.go

@@ -18,7 +18,6 @@ type legacyModelConfig struct {
 	ID                         uint    `json:"id,omitempty" gorm:"primaryKey;autoIncrement"`
 	Sitename                   string  `json:"sitename" gorm:"type:varchar(100);not null"`
 	Description                string  `json:"description" gorm:"type:text"`
-	AllowCors                  bool    `json:"allow_cors" gorm:"column:allow_cors;default:false"`
 	Theme                      string  `json:"theme" gorm:"type:varchar(100);default:'default'"`
 	PrivateSite                bool    `json:"private_site" gorm:"default:false"`
 	ApiKey                     string  `json:"api_key" gorm:"type:varchar(255);default:''"`
@@ -57,7 +56,6 @@ type legacyConfig struct {
 	ID                         uint      `json:"id,omitempty"`
 	Sitename                   string    `json:"sitename"`
 	Description                string    `json:"description"`
-	AllowCors                  bool      `json:"allow_cors"`
 	Theme                      string    `json:"theme"`
 	PrivateSite                bool      `json:"private_site"`
 	ApiKey                     string    `json:"api_key"`
@@ -112,9 +110,6 @@ func Run(ctx Context) error {
 	legacyConfigTable := hasLegacyConfigTable(db)
 
 	if legacyConfigTable {
-		if err := migrateLegacyConfigColumns(db); err != nil {
-			return err
-		}
 		if err := migrateLegacyOidcConfig(db); err != nil {
 			return err
 		}
@@ -162,14 +157,6 @@ func hasTableColumn(db *gorm.DB, tableName, columnName string) bool {
 	return false
 }
 
-func migrateLegacyConfigColumns(db *gorm.DB) error {
-	if hasTableColumn(db, "configs", "allow_cros") {
-		log.Println("[>0.0.5a] Renaming column 'allow_cros' to 'allow_cors' in config table...")
-		return db.Migrator().RenameColumn(&legacyModelConfig{}, "allow_cros", "allow_cors")
-	}
-	return nil
-}
-
 func migrateLegacyLoadNotification(db *gorm.DB) error {
 	if db.Migrator().HasColumn(&models.LoadNotification{}, "client") {
 		log.Println("[>0.1.4] Rebuilding LoadNotification table....")

pkg/migrations/migrations_test.go

@@ -163,7 +163,6 @@ func TestRunMigratesLegacyConfigTableToConfigItems(t *testing.T) {
 	legacy := legacyModelConfig{
 		Sitename:                   "Old Komari",
 		Description:                "legacy description",
-		AllowCors:                  true,
 		Theme:                      "classic",
 		GeoIpEnabled:               true,
 		GeoIpProvider:              "ip-api",
@@ -195,12 +194,9 @@ func TestRunMigratesLegacyConfigTableToConfigItems(t *testing.T) {
 		t.Fatalf("unexpected sitename value: %s", sitename.Value)
 	}
 
-	var allowCors appconfig.ConfigItem
-	if err := db.First(&allowCors, "key = ?", appconfig.AllowCorsKey).Error; err != nil {
-		t.Fatalf("find migrated allow_cors: %v", err)
-	}
-	if allowCors.Value != `true` {
-		t.Fatalf("unexpected allow_cors value: %s", allowCors.Value)
+	var corsOriginCheck appconfig.ConfigItem
+	if err := db.First(&corsOriginCheck, "key = ?", appconfig.CorsOriginCheckEnabledKey).Error; err == nil {
+		t.Fatalf("unexpected migrated cors_origin_check_enabled value: %s", corsOriginCheck.Value)
 	}
 }
 

pkg/rpc/context.go

@@ -14,6 +14,10 @@ import (
 // 为了便于扩展，字段保持冗余：既保存结构体，也保存对应 UUID / Token。
 // 未来如需添加字段（如 IP、UserAgent、TraceID 等）直接在此结构体上扩展即可。
 type ContextMeta struct {
+	// Principal 调用主体(身份 + 能力)。新身份模型的权威来源,由身份识别层统一构建。
+	// Permission 字段保留用于向后兼容,等价于 Principal.PrimaryRole()。
+	Principal *Principal
+
 	// Permission 当前权限分组 guest/client/admin
 	Permission string
 	// User 登录的管理员用户（仅 admin 会话存在）
@@ -24,10 +28,14 @@ type ContextMeta struct {
 	ClientToken string
 	// ClientUUID 解析出的客户端 UUID（若 token 合法）
 	ClientUUID string
+	// SessionToken 当前管理员会话的 session token（仅 admin 会话存在，用于区分当前会话等场景）
+	SessionToken string
 	// RemoteIP 请求来源 IP（可选）
 	RemoteIP string
 	// UserAgent 请求 UA（可选）
 	UserAgent string
+	// TempShareValid 临时分享访问许可是否有效（基于 temp_key cookie 校验，由传输层填充）
+	TempShareValid bool
 }
 
 // 私有类型做 key，避免外部冲突

pkg/rpc/permission.go

@@ -0,0 +1,200 @@
+package rpc
+
+// permission.go
+// 声明式 ACL 权限模型。权限由一组规则 (pattern, minRole) 声明，pattern 支持 "*" 通配符，
+// 匹配完整方法名（如 "admin:addClient"、"rpc.ping"）。判权时在所有匹配规则中取
+// 特异性（specificity）最高者：精确匹配 > 字面前缀更长的通配 > 全局 "*"。
+//
+// 该模型面向插件扩展：插件可用 Allow 声明任意粒度的规则（命名空间级 "ns:*" 或方法级），
+// 无需修改核心代码。
+//
+// 角色采用分级语义：guest < client < admin，规则声明所需最低角色。
+
+import (
+	"strings"
+	"sync"
+)
+
+// 角色常量。与 web/api 中的角色保持一致（guest/client/admin）。
+const (
+	RoleGuest  = "guest"
+	RoleClient = "client"
+	RoleAdmin  = "admin"
+)
+
+// roleLevel 定义角色的权限等级。数值越大权限越高。未知角色按 guest（最低）处理。
+var roleLevel = map[string]int{
+	RoleGuest:  0,
+	RoleClient: 1,
+	RoleAdmin:  2,
+}
+
+// DefaultNamespace 是方法名不含 ":" 时归入的命名空间。
+const DefaultNamespace = "common"
+
+// aclRule 一条声明式 ACL 规则。
+type aclRule struct {
+	pattern string // 方法名匹配模式，支持 "*" 通配
+	minRole string // 匹配时所需的最低角色
+	// 预计算的特异性：精确匹配（无通配）给高位 bonus，其余按字面字符数。
+	specificity int
+	hasWildcard bool
+}
+
+var (
+	muACL   sync.RWMutex
+	aclList []aclRule
+)
+
+func init() {
+	// 默认规则覆盖内置命名空间语义。"*" 兜底要求 admin，确保未声明的方法默认最严格。
+	Allow("*", RoleAdmin)
+	Allow("common:*", RoleGuest)
+	Allow("guest:*", RoleGuest)
+	Allow("rpc.*", RoleGuest) // 内部方法 rpc.ping/rpc.help 等（用 "." 分隔）
+	Allow("rpc:*", RoleGuest)
+	Allow("client:*", RoleClient)
+	Allow("admin:*", RoleAdmin)
+}
+
+// levelOf 返回角色的权限等级，未知角色视为 guest。
+func levelOf(role string) int {
+	if lv, ok := roleLevel[role]; ok {
+		return lv
+	}
+	return roleLevel[RoleGuest]
+}
+
+// NamespaceOf 解析方法名的命名空间。"ns:method" 返回 "ns"；无 ":" 返回 DefaultNamespace。
+func NamespaceOf(method string) string {
+	if i := strings.IndexByte(method, ':'); i >= 0 {
+		return method[:i]
+	}
+	return DefaultNamespace
+}
+
+// computeSpecificity 计算 pattern 的特异性。精确匹配（不含通配符）给一个大 bonus
+// 以保证优先级最高；含通配符的按字面（非 "*"）字符数排序，前缀越长越具体。
+func computeSpecificity(pattern string) (spec int, hasWildcard bool) {
+	literal := 0
+	for _, r := range pattern {
+		if r != '*' {
+			literal++
+		} else {
+			hasWildcard = true
+		}
+	}
+	if !hasWildcard {
+		return 1 << 20, false // 精确匹配最高优先
+	}
+	return literal, true
+}
+
+// Allow 声明一条 ACL 规则：匹配 pattern 的方法允许 minRole 及以上等级的角色调用。
+// 同一 pattern 重复声明时覆盖原规则。供插件声明自定义权限。
+func Allow(pattern, minRole string) {
+	spec, hasWildcard := computeSpecificity(pattern)
+	rule := aclRule{pattern: pattern, minRole: minRole, specificity: spec, hasWildcard: hasWildcard}
+	muACL.Lock()
+	defer muACL.Unlock()
+	for i := range aclList {
+		if aclList[i].pattern == pattern {
+			aclList[i] = rule
+			return
+		}
+	}
+	aclList = append(aclList, rule)
+}
+
+// RegisterNamespace 便捷封装：为整个命名空间声明所需最低角色（等价于 Allow("ns:*", role)）。
+func RegisterNamespace(namespace, requiredRole string) {
+	Allow(namespace+":*", requiredRole)
+}
+
+// wildcardMatch 判断 s 是否匹配只含 "*" 通配符的 pattern。"*" 匹配任意（含空）字符序列。
+func wildcardMatch(pattern, s string) bool {
+	// 双指针 + 回溯，O(len(pattern)+len(s))。
+	var (
+		p, str       = 0, 0
+		star         = -1
+		strBackup    = 0
+		lenP, lenStr = len(pattern), len(s)
+	)
+	for str < lenStr {
+		if p < lenP && (pattern[p] == s[str]) {
+			p++
+			str++
+		} else if p < lenP && pattern[p] == '*' {
+			star = p
+			strBackup = str
+			p++
+		} else if star != -1 {
+			p = star + 1
+			strBackup++
+			str = strBackup
+		} else {
+			return false
+		}
+	}
+	for p < lenP && pattern[p] == '*' {
+		p++
+	}
+	return p == lenP
+}
+
+// resolveMinRole 返回 method 适用规则中特异性最高者的所需角色。无匹配规则时默认 admin。
+func resolveMinRole(method string) string {
+	// 归一化：无命名空间分隔符（既无 ":" 也无 "rpc." 内部前缀）的裸方法名归入默认命名空间，
+	// 以便被 "common:*" 等规则匹配，保持与历史行为一致。
+	if !strings.ContainsAny(method, ":") && !strings.HasPrefix(method, "rpc.") {
+		method = DefaultNamespace + ":" + method
+	}
+	muACL.RLock()
+	defer muACL.RUnlock()
+	bestSpec := -1
+	bestRole := RoleAdmin
+	matched := false
+	for i := range aclList {
+		r := &aclList[i]
+		if !wildcardMatch(r.pattern, method) {
+			continue
+		}
+		// 特异性更高者胜出；相同特异性时取更严格（等级更高）的角色，偏向安全。
+		if r.specificity > bestSpec || (r.specificity == bestSpec && levelOf(r.minRole) > levelOf(bestRole)) {
+			bestSpec = r.specificity
+			bestRole = r.minRole
+			matched = true
+		}
+	}
+	if !matched {
+		return RoleAdmin
+	}
+	return bestRole
+}
+
+// RequiredRole 返回调用 method 所需的最低角色。
+func RequiredRole(method string) string {
+	return resolveMinRole(method)
+}
+
+// CheckPermission 判定 group 角色是否有权调用 method。
+func CheckPermission(group, method string) bool {
+	return CheckPrincipal(PrincipalFromRole(group), method)
+}
+
+// CheckPrincipal 基于主体的能力集判定是否有权调用 method。
+// 采用集合成员语义(而非线性等级):方法要求某最低角色,主体须在其角色集中持有该角色;
+// guest 为公共基线,任何主体均隐式可访问。
+// 该模型使 agent 与 admin 成为正交主体:admin 不再自动获得 client 能力(反之亦然),
+// 从而堵住"admin 会话冒充 agent 调用 client:* 上报方法"等越权路径。
+func CheckPrincipal(p *Principal, method string) bool {
+	if p == nil {
+		p = NewAnonymousPrincipal()
+	}
+	min := resolveMinRole(method)
+	if min == RoleGuest {
+		return true // 公共基线,所有主体可访问
+	}
+	return p.HasRole(min)
+}
+

pkg/rpc/permission_test.go

@@ -0,0 +1,128 @@
+package rpc
+
+import (
+	"context"
+	"testing"
+)
+
+func TestNamespaceOf(t *testing.T) {
+	cases := map[string]string{
+		"getNodes":        DefaultNamespace,
+		"admin:addClient": "admin",
+		"client:report":   "client",
+		"rpc.ping":        DefaultNamespace, // 无 ":"，归入默认命名空间
+		":bare":           "",
+	}
+	for method, want := range cases {
+		if got := NamespaceOf(method); got != want {
+			t.Errorf("NamespaceOf(%q) = %q, want %q", method, got, want)
+		}
+	}
+}
+
+func TestCheckPermission(t *testing.T) {
+	cases := []struct {
+		group  string
+		method string
+		want   bool
+	}{
+		{RoleGuest, "common:getNodes", true},
+		{RoleGuest, "getNodes", true},
+		{RoleGuest, "admin:addClient", false},
+		{RoleGuest, "client:report", false},
+		{RoleClient, "client:report", true},
+		{RoleClient, "admin:addClient", false},
+		{RoleAdmin, "admin:addClient", true},
+		{RoleAdmin, "client:report", false},
+		{RoleAdmin, "common:getNodes", true},
+		// 未知命名空间默认要求 admin
+		{RoleGuest, "plugin:foo", false},
+		{RoleAdmin, "plugin:foo", true},
+	}
+	for _, c := range cases {
+		if got := CheckPermission(c.group, c.method); got != c.want {
+			t.Errorf("CheckPermission(%q, %q) = %v, want %v", c.group, c.method, got, c.want)
+		}
+	}
+}
+
+func TestRegisterNamespace(t *testing.T) {
+	RegisterNamespace("myplugin", RoleClient)
+	if !CheckPermission(RoleClient, "myplugin:doThing") {
+		t.Error("client should be allowed in myplugin namespace after registration")
+	}
+	if CheckPermission(RoleGuest, "myplugin:doThing") {
+		t.Error("guest should not be allowed in myplugin namespace")
+	}
+}
+
+func TestAllowWildcardSpecificity(t *testing.T) {
+	// 命名空间默认 admin，但更具体的方法级规则可放宽。
+	RegisterNamespace("acltest", RoleAdmin)
+	Allow("acltest:public*", RoleGuest) // 前缀通配，比 "acltest:*" 更具体
+
+	if !CheckPermission(RoleGuest, "acltest:publicInfo") {
+		t.Error("guest should be allowed: acltest:public* is more specific than acltest:*")
+	}
+	if CheckPermission(RoleGuest, "acltest:secret") {
+		t.Error("guest should be denied acltest:secret (falls back to acltest:* = admin)")
+	}
+
+	// 精确规则优先于任何通配。
+	Allow("acltest:secret", RoleClient)
+	if !CheckPermission(RoleClient, "acltest:secret") {
+		t.Error("client should be allowed by exact rule acltest:secret")
+	}
+	if CheckPermission(RoleGuest, "acltest:secret") {
+		t.Error("guest still denied acltest:secret (exact rule requires client)")
+	}
+}
+
+func TestAllowOverride(t *testing.T) {
+	Allow("override:x", RoleAdmin)
+	if CheckPermission(RoleGuest, "override:x") {
+		t.Fatal("precondition: guest denied")
+	}
+	Allow("override:x", RoleGuest) // 覆盖同一 pattern
+	if !CheckPermission(RoleGuest, "override:x") {
+		t.Error("override should relax override:x to guest")
+	}
+}
+
+func TestInternalMethodsPermission(t *testing.T) {
+	// 内部 rpc.* 方法对 guest 开放。
+	if !CheckPermission(RoleGuest, "rpc.ping") {
+		t.Error("guest should be allowed to call rpc.ping")
+	}
+	// 裸方法名归入 common，对 guest 开放。
+	if !CheckPermission(RoleGuest, "getNodes") {
+		t.Error("guest should be allowed to call bare common method getNodes")
+	}
+}
+
+func TestUnregister(t *testing.T) {
+	method := "unreg.sample"
+	MustRegister(method, func(ctx context.Context, req *JsonRpcRequest) (any, *JsonRpcError) {
+		return "ok", nil
+	})
+	if getHandler(method) == nil {
+		t.Fatal("method should be registered")
+	}
+	if !Unregister(method) {
+		t.Fatal("Unregister should return true for existing method")
+	}
+	if getHandler(method) != nil {
+		t.Fatal("method should be removed after Unregister")
+	}
+	// 重复注销返回 false
+	if Unregister(method) {
+		t.Fatal("Unregister should return false for missing method")
+	}
+	// 保留前缀禁止注销
+	if Unregister("rpc.ping") {
+		t.Fatal("Unregister must refuse rpc.* reserved methods")
+	}
+	if getHandler("rpc.ping") == nil {
+		t.Fatal("rpc.ping must still be registered")
+	}
+}

pkg/rpc/principal.go

@@ -0,0 +1,118 @@
+package rpc
+
+// principal.go
+// 调用主体(Principal)定义。区分不同主体类型(匿名/agent/用户/API Key),
+// 替代原先的单一 group string,使身份信息更结构化、便于后续能力模型扩展。
+
+// PrincipalType 调用主体类型
+type PrincipalType int
+
+const (
+	// PrincipalAnonymous 匿名访客(未认证)
+	PrincipalAnonymous PrincipalType = iota
+	// PrincipalAgent 通过 client token 认证的 agent 客户端
+	PrincipalAgent
+	// PrincipalUser 通过 session cookie 认证的管理员用户
+	PrincipalUser
+	// PrincipalAPIKey 通过 API Key 认证的调用方
+	PrincipalAPIKey
+)
+
+// Principal 调用主体,携带身份信息和能力。
+type Principal struct {
+	// Type 主体类型
+	Type PrincipalType
+	// UserUUID 用户 UUID(PrincipalUser 时存在)
+	UserUUID string
+	// ClientUUID agent 客户端 UUID(PrincipalAgent 时存在)
+	ClientUUID string
+	// IsAPIKey 是否为 API Key 调用(快速判定,等价于 Type==PrincipalAPIKey)
+	IsAPIKey bool
+	// Roles 角色/能力集。默认由 Type 推导:
+	//   - PrincipalAnonymous → [RoleGuest]
+	//   - PrincipalAgent → [RoleClient]
+	//   - PrincipalUser / PrincipalAPIKey → [RoleAdmin]
+	// 未来可扩展为多角色(只读 admin / API Key scope 等)。
+	Roles []string
+}
+
+// NewAnonymousPrincipal 创建匿名访客主体
+func NewAnonymousPrincipal() *Principal {
+	return &Principal{
+		Type:  PrincipalAnonymous,
+		Roles: []string{RoleGuest},
+	}
+}
+
+// NewAgentPrincipal 创建 agent 客户端主体
+func NewAgentPrincipal(clientUUID string) *Principal {
+	return &Principal{
+		Type:       PrincipalAgent,
+		ClientUUID: clientUUID,
+		Roles:      []string{RoleClient},
+	}
+}
+
+// NewUserPrincipal 创建管理员用户主体
+func NewUserPrincipal(userUUID string) *Principal {
+	return &Principal{
+		Type:     PrincipalUser,
+		UserUUID: userUUID,
+		Roles:    []string{RoleAdmin},
+	}
+}
+
+// NewAPIKeyPrincipal 创建 API Key 调用主体
+func NewAPIKeyPrincipal() *Principal {
+	return &Principal{
+		Type:     PrincipalAPIKey,
+		IsAPIKey: true,
+		Roles:    []string{RoleAdmin},
+	}
+}
+
+// PrimaryRole 返回主体的主要角色(兼容现有单角色模型)。
+// 多角色场景下返回权限等级最高的那个。
+func (p *Principal) PrimaryRole() string {
+	if p == nil || len(p.Roles) == 0 {
+		return RoleGuest
+	}
+	bestRole := p.Roles[0]
+	bestLevel := levelOf(bestRole)
+	for _, r := range p.Roles[1:] {
+		if lv := levelOf(r); lv > bestLevel {
+			bestLevel = lv
+			bestRole = r
+		}
+	}
+	return bestRole
+}
+
+// HasRole 判断主体是否拥有指定角色
+func (p *Principal) HasRole(role string) bool {
+	if p == nil {
+		return false
+	}
+	for _, r := range p.Roles {
+		if r == role {
+			return true
+		}
+	}
+	return false
+}
+
+// PrincipalFromRole 按角色构造一个最小主体,用于内部调用(OnInternalRequest)等
+// 仅知道角色、无具体身份信息的场景。Type 按角色合理推断:
+//   guest → Anonymous, client → Agent, admin → User。
+// 注意:此构造不携带 UUID/token,仅用于权限判定与兜底,不应据此做审计 actor 归属。
+func PrincipalFromRole(role string) *Principal {
+	switch role {
+	case RoleClient:
+		return &Principal{Type: PrincipalAgent, Roles: []string{RoleClient}}
+	case RoleAdmin:
+		return &Principal{Type: PrincipalUser, Roles: []string{RoleAdmin}}
+	default:
+		return NewAnonymousPrincipal()
+	}
+}
+

pkg/rpc/principal_test.go

@@ -0,0 +1,107 @@
+package rpc
+
+import "testing"
+
+func TestPrincipalConstructors(t *testing.T) {
+	cases := []struct {
+		name     string
+		p        *Principal
+		wantType PrincipalType
+		wantRole string
+		wantKey  bool
+	}{
+		{"anonymous", NewAnonymousPrincipal(), PrincipalAnonymous, RoleGuest, false},
+		{"agent", NewAgentPrincipal("c-uuid"), PrincipalAgent, RoleClient, false},
+		{"user", NewUserPrincipal("u-uuid"), PrincipalUser, RoleAdmin, false},
+		{"apikey", NewAPIKeyPrincipal(), PrincipalAPIKey, RoleAdmin, true},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			if tc.p.Type != tc.wantType {
+				t.Errorf("Type = %v, want %v", tc.p.Type, tc.wantType)
+			}
+			if got := tc.p.PrimaryRole(); got != tc.wantRole {
+				t.Errorf("PrimaryRole() = %q, want %q", got, tc.wantRole)
+			}
+			if tc.p.IsAPIKey != tc.wantKey {
+				t.Errorf("IsAPIKey = %v, want %v", tc.p.IsAPIKey, tc.wantKey)
+			}
+		})
+	}
+}
+
+func TestPrincipalConstructorsFields(t *testing.T) {
+	if p := NewAgentPrincipal("c-uuid"); p.ClientUUID != "c-uuid" {
+		t.Errorf("agent ClientUUID = %q, want %q", p.ClientUUID, "c-uuid")
+	}
+	if p := NewUserPrincipal("u-uuid"); p.UserUUID != "u-uuid" {
+		t.Errorf("user UserUUID = %q, want %q", p.UserUUID, "u-uuid")
+	}
+}
+
+func TestPrimaryRolePicksHighest(t *testing.T) {
+	p := &Principal{Roles: []string{RoleGuest, RoleAdmin, RoleClient}}
+	if got := p.PrimaryRole(); got != RoleAdmin {
+		t.Errorf("PrimaryRole() = %q, want %q (highest level)", got, RoleAdmin)
+	}
+}
+
+func TestPrimaryRoleEmptyOrNil(t *testing.T) {
+	if got := (&Principal{}).PrimaryRole(); got != RoleGuest {
+		t.Errorf("empty Roles PrimaryRole() = %q, want %q", got, RoleGuest)
+	}
+	var p *Principal
+	if got := p.PrimaryRole(); got != RoleGuest {
+		t.Errorf("nil principal PrimaryRole() = %q, want %q", got, RoleGuest)
+	}
+}
+
+func TestHasRole(t *testing.T) {
+	p := NewUserPrincipal("u")
+	if !p.HasRole(RoleAdmin) {
+		t.Error("user principal should have admin role")
+	}
+	if p.HasRole(RoleClient) {
+		t.Error("user principal should not have client role")
+	}
+	var nilP *Principal
+	if nilP.HasRole(RoleGuest) {
+		t.Error("nil principal should not have any role")
+	}
+}
+
+func TestCheckPrincipal(t *testing.T) {
+	cases := []struct {
+		name   string
+		p      *Principal
+		method string
+		want   bool
+	}{
+		// agent 主体:可调 client 与 common,不可调 admin
+		{"agent->client", NewAgentPrincipal("c1"), "client:report", true},
+		{"agent->common", NewAgentPrincipal("c1"), "common:getNodes", true},
+		{"agent->admin", NewAgentPrincipal("c1"), "admin:addClient", false},
+		// user 主体:可调 admin 与 common,不可调 client(正交,堵住冒充)
+		{"user->admin", NewUserPrincipal("u1"), "admin:addClient", true},
+		{"user->common", NewUserPrincipal("u1"), "common:getNodes", true},
+		{"user->client", NewUserPrincipal("u1"), "client:report", false},
+		// api key 主体:等同 admin 能力
+		{"apikey->admin", NewAPIKeyPrincipal(), "admin:addClient", true},
+		{"apikey->client", NewAPIKeyPrincipal(), "client:report", false},
+		// 匿名主体:仅公共方法
+		{"anon->common", NewAnonymousPrincipal(), "common:getNodes", true},
+		{"anon->admin", NewAnonymousPrincipal(), "admin:addClient", false},
+		{"anon->client", NewAnonymousPrincipal(), "client:report", false},
+		// nil 主体按匿名处理
+		{"nil->common", nil, "common:getNodes", true},
+		{"nil->admin", nil, "admin:addClient", false},
+	}
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			if got := CheckPrincipal(c.p, c.method); got != c.want {
+				t.Errorf("CheckPrincipal(%s, %q) = %v, want %v", c.name, c.method, got, c.want)
+			}
+		})
+	}
+}
+

pkg/rpc/registry.go

@@ -41,6 +41,28 @@ func MustRegister(method string, h Handler) {
 	}
 }
 
+// Unregister 注销已注册的方法，并清理其元数据。
+// 保留前缀 "rpc." 的内部方法禁止注销。返回是否存在并被移除。
+// 主要供插件卸载时动态移除其注册的方法。
+func Unregister(method string) bool {
+	method = strings.TrimSpace(method)
+	if method == "" || strings.HasPrefix(method, "rpc.") {
+		return false
+	}
+	muHandlers.Lock()
+	_, exists := handlers[method]
+	if exists {
+		delete(handlers, method)
+	}
+	muHandlers.Unlock()
+	if exists {
+		muMetas.Lock()
+		delete(methodMetas, method)
+		muMetas.Unlock()
+	}
+	return exists
+}
+
 // ListMethods 列出当前已注册的方法名（副本）
 func ListMethods() []string {
 	muHandlers.RLock()

pkg/rpc/sensitive.go

@@ -0,0 +1,28 @@
+package rpc
+
+// sensitive.go
+// 敏感操作登记表。被标记的方法（如 admin:exec）在通过角色鉴权后，仍需额外的
+// 敏感操作二次验证（sensitive 2FA）。登记与传输无关，由 RPC 边界统一判定，
+// 确保所有调用入口行为一致。
+
+import "sync"
+
+var (
+	muSensitive      sync.RWMutex
+	sensitiveMethods = map[string]bool{}
+)
+
+// MarkSensitive 标记某方法为敏感操作。method 为完整方法名（如 "admin:exec"）。
+// 重复标记是幂等的。供方法注册处声明。
+func MarkSensitive(method string) {
+	muSensitive.Lock()
+	defer muSensitive.Unlock()
+	sensitiveMethods[method] = true
+}
+
+// IsSensitive 判断方法是否被标记为敏感操作。
+func IsSensitive(method string) bool {
+	muSensitive.RLock()
+	defer muSensitive.RUnlock()
+	return sensitiveMethods[method]
+}

utils/messageSender/telegram/telegram.go

@@ -60,16 +60,19 @@ func (t *TelegramSender) SendTextMessage(message, title string) error {
 	}
 	defer resp.Body.Close()
 
-	if resp.StatusCode != http.StatusOK {
-		return fmt.Errorf("telegram API returned non-OK status: %d", resp.StatusCode)
-	}
-
 	var result struct {
 		Ok          bool   `json:"ok"`
 		Description string `json:"description"`
 	}
 	if err := json.NewDecoder(resp.Body).Decode(&result); err != nil {
-		return err
+		return fmt.Errorf("failed to decode response: %v", err)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		if result.Description != "" {
+			return fmt.Errorf("telegram API error: %s", result.Description)
+		}
+		return fmt.Errorf("telegram API returned non-OK status: %d", resp.StatusCode)
 	}
 
 	if !result.Ok {

utils/notifier/traffic_report.go

@@ -1,8 +1,10 @@
 package notifier
 
 import (
+	"errors"
 	"fmt"
 	"log"
+	"sort"
 	"strings"
 	"time"
 
@@ -11,6 +13,7 @@ import (
 	messageevent "github.com/komari-monitor/komari/database/models/messageEvent"
 	"github.com/komari-monitor/komari/pkg/config"
 	"github.com/komari-monitor/komari/pkg/corn"
+	"github.com/komari-monitor/komari/utils"
 	"github.com/komari-monitor/komari/utils/messageSender"
 	"gorm.io/gorm"
 )
@@ -173,49 +176,135 @@ func getClientTrafficInRange(clientUUID string, trafficType string, start, end t
 	return getClientTrafficInRangeWithDB(dbcore.GetDBInstance(), clientUUID, trafficType, start, end)
 }
 
-func getClientTrafficInRangeWithDB(db *gorm.DB, clientUUID string, trafficType string, start, end time.Time) (int64, error) {
-	type trafficDeltaRecord struct {
-		Time        models.LocalTime `gorm:"column:time"`
-		TrafficUp   int64            `gorm:"column:traffic_up"`
-		TrafficDown int64            `gorm:"column:traffic_down"`
-	}
+type trafficDeltaRecord struct {
+	Time         models.LocalTime `gorm:"column:time"`
+	NetTotalUp   int64            `gorm:"column:net_total_up"`
+	NetTotalDown int64            `gorm:"column:net_total_down"`
+	TrafficUp    int64            `gorm:"column:traffic_up"`
+	TrafficDown  int64            `gorm:"column:traffic_down"`
+}
 
+func getClientTrafficInRangeWithDB(db *gorm.DB, clientUUID string, trafficType string, start, end time.Time) (int64, error) {
 	var recentRecords []trafficDeltaRecord
 	if err := db.Table("records").
-		Select("time, traffic_up, traffic_down").
-		Where("client = ? AND time >= ? AND time <= ?", clientUUID, start, end).
+		Select("time, net_total_up, net_total_down, traffic_up, traffic_down").
+		Where("client = ? AND time >= ? AND time <= ?", clientUUID, models.FromTime(start), models.FromTime(end)).
 		Find(&recentRecords).Error; err != nil {
 		return 0, err
 	}
 
 	var longTermRecords []trafficDeltaRecord
 	if err := db.Table("records_long_term").
-		Select("time, traffic_up, traffic_down").
-		Where("client = ? AND time >= ? AND time <= ?", clientUUID, start, end).
+		Select("time, net_total_up, net_total_down, traffic_up, traffic_down").
+		Where("client = ? AND time >= ? AND time <= ?", clientUUID, models.FromTime(start), models.FromTime(end)).
 		Find(&longTermRecords).Error; err != nil {
 		return 0, err
 	}
 
-	var totalUp int64
-	var totalDown int64
-	longTermSlots := make(map[time.Time]struct{}, len(longTermRecords))
+	records := mergeTrafficRecords(recentRecords, longTermRecords)
 
+	sort.Slice(records, func(i, j int) bool {
+		return records[i].Time.ToTime().Before(records[j].Time.ToTime())
+	})
+
+	previous, err := getPreviousTrafficDeltaRecord(db, clientUUID, start)
+	if err != nil {
+		return 0, err
+	}
+
+	totalUp, totalDown := sumTrafficDeltas(records, previous)
+	return computeUsedByType(strings.ToLower(trafficType), totalUp, totalDown), nil
+}
+
+func mergeTrafficRecords(recentRecords, longTermRecords []trafficDeltaRecord) []trafficDeltaRecord {
+	rawSlots := make(map[time.Time]struct{}, len(recentRecords))
+	for _, record := range recentRecords {
+		rawSlots[record.Time.ToTime().Truncate(15*time.Minute)] = struct{}{}
+	}
+
+	longTermSlots := make(map[time.Time]struct{}, len(longTermRecords))
+	records := make([]trafficDeltaRecord, 0, len(longTermRecords)+len(recentRecords))
 	for _, record := range longTermRecords {
-		totalUp += record.TrafficUp
-		totalDown += record.TrafficDown
-		longTermSlots[record.Time.ToTime()] = struct{}{}
+		slot := record.Time.ToTime().Truncate(15 * time.Minute)
+		if _, hasRawSlot := rawSlots[slot]; hasRawSlot && record.TrafficUp == 0 && record.TrafficDown == 0 {
+			continue
+		}
+		longTermSlots[slot] = struct{}{}
+		records = append(records, record)
 	}
 
 	for _, record := range recentRecords {
-		// records_long_term stores one aggregated row per 15-minute slot.
-		// Skip raw rows only when that exact slot is already present there.
 		slot := record.Time.ToTime().Truncate(15 * time.Minute)
 		if _, exists := longTermSlots[slot]; exists {
 			continue
 		}
-		totalUp += record.TrafficUp
-		totalDown += record.TrafficDown
+		records = append(records, record)
 	}
 
-	return computeUsedByType(strings.ToLower(trafficType), totalUp, totalDown), nil
+	return records
+}
+
+func getPreviousTrafficDeltaRecord(db *gorm.DB, clientUUID string, before time.Time) (*trafficDeltaRecord, error) {
+	record, err := latestTrafficDeltaRecordBefore(db.Table("records"), clientUUID, before)
+	if err != nil {
+		return nil, err
+	}
+
+	longTermRecord, err := latestTrafficDeltaRecordBefore(db.Table("records_long_term"), clientUUID, before)
+	if err != nil {
+		return nil, err
+	}
+
+	if record == nil {
+		return longTermRecord, nil
+	}
+	if longTermRecord == nil {
+		return record, nil
+	}
+	if longTermRecord.Time.ToTime().After(record.Time.ToTime()) {
+		return longTermRecord, nil
+	}
+	return record, nil
+}
+
+func latestTrafficDeltaRecordBefore(query *gorm.DB, clientUUID string, before time.Time) (*trafficDeltaRecord, error) {
+	var record trafficDeltaRecord
+	err := query.
+		Select("time, net_total_up, net_total_down, traffic_up, traffic_down").
+		Where("client = ? AND time < ?", clientUUID, models.FromTime(before)).
+		Order("time DESC").
+		First(&record).Error
+	if err == nil {
+		return &record, nil
+	}
+	if errors.Is(err, gorm.ErrRecordNotFound) {
+		return nil, nil
+	}
+	return nil, err
+}
+
+func sumTrafficDeltas(records []trafficDeltaRecord, previous *trafficDeltaRecord) (int64, int64) {
+	var totalUp int64
+	var totalDown int64
+
+	for i := range records {
+		up := records[i].TrafficUp
+		down := records[i].TrafficDown
+		if previous != nil {
+			up = trafficDeltaOrFallback(up, records[i].NetTotalUp, previous.NetTotalUp)
+			down = trafficDeltaOrFallback(down, records[i].NetTotalDown, previous.NetTotalDown)
+		}
+		totalUp += up
+		totalDown += down
+		previous = &records[i]
+	}
+
+	return totalUp, totalDown
+}
+
+func trafficDeltaOrFallback(storedDelta, currentTotal, previousTotal int64) int64 {
+	if storedDelta > 0 {
+		return storedDelta
+	}
+	return utils.ComputeTrafficDelta(currentTotal, previousTotal)
 }

utils/notifier/traffic_report_test.go

@@ -41,6 +41,37 @@ func TestGetClientTrafficInRangeAvoidsOverlappingRawAndLongTermRows(t *testing.T
 	assert.Equal(t, int64(380), used)
 }
 
+func TestGetClientTrafficInRangeNormalizesLongTermSlotForOverlap(t *testing.T) {
+	db, err := gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
+	assert.NoError(t, err)
+	assert.NoError(t, db.AutoMigrate(&models.Record{}))
+	assert.NoError(t, db.Table("records_long_term").AutoMigrate(&models.Record{}))
+
+	clientUUID := "client-overlap-normalized"
+	start := time.Date(2026, 6, 1, 0, 0, 0, 0, time.UTC)
+	sharedSlot := start.Add(15 * time.Minute)
+
+	assert.NoError(t, db.Table("records_long_term").Create(&models.Record{
+		Client:      clientUUID,
+		Time:        models.FromTime(sharedSlot.Add(8 * time.Minute)),
+		TrafficUp:   100,
+		TrafficDown: 200,
+	}).Error)
+
+	rawRecords := []models.Record{
+		{Client: clientUUID, Time: models.FromTime(sharedSlot.Add(1 * time.Minute)), TrafficUp: 40, TrafficDown: 80},
+		{Client: clientUUID, Time: models.FromTime(sharedSlot.Add(5 * time.Minute)), TrafficUp: 60, TrafficDown: 120},
+		{Client: clientUUID, Time: models.FromTime(sharedSlot.Add(16 * time.Minute)), TrafficUp: 30, TrafficDown: 50},
+	}
+	for _, record := range rawRecords {
+		assert.NoError(t, db.Create(&record).Error)
+	}
+
+	used, err := getClientTrafficInRangeWithDB(db, clientUUID, "sum", start, sharedSlot.Add(30*time.Minute))
+	assert.NoError(t, err)
+	assert.Equal(t, int64(380), used)
+}
+
 func TestGetClientTrafficInRangeSumsPersistedDeltasAcrossCounterReset(t *testing.T) {
 	db, err := gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
 	assert.NoError(t, err)
@@ -63,3 +94,94 @@ func TestGetClientTrafficInRangeSumsPersistedDeltasAcrossCounterReset(t *testing
 	assert.NoError(t, err)
 	assert.Equal(t, int64(175), used)
 }
+
+func TestGetClientTrafficInRangeFallsBackForPersistedZeroDeltas(t *testing.T) {
+	db, err := gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
+	assert.NoError(t, err)
+	assert.NoError(t, db.AutoMigrate(&models.Record{}))
+	assert.NoError(t, db.Table("records_long_term").AutoMigrate(&models.Record{}))
+
+	clientUUID := "client-zero-deltas"
+	start := time.Date(2026, 6, 3, 0, 0, 0, 0, time.UTC)
+	records := []models.Record{
+		{Client: clientUUID, Time: models.FromTime(start.Add(-5 * time.Minute)), NetTotalUp: 100, NetTotalDown: 200},
+		{Client: clientUUID, Time: models.FromTime(start.Add(0 * time.Minute)), NetTotalUp: 130, NetTotalDown: 250},
+		{Client: clientUUID, Time: models.FromTime(start.Add(5 * time.Minute)), NetTotalUp: 160, NetTotalDown: 310},
+		{Client: clientUUID, Time: models.FromTime(start.Add(10 * time.Minute)), NetTotalUp: 10, NetTotalDown: 30},
+		{Client: clientUUID, Time: models.FromTime(start.Add(15 * time.Minute)), NetTotalUp: 25, NetTotalDown: 40},
+	}
+	for _, record := range records {
+		assert.NoError(t, db.Create(&record).Error)
+	}
+
+	used, err := getClientTrafficInRangeWithDB(db, clientUUID, "sum", start, start.Add(20*time.Minute))
+	assert.NoError(t, err)
+	assert.Equal(t, int64(235), used)
+}
+
+func TestGetClientTrafficInRangeFallsBackForLongTermZeroDeltas(t *testing.T) {
+	db, err := gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
+	assert.NoError(t, err)
+	assert.NoError(t, db.AutoMigrate(&models.Record{}))
+	assert.NoError(t, db.Table("records_long_term").AutoMigrate(&models.Record{}))
+
+	clientUUID := "client-long-term-zero-deltas"
+	start := time.Date(2026, 6, 4, 0, 0, 0, 0, time.UTC)
+	assert.NoError(t, db.Create(&models.Record{
+		Client:       clientUUID,
+		Time:         models.FromTime(start.Add(-15 * time.Minute)),
+		NetTotalUp:   100,
+		NetTotalDown: 200,
+	}).Error)
+
+	longTermRecords := []models.Record{
+		{Client: clientUUID, Time: models.FromTime(start), NetTotalUp: 140, NetTotalDown: 260},
+		{Client: clientUUID, Time: models.FromTime(start.Add(15 * time.Minute)), NetTotalUp: 180, NetTotalDown: 330},
+	}
+	for _, record := range longTermRecords {
+		assert.NoError(t, db.Table("records_long_term").Create(&record).Error)
+	}
+
+	used, err := getClientTrafficInRangeWithDB(db, clientUUID, "sum", start, start.Add(30*time.Minute))
+	assert.NoError(t, err)
+	assert.Equal(t, int64(210), used)
+}
+
+func TestGetClientTrafficInRangePrefersRawSlotOverZeroLongTermSlot(t *testing.T) {
+	db, err := gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
+	assert.NoError(t, err)
+	assert.NoError(t, db.AutoMigrate(&models.Record{}))
+	assert.NoError(t, db.Table("records_long_term").AutoMigrate(&models.Record{}))
+
+	clientUUID := "client-zero-long-term-with-raw-reset"
+	start := time.Date(2026, 6, 5, 0, 0, 0, 0, time.UTC)
+	slot := start.Add(15 * time.Minute)
+	assert.NoError(t, db.Create(&models.Record{
+		Client:       clientUUID,
+		Time:         models.FromTime(start.Add(-5 * time.Minute)),
+		NetTotalUp:   100,
+		NetTotalDown: 200,
+	}).Error)
+
+	assert.NoError(t, db.Table("records_long_term").Create(&models.Record{
+		Client:       clientUUID,
+		Time:         models.FromTime(slot),
+		NetTotalUp:   15,
+		NetTotalDown: 25,
+		TrafficUp:    0,
+		TrafficDown:  0,
+	}).Error)
+
+	rawRecords := []models.Record{
+		{Client: clientUUID, Time: models.FromTime(slot.Add(1 * time.Minute)), NetTotalUp: 130, NetTotalDown: 240},
+		{Client: clientUUID, Time: models.FromTime(slot.Add(5 * time.Minute)), NetTotalUp: 10, NetTotalDown: 20},
+		{Client: clientUUID, Time: models.FromTime(slot.Add(10 * time.Minute)), NetTotalUp: 15, NetTotalDown: 25},
+	}
+	for _, record := range rawRecords {
+		assert.NoError(t, db.Create(&record).Error)
+	}
+
+	used, err := getClientTrafficInRangeWithDB(db, clientUUID, "sum", start, slot.Add(15*time.Minute))
+	assert.NoError(t, err)
+	assert.Equal(t, int64(110), used)
+}

utils/utils.go

@@ -5,8 +5,8 @@ import (
 	"strings"
 	"time"
 
-	"github.com/komari-monitor/komari/protocol/v1"
 	"github.com/komari-monitor/komari/database/models"
+	"github.com/komari-monitor/komari/protocol/v1"
 )
 
 // AverageReport 根据 topPercentage 参数计算报告的平均值。
@@ -26,6 +26,8 @@ func AverageReport(uuid string, time time.Time, records []v1.Report, topPercenta
 		}
 	}
 
+	latestTotalUp, latestTotalDown := latestNetworkTotals(records)
+
 	// 定义一个辅助函数，用于排序和求和
 	sumAndSort := func(getFloat32Value func(v1.Report) float32, getInt64Value func(v1.Report) int64, isFloat bool) (float32, int64) {
 		if isFloat {
@@ -120,8 +122,8 @@ func AverageReport(uuid string, time time.Time, records []v1.Report, topPercenta
 		DiskTotal:      records[0].Disk.Total,
 		NetIn:          sumNETIn / int64(recordsToAverageCount),
 		NetOut:         sumNETOut / int64(recordsToAverageCount),
-		NetTotalUp:     sumNETTotalUp / int64(recordsToAverageCount),
-		NetTotalDown:   sumNETTotalDown / int64(recordsToAverageCount),
+		NetTotalUp:     latestTotalUp,
+		NetTotalDown:   latestTotalDown,
 		Process:        sumPROCESS / recordsToAverageCount,
 		Connections:    sumConnections / recordsToAverageCount,
 		ConnectionsUdp: sumConnectionsUDP / recordsToAverageCount,
@@ -129,6 +131,30 @@ func AverageReport(uuid string, time time.Time, records []v1.Report, topPercenta
 	return newRecord
 }
 
+func latestNetworkTotals(records []v1.Report) (int64, int64) {
+	if len(records) == 0 {
+		return 0, 0
+	}
+	latestIndex := 0
+	for i := 1; i < len(records); i++ {
+		if records[i].UpdatedAt.After(records[latestIndex].UpdatedAt) ||
+			records[i].UpdatedAt.Equal(records[latestIndex].UpdatedAt) {
+			latestIndex = i
+		}
+	}
+	return records[latestIndex].Network.TotalUp, records[latestIndex].Network.TotalDown
+}
+
+func ComputeTrafficDelta(current, previous int64) int64 {
+	if current < 0 || previous < 0 {
+		return 0
+	}
+	if current >= previous {
+		return current - previous
+	}
+	return current
+}
+
 // AverageGPUReports 使用与 AverageReport 相同的聚合逻辑处理GPU数据
 // 返回每个GPU设备的聚合记录
 func AverageGPUReports(uuid string, time time.Time, reports []v1.Report, topPercentage float64) []models.GPURecord {

utils/utils_test.go

@@ -0,0 +1,48 @@
+package utils
+
+import (
+	"testing"
+	"time"
+
+	"github.com/komari-monitor/komari/protocol/v1"
+	"github.com/stretchr/testify/assert"
+)
+
+func TestAverageReportUsesLatestNetworkTotals(t *testing.T) {
+	base := time.Date(2026, 6, 13, 12, 0, 0, 0, time.UTC)
+	record := AverageReport("client-latest-total", base.Add(time.Minute), []v1.Report{
+		{
+			UpdatedAt: base.Add(10 * time.Second),
+			Network:   v1.NetworkReport{TotalUp: 300, TotalDown: 500},
+		},
+		{
+			UpdatedAt: base.Add(50 * time.Second),
+			Network:   v1.NetworkReport{TotalUp: 360, TotalDown: 620},
+		},
+	}, 0)
+
+	assert.Equal(t, int64(360), record.NetTotalUp)
+	assert.Equal(t, int64(620), record.NetTotalDown)
+}
+
+func TestComputeTrafficDelta(t *testing.T) {
+	tests := []struct {
+		name     string
+		current  int64
+		previous int64
+		want     int64
+	}{
+		{name: "previous zero counts current delta", current: 120, previous: 0, want: 120},
+		{name: "monotonic counter uses difference", current: 250, previous: 200, want: 50},
+		{name: "same counter", current: 100, previous: 100, want: 0},
+		{name: "counter reset uses current", current: 15, previous: 250, want: 15},
+		{name: "negative current remains guarded", current: -1, previous: 100, want: 0},
+		{name: "negative previous remains guarded", current: 15, previous: -1, want: 0},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			assert.Equal(t, test.want, ComputeTrafficDelta(test.current, test.previous))
+		})
+	}
+}

web/agent/v2_events.go

@@ -39,7 +39,9 @@ func getV2EventQueueLocked(uuid string) *v2EventQueue {
 func DispatchV2Event(uuid, method string, params any) bool {
 	if conn := GetConnectedClients()[uuid]; conn != nil {
 		payload := v2.Request{JSONRPC: v2.Version, Method: method, Params: params}
-		return conn.WriteJSON(payload) == nil
+		if conn.WriteJSON(payload) == nil {
+			return true
+		}
 	}
 	if !IsV2Client(uuid) {
 		return false
@@ -54,7 +56,9 @@ func DispatchPing(uuid string, legacy any, params v2.PingParams) bool {
 		if IsV2Client(uuid) {
 			payload = v2.Request{JSONRPC: v2.Version, Method: v2.MethodAgentPing, Params: params}
 		}
-		return conn.WriteJSON(payload) == nil
+		if conn.WriteJSON(payload) == nil {
+			return true
+		}
 	}
 	if !IsV2Client(uuid) {
 		return false
@@ -183,6 +187,24 @@ func TakeV2Events(uuid string, ackIDs []string, limit int) []v2.Event {
 	q := getV2EventQueueLocked(uuid)
 	ackV2EventsLocked(q, ackIDs)
 	pruneExpiredV2EventsLocked(q)
+	return takeV2EventsLocked(q, limit)
+}
+
+func AckV2Events(uuid string, ackIDs []string) {
+	if len(ackIDs) == 0 {
+		return
+	}
+	v2EventMu.Lock()
+	defer v2EventMu.Unlock()
+
+	q := v2EventQueues[uuid]
+	if q == nil {
+		return
+	}
+	ackV2EventsLocked(q, ackIDs)
+}
+
+func takeV2EventsLocked(q *v2EventQueue, limit int) []v2.Event {
 	if limit <= 0 || limit > len(q.events) {
 		limit = len(q.events)
 	}
@@ -192,13 +214,15 @@ func TakeV2Events(uuid string, ackIDs []string, limit int) []v2.Event {
 }
 
 func WaitV2Events(uuid string, ackIDs []string, timeout time.Duration) []v2.Event {
-	events := TakeV2Events(uuid, ackIDs, v2EventQueueLimit)
-	if len(events) > 0 || timeout <= 0 {
-		return events
-	}
-
 	v2EventMu.Lock()
 	q := getV2EventQueueLocked(uuid)
+	ackV2EventsLocked(q, ackIDs)
+	pruneExpiredV2EventsLocked(q)
+	events := takeV2EventsLocked(q, v2EventQueueLimit)
+	if len(events) > 0 || timeout <= 0 {
+		v2EventMu.Unlock()
+		return events
+	}
 	signal := q.signal
 	v2EventMu.Unlock()
 

web/api/Auth.go

@@ -13,6 +13,7 @@ import (
 	"github.com/komari-monitor/komari/database/accounts"
 	"github.com/komari-monitor/komari/database/clients"
 	"github.com/komari-monitor/komari/pkg/config"
+	"github.com/komari-monitor/komari/pkg/rpc"
 	"gorm.io/gorm"
 )
 
@@ -24,46 +25,31 @@ const (
 
 // IdentityMiddleware 统一身份识别中间件，在路由栈最外层运行。
 // 负责识别当前请求者身份（Admin / Client / Guest），并写入 Context。
+// 身份识别统一委托给 IdentifyPrincipal;同时保留旧的 c.Set 键(role/uuid/
+// api_key/session/client_uuid)以兼容现有 handler 与中间件。
 func IdentityMiddleware() gin.HandlerFunc {
 	return func(c *gin.Context) {
-		// 1. API Key 认证
-		apiKey := c.GetHeader("Authorization")
-		if isApiKeyValid(apiKey) {
-			c.Set("role", RoleAdmin)
-			c.Set("api_key", apiKey[7:])
-			c.Set("uuid", "00000000-0000-0000-0000-000000000000") // API Key
-			c.Next()
-			return
-		}
+		p := IdentifyPrincipal(c)
+		SetPrincipal(c, p)
 
-		// 2. Session 认证
-		session, err := c.Cookie("session_token")
-		if err == nil && session != "" {
-			uuid, err := accounts.GetSession(session)
-			if err == nil {
-				c.Set("role", RoleAdmin)
+		// 写入兼容字段。
+		c.Set("role", p.PrimaryRole())
+		switch p.Type {
+		case rpc.PrincipalAPIKey:
+			// 旧逻辑:API Key 时记录裸 key 与固定占位 uuid。
+			apiKey := c.GetHeader("Authorization")
+			c.Set("api_key", apiKey[len("Bearer "):])
+			c.Set("uuid", "00000000-0000-0000-0000-000000000000")
+		case rpc.PrincipalUser:
+			if session, err := c.Cookie("session_token"); err == nil && session != "" {
 				c.Set("session", session)
-				c.Set("uuid", uuid)
 				accounts.UpdateLatest(session, c.Request.UserAgent(), c.ClientIP())
-				c.Next()
-				return
 			}
+			c.Set("uuid", p.UserUUID)
+		case rpc.PrincipalAgent:
+			c.Set("client_uuid", p.ClientUUID)
 		}
 
-		// 3. Client Token 认证
-		token := extractClientToken(c)
-		if token != "" {
-			uuid, err := checkTokenAndGetUUID(token)
-			if err == nil && uuid != "" {
-				c.Set("role", RoleClient)
-				c.Set("client_uuid", uuid)
-				c.Next()
-				return
-			}
-		}
-
-		// 4. 未识别身份，标记为访客
-		c.Set("role", RoleGuest)
 		c.Next()
 	}
 }
@@ -183,6 +169,10 @@ func extractClientToken(c *gin.Context) string {
 	if token != "" {
 		return token
 	}
+	// rpc2 约定:agent 经 ?Authorization=<token> 传入 client token。
+	if token := c.Query("Authorization"); token != "" {
+		return token
+	}
 
 	if c.Request.Method != http.MethodGet {
 		bodyBytes, err := io.ReadAll(c.Request.Body)

web/api/AuthSensitive.go

@@ -17,34 +17,33 @@ func RequireSensitive2FA() gin.HandlerFunc {
 			c.Abort()
 			return
 		}
+		// 标记本请求已通过敏感操作校验，避免下游（RPC 边界）重复校验。
+		c.Set("sensitive_2fa_verified", true)
 		c.Next()
 	}
 }
 
-func VerifySensitive2FA(c *gin.Context) error {
-	if _, ok := c.Get("api_key"); ok {
+// VerifySensitive2FACore 传输无关的 2FA 校验核心。
+// 输入原始值:userUUID、2FA code、是否为 API Key。
+// API Key 豁免;未启用 2FA 的用户放行;其余需要有效 code。
+func VerifySensitive2FACore(userUUID, code string, isAPIKey bool) error {
+	if isAPIKey {
 		return nil
 	}
-	uuid, ok := c.Get("uuid")
-	if !ok {
+	if userUUID == "" {
 		return err2FARequired()
 	}
-	uuidString, ok := uuid.(string)
-	if !ok || uuidString == "" {
-		return err2FARequired()
-	}
-	user, err := accounts.GetUserByUUID(uuidString)
+	user, err := accounts.GetUserByUUID(userUUID)
 	if err != nil {
 		return err
 	}
 	if user.TwoFactor == "" {
 		return nil
 	}
-	code := get2FACode(c)
 	if code == "" {
 		return err2FARequired()
 	}
-	valid, err := accounts.Verify2Fa(uuidString, code)
+	valid, err := accounts.Verify2Fa(userUUID, code)
 	if err != nil {
 		return err
 	}
@@ -54,6 +53,14 @@ func VerifySensitive2FA(c *gin.Context) error {
 	return nil
 }
 
+// VerifySensitive2FA gin 适配层:从 gin.Context 提取参数后委托核心校验。
+func VerifySensitive2FA(c *gin.Context) error {
+	_, isAPIKey := c.Get("api_key")
+	uuidRaw, _ := c.Get("uuid")
+	uuid, _ := uuidRaw.(string)
+	return VerifySensitive2FACore(uuid, get2FACode(c), isAPIKey)
+}
+
 func get2FACode(c *gin.Context) string {
 	if code, ok := c.Get("2fa_code"); ok {
 		if codeString, ok := code.(string); ok && codeString != "" {

web/api/admin/client.go

@@ -1,137 +0,0 @@
-package admin
-
-import (
-	"net/http"
-
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/auditlog"
-	"github.com/komari-monitor/komari/database/clients"
-	"github.com/komari-monitor/komari/database/records"
-	agent_runtime "github.com/komari-monitor/komari/web/agent"
-)
-
-func AddClient(c *gin.Context) {
-	var req struct {
-		Name string `json:"name"`
-	}
-	if err := c.ShouldBindJSON(&req); err != nil || req.Name == "" {
-		uuid, token, err := clients.CreateClient()
-		if err != nil {
-			c.JSON(http.StatusInternalServerError, gin.H{"status": "error", "message": err.Error()})
-			return
-		}
-		c.JSON(http.StatusOK, gin.H{"status": "success", "uuid": uuid, "token": token})
-		return
-	}
-	uuid, token, err := clients.CreateClientWithName(req.Name)
-	if err != nil {
-		c.JSON(http.StatusInternalServerError, gin.H{"status": "error", "message": err.Error()})
-		return
-	}
-	user_uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), user_uuid.(string), "create client:"+uuid, "info")
-	c.JSON(http.StatusOK, gin.H{"status": "success", "uuid": uuid, "token": token, "message": ""})
-}
-
-func EditClient(c *gin.Context) {
-	var req = make(map[string]interface{})
-	uuid := c.Param("uuid")
-	if err := c.ShouldBindJSON(&req); err != nil {
-		c.JSON(http.StatusBadRequest, gin.H{"status": "error", "message": err.Error()})
-		return
-	}
-	if uuid == "" {
-		c.JSON(http.StatusBadRequest, gin.H{"status": "error", "message": "Invalid or missing UUID"})
-		return
-	}
-	req["uuid"] = uuid
-	err := clients.SaveClient(req)
-	if err != nil {
-		c.JSON(http.StatusInternalServerError, gin.H{"status": "error", "message": err.Error()})
-		return
-	}
-	user_uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), user_uuid.(string), "edit client:"+uuid, "info")
-	c.JSON(http.StatusOK, gin.H{"status": "success"})
-}
-
-func RemoveClient(c *gin.Context) {
-	uuid := c.Param("uuid")
-	err := clients.DeleteClient(uuid)
-	if err != nil {
-		c.JSON(500, gin.H{
-			"status": "error",
-			"error":  "Failed to delete client" + err.Error(),
-		})
-		return
-	}
-	user_uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), user_uuid.(string), "delete client:"+uuid, "warn")
-	c.JSON(200, gin.H{"status": "success"})
-	agent_runtime.DeleteConnectedClients(uuid)
-	agent_runtime.DeleteLatestReport(uuid)
-}
-
-func ClearRecord(c *gin.Context) {
-	if err := records.DeleteAll(); err != nil {
-		c.JSON(500, gin.H{
-			"status":  "error",
-			"message": "Failed to delete Record" + err.Error(),
-		})
-		return
-	}
-	user_uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), user_uuid.(string), "clear records", "warn")
-	c.JSON(200, gin.H{"status": "success"})
-}
-
-func GetClient(c *gin.Context) {
-	uuid := c.Param("uuid")
-	if uuid == "" {
-		c.JSON(400, gin.H{
-			"status":  "error",
-			"message": "Invalid or missing UUID",
-		})
-		return
-	}
-
-	result, err := clients.GetClientByUUID(uuid)
-	if err != nil {
-		c.JSON(http.StatusInternalServerError, gin.H{
-			"status":  "error",
-			"message": err.Error(),
-		})
-		return
-	}
-
-	c.JSON(http.StatusOK, result)
-}
-
-func ListClients(c *gin.Context) {
-	cls, err := clients.GetAllClientBasicInfo()
-	if err != nil {
-		c.JSON(http.StatusInternalServerError, gin.H{"status": "error", "message": err.Error()})
-		return
-	}
-
-	c.JSON(http.StatusOK, cls)
-}
-
-func GetClientToken(c *gin.Context) {
-	uuid := c.Param("uuid")
-	if uuid == "" {
-		c.JSON(400, gin.H{
-			"status":  "error",
-			"message": "Invalid or missing UUID",
-		})
-		return
-	}
-
-	token, err := clients.GetClientTokenByUUID(uuid)
-	if err != nil {
-		c.JSON(http.StatusInternalServerError, gin.H{"status": "error", "message": err.Error()})
-		return
-	}
-
-	c.JSON(http.StatusOK, gin.H{"status": "success", "token": token, "message:": ""})
-}

web/api/admin/client_weight.go

@@ -1,28 +0,0 @@
-package admin
-
-import (
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/auditlog"
-	"github.com/komari-monitor/komari/database/dbcore"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-func OrderWeight(c *gin.Context) {
-	var req = make(map[string]int)
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, 400, "Invalid or missing request body: "+err.Error())
-		return
-	}
-	db := dbcore.GetDBInstance()
-	for uuid, weight := range req {
-		err := db.Model(&models.Client{}).Where("uuid = ?", uuid).Update("weight", weight).Error
-		if err != nil {
-			api.RespondError(c, 500, "Failed to update client weight: "+err.Error())
-			return
-		}
-	}
-	uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), uuid.(string), "order clients", "info")
-	api.RespondSuccess(c, nil)
-}

web/api/admin/clipboard/clipboard.go

@@ -1,115 +0,0 @@
-package clipboard
-
-import (
-	"net/http"
-	"strconv"
-
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/auditlog"
-	clipboardDB "github.com/komari-monitor/komari/database/clipboard"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-// GetClipboard retrieves a clipboard entry by ID
-func GetClipboard(c *gin.Context) {
-	idStr := c.Param("id")
-	id, err := strconv.Atoi(idStr)
-	if err != nil {
-		api.RespondError(c, http.StatusBadRequest, "Invalid ID")
-		return
-	}
-	cb, err := clipboardDB.GetClipboardByID(id)
-	if err != nil {
-		api.RespondError(c, http.StatusInternalServerError, "Failed to get clipboard: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, cb)
-}
-
-// ListClipboard lists all clipboard entries
-func ListClipboard(c *gin.Context) {
-	list, err := clipboardDB.ListClipboard()
-	if err != nil {
-		api.RespondError(c, http.StatusInternalServerError, "Failed to list clipboard: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, list)
-}
-
-// CreateClipboard creates a new clipboard entry
-func CreateClipboard(c *gin.Context) {
-	var req models.Clipboard
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, http.StatusBadRequest, "Invalid request: "+err.Error())
-		return
-	}
-	if err := clipboardDB.CreateClipboard(&req); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, "Failed to create clipboard: "+err.Error())
-		return
-	}
-	userUUID, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), userUUID.(string), "create clipboard:"+strconv.Itoa(req.Id), "info")
-	api.RespondSuccess(c, req)
-}
-
-// UpdateClipboard updates an existing clipboard entry
-func UpdateClipboard(c *gin.Context) {
-	idStr := c.Param("id")
-	id, err := strconv.Atoi(idStr)
-	if err != nil {
-		api.RespondError(c, http.StatusBadRequest, "Invalid ID")
-		return
-	}
-	var fields map[string]interface{}
-	if err := c.ShouldBindJSON(&fields); err != nil {
-		api.RespondError(c, http.StatusBadRequest, "Invalid request: "+err.Error())
-		return
-	}
-	if err := clipboardDB.UpdateClipboardFields(id, fields); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, "Failed to update clipboard: "+err.Error())
-		return
-	}
-	userUUID, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), userUUID.(string), "update clipboard:"+strconv.Itoa(id), "info")
-	api.RespondSuccess(c, nil)
-}
-
-// DeleteClipboard deletes a clipboard entry
-func DeleteClipboard(c *gin.Context) {
-	idStr := c.Param("id")
-	id, err := strconv.Atoi(idStr)
-	if err != nil {
-		api.RespondError(c, http.StatusBadRequest, "Invalid ID")
-		return
-	}
-	if err := clipboardDB.DeleteClipboard(id); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, "Failed to delete clipboard: "+err.Error())
-		return
-	}
-	userUUID, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), userUUID.(string), "delete clipboard:"+strconv.Itoa(id), "warn")
-	api.RespondSuccess(c, nil)
-}
-
-// BatchDeleteClipboard deletes multiple clipboard entries
-func BatchDeleteClipboard(c *gin.Context) {
-	var req struct {
-		IDs []int `json:"ids" binding:"required"`
-	}
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, http.StatusBadRequest, "Invalid request: "+err.Error())
-		return
-	}
-	if len(req.IDs) == 0 {
-		api.RespondError(c, http.StatusBadRequest, "IDs cannot be empty")
-		return
-	}
-	if err := clipboardDB.DeleteClipboardBatch(req.IDs); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, "Failed to batch delete clipboard: "+err.Error())
-		return
-	}
-	userUUID, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), userUUID.(string), "batch delete clipboard: "+strconv.Itoa(len(req.IDs))+" items", "warn")
-	api.RespondSuccess(c, nil)
-}

web/api/admin/cloudflared.go

@@ -1,104 +0,0 @@
-package admin
-
-import (
-	"net/http"
-	"strings"
-
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/accounts"
-	"github.com/komari-monitor/komari/database/auditlog"
-	"github.com/komari-monitor/komari/pkg/config"
-	"github.com/komari-monitor/komari/utils/cloudflared"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-const cloudflaredStopConfirmText = "STOP CLOUDFLARED"
-
-func GetCloudflaredStatus(c *gin.Context) {
-	api.RespondSuccess(c, cloudflared.Status())
-}
-
-func StartCloudflared(c *gin.Context) {
-	var req struct {
-		Token string `json:"token"`
-	}
-
-	if err := c.ShouldBindJSON(&req); err != nil && err.Error() != "EOF" {
-		api.RespondError(c, http.StatusBadRequest, "Invalid request body: "+err.Error())
-		return
-	}
-
-	token := strings.TrimSpace(req.Token)
-	if token != "" {
-		if err := cloudflared.SaveToken(token); err != nil {
-			api.RespondError(c, http.StatusInternalServerError, "Failed to save Cloudflare Tunnel token: "+err.Error())
-			return
-		}
-	}
-	if err := cloudflared.Start(token); err != nil {
-		api.RespondError(c, http.StatusBadRequest, err.Error())
-		return
-	}
-	if uuid, ok := c.Get("uuid"); ok {
-		auditlog.Log(c.ClientIP(), uuid.(string), "started cloudflared tunnel", "warn")
-	}
-	api.RespondSuccess(c, cloudflared.Status())
-}
-
-func StopCloudflared(c *gin.Context) {
-	var req struct {
-		CurrentPassword string `json:"current_password"`
-		ConfirmText     string `json:"confirm_text"`
-	}
-
-	if err := c.ShouldBindJSON(&req); err != nil && err.Error() != "EOF" {
-		api.RespondError(c, http.StatusBadRequest, "Invalid request body: "+err.Error())
-		return
-	}
-
-	disablePasswordLogin, _ := config.GetAs[bool](config.DisablePasswordLoginKey, false)
-	if !disablePasswordLogin {
-		rawUUID, exists := c.Get("uuid")
-		if !exists {
-			api.RespondError(c, http.StatusUnauthorized, "Unauthorized.")
-			return
-		}
-
-		user, err := accounts.GetUserByUUID(rawUUID.(string))
-		if err != nil {
-			api.RespondError(c, http.StatusUnauthorized, "Failed to verify current user")
-			return
-		}
-		if strings.TrimSpace(req.CurrentPassword) == "" {
-			api.RespondError(c, http.StatusBadRequest, "Current password is required")
-			return
-		}
-		if _, ok := accounts.CheckPassword(user.Username, req.CurrentPassword); !ok {
-			api.RespondError(c, http.StatusUnauthorized, "Current password is incorrect")
-			return
-		}
-	} else if strings.TrimSpace(req.ConfirmText) != cloudflaredStopConfirmText {
-		api.RespondError(c, http.StatusBadRequest, "Type STOP CLOUDFLARED to confirm stopping cloudflared")
-		return
-	}
-
-	if err := cloudflared.Stop(); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, "Failed to stop cloudflared: "+err.Error())
-		return
-	}
-	if uuid, ok := c.Get("uuid"); ok {
-		auditlog.Log(c.ClientIP(), uuid.(string), "stopped cloudflared tunnel", "warn")
-	}
-	api.RespondSuccess(c, cloudflared.Status())
-}
-
-func RemoveCloudflaredToken(c *gin.Context) {
-	if err := cloudflared.RemoveToken(); err != nil {
-		api.RespondError(c, http.StatusBadRequest, "Failed to remove Cloudflare Tunnel token: "+err.Error())
-		return
-	}
-	if uuid, ok := c.Get("uuid"); ok {
-		auditlog.Log(c.ClientIP(), uuid.(string), "removed cloudflared tunnel token", "warn")
-	}
-	api.RespondSuccess(c, cloudflared.Status())
-}

web/api/admin/exec.go

@@ -1,112 +0,0 @@
-package admin
-
-import (
-	"encoding/json"
-	"strings"
-	"time"
-
-	"github.com/gin-gonic/gin"
-	"github.com/gorilla/websocket"
-	"github.com/komari-monitor/komari/database/auditlog"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/database/tasks"
-	v2 "github.com/komari-monitor/komari/protocol/v2"
-	"github.com/komari-monitor/komari/utils"
-	agent_runtime "github.com/komari-monitor/komari/web/agent"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-// 接受数据类型：
-// - command: string
-// - clients: []string (客户端 UUID 列表)
-func Exec(c *gin.Context) {
-	var req struct {
-		Command string   `json:"command" binding:"required"`
-		Clients []string `json:"clients" binding:"required"`
-	}
-	var onlineClients []string
-	var queuedClients []string
-	var offlineClients []string
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, 400, "Invalid or missing request body: "+err.Error())
-		return
-	}
-	if strings.TrimSpace(req.Command) == "" {
-		api.RespondError(c, 400, "Command cannot be empty")
-		return
-	}
-	// for uuid := range agent_runtime.GetConnectedClients() {
-	// 	if contain(req.Clients, uuid) {
-	// 		onlineClients = append(onlineClients, uuid)
-	// 	}
-	// 	// else {
-	// 	// 	api.RespondError(c, 400, "Client not connected: "+uuid)
-	// 	// 	return
-	// 	// }
-	// }
-	for _, uuid := range req.Clients {
-		if client := agent_runtime.GetConnectedClients()[uuid]; client != nil {
-			onlineClients = append(onlineClients, uuid)
-		} else if agent_runtime.IsAgentOnline(uuid) {
-			queuedClients = append(queuedClients, uuid)
-		} else {
-			offlineClients = append(offlineClients, uuid)
-		}
-	}
-	if len(onlineClients) == 0 && len(queuedClients) == 0 {
-		api.RespondError(c, 400, "No clients connected")
-		return
-	}
-	taskId := utils.GenerateRandomString(16)
-	taskClients := append(append([]string{}, onlineClients...), queuedClients...)
-	taskClients = append(taskClients, offlineClients...)
-	if err := tasks.CreateTask(taskId, taskClients, req.Command); err != nil {
-		api.RespondError(c, 500, "Failed to create task: "+err.Error())
-		return
-	}
-	for _, uuid := range onlineClients {
-		legacy := struct {
-			Message string `json:"message"`
-			Command string `json:"command"`
-			TaskId  string `json:"task_id"`
-		}{Message: "exec", Command: req.Command, TaskId: taskId}
-		payload, _ := json.Marshal(legacy)
-		if agent_runtime.IsV2Client(uuid) {
-			payload, _ = json.Marshal(v2.Request{JSONRPC: v2.Version, Method: v2.MethodAgentExec, Params: v2.ExecParams{TaskID: taskId, Command: req.Command}})
-		}
-		client := agent_runtime.GetConnectedClients()[uuid]
-		if client != nil {
-			if err := client.WriteMessage(websocket.TextMessage, payload); err != nil {
-				api.RespondError(c, 400, "Client connection is broke: "+uuid)
-				return
-			}
-		} else {
-			api.RespondError(c, 400, "Client connection is null: "+uuid)
-			return
-		}
-	}
-	for _, uuid := range queuedClients {
-		agent_runtime.DispatchV2Event(uuid, v2.MethodAgentExec, v2.ExecParams{TaskID: taskId, Command: req.Command})
-	}
-	uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), uuid.(string), "REC, task id: "+taskId, "warn")
-	api.RespondSuccess(c, gin.H{
-		"task_id":        taskId,
-		"clients":        onlineClients,
-		"queued_clients": queuedClients,
-	})
-	if len(offlineClients) > 0 {
-		for _, uuid := range offlineClients {
-			tasks.SaveTaskResult(taskId, uuid, "Client offline!", -1, models.FromTime(time.Now()))
-		}
-	}
-}
-
-// func contain(clients []string, uuid string) bool {
-// 	for _, client := range clients {
-// 		if client == uuid {
-// 			return true
-// 		}
-// 	}
-// 	return false
-// }

web/api/admin/log/log.go

@@ -1,49 +0,0 @@
-package log
-
-import (
-	"strconv"
-
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/dbcore"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-func GetLogs(c *gin.Context) {
-	limit := c.Query("limit")
-	if limit == "" {
-		limit = "100" // Default to 100 logs if not specified
-	}
-	page := c.Query("page")
-	if page == "" {
-		page = "1" // Default to page 1 if not specified
-	}
-	// Convert limit and page to integers
-	// If conversion fails, return an error
-	limitInt, err := strconv.Atoi(limit)
-	if err != nil || limitInt <= 0 {
-		api.RespondError(c, 400, "Invalid limit: "+limit)
-		return
-	}
-	pageInt, err := strconv.Atoi(page)
-	if err != nil || pageInt <= 0 {
-		api.RespondError(c, 400, "Invalid page: "+page)
-		return
-	}
-	db := dbcore.GetDBInstance()
-	var logs []models.Log
-	// 添加分页：计算偏移量并限制数量
-	offset := (pageInt - 1) * limitInt
-
-	var total int64
-	if err := db.Model(&models.Log{}).Count(&total).Error; err != nil {
-		api.RespondError(c, 500, "Failed to count logs: "+err.Error())
-		return
-	}
-
-	if err := db.Order("time desc").Limit(limitInt).Offset(offset).Find(&logs).Error; err != nil {
-		api.RespondError(c, 500, "Failed to retrieve logs: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, gin.H{"logs": logs, "total": total})
-}

web/api/admin/messageSender.go

@@ -1,64 +0,0 @@
-package admin
-
-import (
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/pkg/config"
-	"github.com/komari-monitor/komari/utils/messageSender"
-	"github.com/komari-monitor/komari/utils/messageSender/factory"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-func GetMessageSenderProvider(c *gin.Context) {
-	provider := c.Query("provider")
-	if provider != "" {
-		// 如果指定了provider，返回单个提供者的配置
-		config, err := database.GetMessageSenderConfigByName(provider)
-		if err != nil {
-			api.RespondError(c, 404, "Provider not found: "+err.Error())
-			return
-		}
-		api.RespondSuccess(c, config)
-		return
-	}
-	// 否则返回所有提供者的配置项模板
-	providers := factory.GetSenderConfigs()
-	if len(providers) == 0 {
-		api.RespondError(c, 404, "No message sender providers found")
-		return
-	}
-	api.RespondSuccess(c, providers)
-}
-
-func SetMessageSenderProvider(c *gin.Context) {
-	var senderConfig models.MessageSenderProvider
-	if err := c.ShouldBindJSON(&senderConfig); err != nil {
-		api.RespondError(c, 400, "Invalid configuration: "+err.Error())
-		return
-	}
-	if senderConfig.Name == "" {
-		api.RespondError(c, 400, "Provider name is required")
-		return
-	}
-	_, exists := factory.GetConstructor(senderConfig.Name)
-	if !exists {
-		api.RespondError(c, 404, "Provider not found: "+senderConfig.Name)
-		return
-	}
-	if err := database.SaveMessageSenderConfig(&senderConfig); err != nil {
-		api.RespondError(c, 500, "Failed to save message sender provider configuration: "+err.Error())
-		return
-	}
-	NotificationMethod, _ := config.GetAs[string](config.NotificationMethodKey, "none")
-
-	// 正在使用，重载
-	if NotificationMethod == senderConfig.Name {
-		err := messageSender.LoadProvider(senderConfig.Name, senderConfig.Addition)
-		if err != nil {
-			api.RespondError(c, 500, "Failed to load message sender provider: "+err.Error())
-			return
-		}
-	}
-	api.RespondSuccess(c, gin.H{"message": "Message sender provider set successfully"})
-}

web/api/admin/notification/load.go

@@ -1,90 +0,0 @@
-package notification
-
-import (
-	"net/http"
-
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/database/notification"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-// POST body: clients []string, name string, metric string, threshold float32, ratio float32, interval int
-func AddLoadNotification(c *gin.Context) {
-	var req struct {
-		Clients   []string `json:"clients" binding:"required"`
-		Name      string   `json:"name"`
-		Metric    string   `json:"metric" binding:"required"`
-		Threshold float32  `json:"threshold" binding:"required"` // 阈值百分比
-		Ratio     float32  `json:"ratio" binding:"required"`     // 达标时间比
-		Interval  int      `json:"interval" binding:"required"`  // 间隔时间，单位秒
-	}
-
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, http.StatusBadRequest, err.Error())
-		return
-	}
-
-	if req.Interval > 4*60 || req.Interval <= 0 {
-		api.RespondError(c, http.StatusBadRequest, "Interval must be between 1 and 240 minutes")
-		return
-	}
-	if req.Ratio <= 0 || req.Ratio > 1 {
-		api.RespondError(c, http.StatusBadRequest, "Ratio must be between 0 and 1")
-		return
-	}
-
-	if taskID, err := notification.AddLoadNotification(req.Clients, req.Name, req.Metric, req.Threshold, req.Ratio, req.Interval); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, err.Error())
-	} else {
-		api.RespondSuccess(c, gin.H{"task_id": taskID})
-	}
-}
-
-// POST body: id []uint
-func DeleteLoadNotification(c *gin.Context) {
-	var req struct {
-		ID []uint `json:"id" binding:"required"`
-	}
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, http.StatusBadRequest, err.Error())
-		return
-	}
-
-	if err := notification.DeleteLoadNotification(req.ID); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, err.Error())
-	} else {
-		api.RespondSuccess(c, nil)
-	}
-}
-
-// POST body: id []uint, updates map[string]interface{}
-func EditLoadNotification(c *gin.Context) {
-	var req struct {
-		Notifications []*models.LoadNotification `json:"notifications" binding:"required"`
-	}
-
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, http.StatusBadRequest, "Invalid request data")
-		return
-	}
-
-	if err := notification.EditLoadNotification(req.Notifications); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, err.Error())
-	} else {
-		// for _, notification := range req.Notifications {
-		// 	notification.DeleteLoadNotification([]uint{notification.Id})
-		// }
-		api.RespondSuccess(c, nil)
-	}
-}
-
-func GetAllLoadNotifications(c *gin.Context) {
-	notifications, err := notification.GetAllLoadNotifications()
-	if err != nil {
-		api.RespondError(c, http.StatusInternalServerError, err.Error())
-		return
-	}
-
-	api.RespondSuccess(c, notifications)
-}

web/api/admin/notification/offline.go

@@ -1,109 +0,0 @@
-package notification
-
-import (
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/dbcore"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/web/api"
-	"gorm.io/gorm/clause"
-)
-
-// POST body : []uuid
-func EnableOfflineNotification(c *gin.Context) {
-	var uuids []string
-	if err := c.ShouldBindJSON(&uuids); err != nil {
-		api.RespondError(c, 400, "Invalid request body: "+err.Error())
-		return
-	}
-	var notifications []models.OfflineNotification
-	for _, uuid := range uuids {
-		notifications = append(notifications, models.OfflineNotification{
-			Client: uuid,
-			Enable: true,
-		})
-	}
-	err := dbcore.GetDBInstance().Model(&models.OfflineNotification{}).
-		Clauses(clause.OnConflict{
-			Columns:   []clause.Column{{Name: "client"}},
-			DoUpdates: clause.AssignmentColumns([]string{"enable"}),
-		}).
-		Select("client", "enable").
-		Create(notifications).Error
-	if err != nil {
-		api.RespondError(c, 500, "Failed to enable offline notifications: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, nil)
-}
-
-// POST body : []uuid
-func DisableOfflineNotification(c *gin.Context) {
-	var uuids []string
-	if err := c.ShouldBindJSON(&uuids); err != nil {
-		api.RespondError(c, 400, "Invalid request body: "+err.Error())
-		return
-	}
-	var notifications []models.OfflineNotification
-	for _, uuid := range uuids {
-		notifications = append(notifications, models.OfflineNotification{
-			Client: uuid,
-			Enable: false,
-		})
-	}
-	err := dbcore.GetDBInstance().Model(&models.OfflineNotification{}).
-		Clauses(clause.OnConflict{
-			Columns:   []clause.Column{{Name: "client"}},
-			DoUpdates: clause.AssignmentColumns([]string{"enable"}),
-		}).
-		Select("client", "enable").
-		Create(notifications).Error
-	if err != nil {
-		api.RespondError(c, 500, "Failed to disable offline notifications: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, nil)
-}
-
-func EditOfflineNotification(c *gin.Context) {
-	var notifications []models.OfflineNotification
-	if err := c.ShouldBindJSON(&notifications); err != nil {
-		api.RespondError(c, 400, "Invalid request body: "+err.Error())
-		return
-	}
-	if len(notifications) == 0 {
-		api.RespondError(c, 400, "At least one notification is required")
-		return
-	}
-	for _, noti := range notifications {
-		if noti.Client == "" {
-			api.RespondError(c, 400, "Client UUID cannot be empty")
-			return
-		}
-		if noti.GracePeriod <= 0 {
-			api.RespondError(c, 400, "GracePeriod must be a positive integer")
-			return
-		}
-	}
-	err := dbcore.GetDBInstance().Model(&models.OfflineNotification{}).
-		Clauses(clause.OnConflict{
-			Columns:   []clause.Column{{Name: "client"}},
-			DoUpdates: clause.AssignmentColumns([]string{"enable", "grace_period"}),
-		}).
-		Select("*").
-		Create(notifications).Error
-	if err != nil {
-		api.RespondError(c, 500, "Failed to edit offline notifications: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, nil)
-}
-
-func ListOfflineNotifications(c *gin.Context) {
-	var notifications []models.OfflineNotification
-	err := dbcore.GetDBInstance().Model(&models.OfflineNotification{}).Find(&notifications).Error
-	if err != nil {
-		api.RespondError(c, 500, "Failed to list offline notifications: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, notifications)
-}

web/api/admin/notification/traffic_report.go

@@ -1,71 +0,0 @@
-package notification
-
-import (
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/database/notification"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-// ListTrafficReportNotifications GET /api/admin/notification/traffic-report
-func ListTrafficReportNotifications(c *gin.Context) {
-	notifications, err := notification.ListTrafficReportNotifications()
-	if err != nil {
-		api.RespondError(c, 500, "Failed to list traffic report notifications: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, notifications)
-}
-
-// EditTrafficReportNotifications POST /api/admin/notification/traffic-report/edit
-// Body: []TrafficReportNotification
-func EditTrafficReportNotifications(c *gin.Context) {
-	var notifications []models.TrafficReportNotification
-	if err := c.ShouldBindJSON(&notifications); err != nil {
-		api.RespondError(c, 400, "Invalid request body: "+err.Error())
-		return
-	}
-	if len(notifications) == 0 {
-		api.RespondError(c, 400, "At least one notification is required")
-		return
-	}
-	if err := notification.ValidateTrafficReportNotifications(notifications); err != nil {
-		api.RespondError(c, 400, err.Error())
-		return
-	}
-	if err := notification.EditTrafficReportNotifications(notifications); err != nil {
-		api.RespondError(c, 500, "Failed to edit traffic report notifications: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, nil)
-}
-
-// EnableTrafficReportNotifications POST /api/admin/notification/traffic-report/enable
-// Body: []string (uuid list)
-func EnableTrafficReportNotifications(c *gin.Context) {
-	var uuids []string
-	if err := c.ShouldBindJSON(&uuids); err != nil {
-		api.RespondError(c, 400, "Invalid request body: "+err.Error())
-		return
-	}
-	if err := notification.EnableTrafficReportNotifications(uuids); err != nil {
-		api.RespondError(c, 500, "Failed to enable traffic report notifications: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, nil)
-}
-
-// DisableTrafficReportNotifications POST /api/admin/notification/traffic-report/disable
-// Body: []string (uuid list)
-func DisableTrafficReportNotifications(c *gin.Context) {
-	var uuids []string
-	if err := c.ShouldBindJSON(&uuids); err != nil {
-		api.RespondError(c, 400, "Invalid request body: "+err.Error())
-		return
-	}
-	if err := notification.DisableTrafficReportNotifications(uuids); err != nil {
-		api.RespondError(c, 500, "Failed to disable traffic report notifications: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, nil)
-}

web/api/admin/oauth.go

@@ -2,15 +2,14 @@ package admin
 
 import (
 	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database"
 	"github.com/komari-monitor/komari/database/accounts"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/pkg/config"
 	"github.com/komari-monitor/komari/web/api"
-	"github.com/komari-monitor/komari/web/oauth"
-	"github.com/komari-monitor/komari/web/oauth/factory"
 )
 
+// oauth.go
+// 外部账号绑定/解绑：绑定走 302 重定向，保留为 REST handler（不走 RPC 桥）。
+// OIDC provider 配置的 get/set 已迁移为 RPC2 方法（见 web/rpc/jsonrpc/admin.provider.go）。
+
 func BindingExternalAccount(c *gin.Context) {
 	session, _ := c.Cookie("session_token")
 	user, err := accounts.GetUserBySession(session)
@@ -18,10 +17,10 @@ func BindingExternalAccount(c *gin.Context) {
 		api.RespondError(c, 500, "No user found: "+err.Error())
 		return
 	}
-
 	c.SetCookie("binding_external_account", user.UUID, 3600, "/", "", false, true)
 	c.Redirect(302, "/api/oauth")
 }
+
 func UnbindExternalAccount(c *gin.Context) {
 	session, _ := c.Cookie("session_token")
 	user, err := accounts.GetUserBySession(session)
@@ -29,65 +28,9 @@ func UnbindExternalAccount(c *gin.Context) {
 		api.RespondError(c, 500, "No user found: "+err.Error())
 		return
 	}
-
-	err = accounts.UnbindExternalAccount(user.UUID)
-	if err != nil {
+	if err := accounts.UnbindExternalAccount(user.UUID); err != nil {
 		api.RespondError(c, 500, "Failed to unbind external account: "+err.Error())
 		return
 	}
-
 	api.RespondSuccess(c, nil)
 }
-
-func GetOidcProvider(c *gin.Context) {
-	provider := c.Query("provider")
-	if provider != "" {
-		// 如果指定了provider，返回单个提供者的配置
-		config, err := database.GetOidcConfigByName(provider)
-		if err != nil {
-			api.RespondError(c, 404, "Provider not found: "+err.Error())
-			return
-		}
-		api.RespondSuccess(c, config)
-		return
-	}
-	// 否则返回所有提供者的配置
-	providers := factory.GetProviderConfigs()
-	if len(providers) == 0 {
-		api.RespondError(c, 404, "No OIDC providers found")
-		return
-	}
-	api.RespondSuccess(c, providers)
-}
-
-func SetOidcProvider(c *gin.Context) {
-	var oidcConfig models.OidcProvider
-	if err := c.ShouldBindJSON(&oidcConfig); err != nil {
-		api.RespondError(c, 400, "Invalid configuration: "+err.Error())
-		return
-	}
-	if oidcConfig.Name == "" {
-		api.RespondError(c, 400, "Provider name is required")
-		return
-	}
-	_, exists := factory.GetConstructor(oidcConfig.Name)
-	if !exists {
-		api.RespondError(c, 404, "Provider not found: "+oidcConfig.Name)
-		return
-	}
-
-	if err := database.SaveOidcConfig(&oidcConfig); err != nil {
-		api.RespondError(c, 500, "Failed to save OIDC provider configuration: "+err.Error())
-		return
-	}
-	OAuthProvider, _ := config.GetAs[string](config.OAuthProviderKey, "github")
-	// 正在使用，重载
-	if OAuthProvider == oidcConfig.Name {
-		err := oauth.LoadProvider(oidcConfig.Name, oidcConfig.Addition)
-		if err != nil {
-			api.RespondError(c, 500, "Failed to load OIDC provider: "+err.Error())
-			return
-		}
-	}
-	api.RespondSuccess(c, gin.H{"message": "OIDC provider set successfully"})
-}

web/api/admin/ping.go

@@ -1,120 +0,0 @@
-package admin
-
-import (
-	"net/http"
-	"strconv"
-
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/database/tasks"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-// AddPingTask 处理新增延迟监测任务请求。
-// POST body: clients []string, default_on bool, target, task_type string, interval int
-func AddPingTask(c *gin.Context) {
-	var req struct {
-		Clients   []string `json:"clients"`
-		DefaultOn bool     `json:"default_on"`
-		Name      string   `json:"name" binding:"required"`
-		Target    string   `json:"target" binding:"required"`
-		TaskType  string   `json:"type" binding:"required"`     // icmp, tcp, http
-		Interval  int      `json:"interval" binding:"required"` // 间隔时间，单位秒
-	}
-
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, http.StatusBadRequest, err.Error())
-		return
-	}
-	if !req.DefaultOn && len(req.Clients) == 0 {
-		api.RespondError(c, http.StatusBadRequest, "clients is required when default_on is false")
-		return
-	}
-
-	if taskID, err := tasks.AddPingTask(req.Clients, req.DefaultOn, req.Name, req.Target, req.TaskType, req.Interval); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, err.Error())
-	} else {
-		api.RespondSuccess(c, gin.H{"task_id": taskID})
-	}
-}
-
-// POST body: id []uint
-func DeletePingTask(c *gin.Context) {
-	var req struct {
-		ID []uint `json:"id" binding:"required"`
-	}
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, http.StatusBadRequest, err.Error())
-		return
-	}
-
-	if err := tasks.DeletePingTask(req.ID); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, err.Error())
-	} else {
-		api.RespondSuccess(c, nil)
-	}
-}
-
-// EditPingTask 处理延迟监测任务编辑请求。
-// POST body: id []uint, updates map[string]interface{}
-func EditPingTask(c *gin.Context) {
-	var req struct {
-		Tasks []*models.PingTask `json:"tasks" binding:"required"`
-	}
-
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, http.StatusBadRequest, "Invalid request data")
-		return
-	}
-	for _, task := range req.Tasks {
-		// 编辑时只拦截空任务对象，允许把任务临时保存为未绑定服务器状态。
-		if task == nil {
-			api.RespondError(c, http.StatusBadRequest, "Invalid request data")
-			return
-		}
-	}
-
-	if err := tasks.EditPingTask(req.Tasks); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, err.Error())
-	} else {
-		// for _, task := range req.Tasks {
-		// 	tasks.DeletePingRecords([]uint{task.Id})
-		// }
-		api.RespondSuccess(c, nil)
-	}
-}
-
-func GetAllPingTasks(c *gin.Context) {
-	tasks, err := tasks.GetAllPingTasks()
-	if err != nil {
-		api.RespondError(c, http.StatusInternalServerError, err.Error())
-		return
-	}
-
-	api.RespondSuccess(c, tasks)
-}
-
-func OrderPingTask(c *gin.Context) {
-	var req map[string]int
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, http.StatusBadRequest, "Invalid or missing request body: "+err.Error())
-		return
-	}
-
-	order := make(map[uint]int, len(req))
-	for idStr, weight := range req {
-		id, err := strconv.ParseUint(idStr, 10, 64)
-		if err != nil {
-			api.RespondError(c, http.StatusBadRequest, "Invalid task id: "+idStr)
-			return
-		}
-		order[uint(id)] = weight
-	}
-
-	if err := tasks.UpdatePingTaskOrder(order); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, err.Error())
-		return
-	}
-
-	api.RespondSuccess(c, nil)
-}

web/api/admin/sessions.go

@@ -1,50 +0,0 @@
-package admin
-
-import (
-	"github.com/komari-monitor/komari/database/accounts"
-	"github.com/komari-monitor/komari/database/auditlog"
-	"github.com/komari-monitor/komari/web/api"
-
-	"github.com/gin-gonic/gin"
-)
-
-func GetSessions(c *gin.Context) {
-
-	ss, err := accounts.GetAllSessions()
-	if err != nil {
-		api.RespondError(c, 500, "Failed to retrieve sessions: "+err.Error())
-		return
-	}
-	current, _ := c.Cookie("session_token")
-	c.JSON(200, gin.H{"status": "success", "current": current, "data": ss})
-}
-
-func DeleteSession(c *gin.Context) {
-	var req struct {
-		Session string `json:"session" binding:"required"`
-	}
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, 400, "Invalid request: "+err.Error())
-		return
-	}
-	err := accounts.DeleteSession(req.Session)
-	if err != nil {
-		api.RespondError(c, 500, "Failed to delete session: "+err.Error())
-		return
-	}
-	uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), uuid.(string), "delete session", "info")
-	api.RespondSuccess(c, nil)
-}
-
-func DeleteAllSession(c *gin.Context) {
-
-	err := accounts.DeleteAllSessions()
-	if err != nil {
-		api.RespondError(c, 500, "Failed to delete all sessions: "+err.Error())
-		return
-	}
-	uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), uuid.(string), "delete all sessions", "warn")
-	api.RespondSuccess(c, nil)
-}

web/api/admin/settings.go

@@ -1,54 +0,0 @@
-package admin
-
-import (
-	"github.com/komari-monitor/komari/database/auditlog"
-	"github.com/komari-monitor/komari/database/records"
-	"github.com/komari-monitor/komari/database/tasks"
-	"github.com/komari-monitor/komari/pkg/config"
-	"github.com/komari-monitor/komari/web/api"
-
-	"github.com/gin-gonic/gin"
-)
-
-// GetSettings 获取自定义配置
-func GetSettings(c *gin.Context) {
-	cst, err := config.GetAll()
-	if err != nil {
-		api.RespondError(c, 500, "Failed to get settings: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, cst)
-}
-
-// EditSettings 更新自定义配置
-func EditSettings(c *gin.Context) {
-	cfg := make(map[string]interface{})
-	if err := c.ShouldBindJSON(&cfg); err != nil {
-		api.RespondError(c, 400, "Invalid or missing request body: "+err.Error())
-		return
-	}
-
-	if err := config.SetMany(cfg); err != nil {
-		api.RespondError(c, 500, "Failed to update settings: "+err.Error())
-		return
-	}
-
-	uuid, _ := c.Get("uuid")
-	message := "update settings: "
-	for key := range cfg {
-		message += key + ", "
-	}
-	if len(message) > 2 {
-		message = message[:len(message)-2]
-	}
-	auditlog.Log(c.ClientIP(), uuid.(string), message, "info")
-	api.RespondSuccess(c, nil)
-}
-
-func ClearAllRecords(c *gin.Context) {
-	records.DeleteAll()
-	tasks.DeleteAllPingRecords()
-	uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), uuid.(string), "clear all records", "info")
-	api.RespondSuccess(c, nil)
-}

web/api/admin/settings_xtermjs_test.go

@@ -1,375 +0,0 @@
-// Do not use t.Parallel(): config.SetDb mutates package-global state.
-package admin
-
-import (
-	"bytes"
-	"database/sql"
-	"encoding/json"
-	"fmt"
-	"net/http"
-	"net/http/httptest"
-	"strings"
-	"sync"
-	"testing"
-
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/cmd/flags"
-	"github.com/komari-monitor/komari/database/dbcore"
-	"github.com/komari-monitor/komari/pkg/config"
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-	"gorm.io/driver/sqlite"
-	"gorm.io/gorm"
-)
-
-var xtermJSAuditDBOnce sync.Once
-
-type xtermJSAPIResponse struct {
-	Status  string          `json:"status"`
-	Message string          `json:"message"`
-	Data    json.RawMessage `json:"data"`
-}
-
-func ensureXtermJSAuditDB() {
-	xtermJSAuditDBOnce.Do(func() {
-		flags.DatabaseType = "sqlite"
-		flags.DatabaseFile = "file:xtermjs_audit?mode=memory&cache=shared"
-		_ = dbcore.GetDBInstance()
-	})
-}
-
-func setupXtermJSTestDB(t *testing.T) (*gorm.DB, *sql.DB) {
-	t.Helper()
-
-	gin.SetMode(gin.TestMode)
-
-	testDBName := strings.NewReplacer("/", "_", " ", "_").Replace(t.Name())
-	testDSN := fmt.Sprintf("file:%s?mode=memory&cache=shared", testDBName)
-
-	testDB, err := gorm.Open(sqlite.Open(testDSN), &gorm.Config{})
-	require.NoError(t, err)
-
-	sqlDB, err := testDB.DB()
-	require.NoError(t, err)
-
-	// Initialize the shared audit DB first, then restore the per-test config DB.
-	ensureXtermJSAuditDB()
-	config.SetDb(testDB)
-
-	t.Cleanup(func() {
-		_ = sqlDB.Close()
-	})
-
-	return testDB, sqlDB
-}
-
-func newXtermJSRouter() *gin.Engine {
-	router := gin.New()
-	router.Use(func(c *gin.Context) {
-		c.Set("uuid", "test-uuid")
-		c.Next()
-	})
-	router.GET("/xtermjs", GetXtermJSSettings)
-	router.POST("/xtermjs", SetXtermJSSettings)
-	return router
-}
-
-func performXtermJSRequest(t *testing.T, router *gin.Engine, method, target string, body []byte) *httptest.ResponseRecorder {
-	t.Helper()
-
-	req, err := http.NewRequest(method, target, bytes.NewReader(body))
-	require.NoError(t, err)
-	if len(body) > 0 {
-		req.Header.Set("Content-Type", "application/json")
-	}
-
-	w := httptest.NewRecorder()
-	router.ServeHTTP(w, req)
-	return w
-}
-
-func decodeXtermJSResponse(t *testing.T, w *httptest.ResponseRecorder) xtermJSAPIResponse {
-	t.Helper()
-
-	var resp xtermJSAPIResponse
-	require.NoError(t, json.Unmarshal(w.Body.Bytes(), &resp))
-	return resp
-}
-
-func decodeXtermJSSettings(t *testing.T, raw json.RawMessage) XtermJSSettings {
-	t.Helper()
-
-	var settings XtermJSSettings
-	require.NoError(t, json.Unmarshal(raw, &settings))
-	return settings
-}
-
-func ptr[T any](v T) *T {
-	return &v
-}
-
-func TestGetXtermJSSettingsReturnsDefaultWhenMissing(t *testing.T) {
-	testDB, _ := setupXtermJSTestDB(t)
-	router := newXtermJSRouter()
-
-	w := performXtermJSRequest(t, router, http.MethodGet, "/xtermjs", nil)
-	require.Equal(t, http.StatusOK, w.Code)
-
-	resp := decodeXtermJSResponse(t, w)
-	assert.Equal(t, "success", resp.Status)
-	assert.Empty(t, resp.Message)
-
-	got := decodeXtermJSSettings(t, resp.Data)
-	expected := defaultXtermJSSettings()
-	assert.Equal(t, expected, got)
-
-	var stored config.ConfigItem
-	require.NoError(t, testDB.First(&stored, "key = ?", config.XtermjsSettingsKey).Error)
-
-	var persisted XtermJSSettings
-	require.NoError(t, json.Unmarshal([]byte(stored.Value), &persisted))
-	assert.Equal(t, expected, persisted)
-}
-
-func TestXtermJSSettingsRoundTripAndThemeVariants(t *testing.T) {
-	t.Run("full config round trip", func(t *testing.T) {
-		setupXtermJSTestDB(t)
-		router := newXtermJSRouter()
-
-		input := XtermJSSettings{
-			TerminalOptions: &TerminalOptions{
-				CursorBlink:     ptr(false),
-				ConvertEol:      ptr(false),
-				FontFamily:      "JetBrains Mono",
-				FontSize:        18,
-				MacOptionIsMeta: ptr(false),
-				Scrollback:      ptr(9000),
-				Theme: &ThemeConfig{
-					Foreground:          "#f8f8f2",
-					Background:          "#282a36",
-					Cursor:              "#f8f8f2",
-					SelectionBackground: "#44475a",
-					Blue:                "#6272a4",
-					ExtendedAnsi:        []string{"#111111", "#222222"},
-				},
-			},
-			TerminalPadding:       ptr(20),
-			TransparentBackground: true,
-			CustomCss:             "body { color: red; }",
-		}
-
-		body, err := json.Marshal(input)
-		require.NoError(t, err)
-
-		w := performXtermJSRequest(t, router, http.MethodPost, "/xtermjs", body)
-		require.Equal(t, http.StatusOK, w.Code)
-
-		resp := decodeXtermJSResponse(t, w)
-		assert.Equal(t, "success", resp.Status)
-		assert.Equal(t, "settings saved", resp.Message)
-
-		getResp := decodeXtermJSResponse(t, performXtermJSRequest(t, router, http.MethodGet, "/xtermjs", nil))
-		got := decodeXtermJSSettings(t, getResp.Data)
-		assert.Equal(t, input, got)
-	})
-
-	t.Run("theme null stays null", func(t *testing.T) {
-		setupXtermJSTestDB(t)
-		router := newXtermJSRouter()
-
-		body := []byte(`{"terminalOptions":{"cursorBlink":true,"convertEol":true,"fontFamily":"JetBrains Mono","fontSize":17,"macOptionIsMeta":true,"scrollback":1000,"theme":null},"terminalPadding":16,"transparentBackground":false,"customCss":"."}`)
-		w := performXtermJSRequest(t, router, http.MethodPost, "/xtermjs", body)
-		require.Equal(t, http.StatusOK, w.Code)
-
-		got := decodeXtermJSSettings(t, decodeXtermJSResponse(t, performXtermJSRequest(t, router, http.MethodGet, "/xtermjs", nil)).Data)
-		require.NotNil(t, got.TerminalOptions)
-		assert.Nil(t, got.TerminalOptions.Theme)
-	})
-
-	t.Run("terminalOptions null uses defaults", func(t *testing.T) {
-		setupXtermJSTestDB(t)
-		router := newXtermJSRouter()
-
-		body := []byte(`{"terminalOptions":null,"terminalPadding":24,"transparentBackground":true,"customCss":"body { background: black; }"}`)
-		w := performXtermJSRequest(t, router, http.MethodPost, "/xtermjs", body)
-		require.Equal(t, http.StatusOK, w.Code)
-
-		got := decodeXtermJSSettings(t, decodeXtermJSResponse(t, performXtermJSRequest(t, router, http.MethodGet, "/xtermjs", nil)).Data)
-		expected := defaultXtermJSSettings()
-		expected.TerminalPadding = ptr(24)
-		expected.TransparentBackground = true
-		expected.CustomCss = "body { background: black; }"
-		assert.Equal(t, expected, got)
-	})
-}
-
-func TestXtermJSSettingsUnknownThemeFieldsAreDropped(t *testing.T) {
-	testDB, _ := setupXtermJSTestDB(t)
-	router := newXtermJSRouter()
-
-	body := []byte(`{"terminalOptions":{"fontFamily":"JetBrains Mono","fontSize":16,"theme":{"foreground":"#ffffff","unknownField":"keep-out","extendedAnsi":["#111111"]}}}`)
-	w := performXtermJSRequest(t, router, http.MethodPost, "/xtermjs", body)
-	require.Equal(t, http.StatusOK, w.Code)
-
-	resp := decodeXtermJSResponse(t, w)
-	assert.Equal(t, "success", resp.Status)
-
-	getResp := decodeXtermJSResponse(t, performXtermJSRequest(t, router, http.MethodGet, "/xtermjs", nil))
-	got := decodeXtermJSSettings(t, getResp.Data)
-	require.NotNil(t, got.TerminalOptions)
-	require.NotNil(t, got.TerminalOptions.Theme)
-	assert.Equal(t, "#ffffff", got.TerminalOptions.Theme.Foreground)
-
-	var stored config.ConfigItem
-	require.NoError(t, testDB.First(&stored, "key = ?", config.XtermjsSettingsKey).Error)
-	assert.NotContains(t, stored.Value, "unknownField")
-}
-
-func TestXtermJSSettingsValidationAndFallbacks(t *testing.T) {
-	t.Run("extendedAnsi and body validation", func(t *testing.T) {
-		cases := []struct {
-			name string
-			body []byte
-		}{
-			{
-				name: "empty string",
-				body: []byte(`{"terminalOptions":{"theme":{"extendedAnsi":[""]}}}`),
-			},
-			{
-				name: "space string",
-				body: []byte(`{"terminalOptions":{"theme":{"extendedAnsi":["   "]}}}`),
-			},
-			{
-				name: "illegal type",
-				body: []byte(`{"terminalOptions":{"theme":{"extendedAnsi":[1]}}}`),
-			},
-			{
-				name: "invalid json",
-				body: []byte(`{"terminalOptions":`),
-			},
-		}
-
-		for _, tc := range cases {
-			t.Run(tc.name, func(t *testing.T) {
-				setupXtermJSTestDB(t)
-				router := newXtermJSRouter()
-
-				w := performXtermJSRequest(t, router, http.MethodPost, "/xtermjs", tc.body)
-				require.Equal(t, http.StatusBadRequest, w.Code)
-
-				resp := decodeXtermJSResponse(t, w)
-				assert.Equal(t, "error", resp.Status)
-				assert.NotEmpty(t, resp.Message)
-			})
-		}
-	})
-
-	t.Run("fallbacks are normalized on save and read", func(t *testing.T) {
-		testDB, _ := setupXtermJSTestDB(t)
-		router := newXtermJSRouter()
-
-		neg := -1
-		body, err := json.Marshal(XtermJSSettings{
-			TerminalOptions: &TerminalOptions{
-				CursorBlink:     ptr(true),
-				ConvertEol:      ptr(true),
-				FontFamily:      "   ",
-				FontSize:        0,
-				MacOptionIsMeta: ptr(true),
-				Scrollback:      &neg,
-				Theme: &ThemeConfig{
-					Foreground: "   ",
-				},
-			},
-			TerminalPadding:       &neg,
-			TransparentBackground: true,
-			CustomCss:             "body { color: blue; }",
-		})
-		require.NoError(t, err)
-
-		w := performXtermJSRequest(t, router, http.MethodPost, "/xtermjs", body)
-		require.Equal(t, http.StatusOK, w.Code)
-
-		got := decodeXtermJSSettings(t, decodeXtermJSResponse(t, performXtermJSRequest(t, router, http.MethodGet, "/xtermjs", nil)).Data)
-		expected := defaultXtermJSSettings()
-		expected.TransparentBackground = true
-		expected.CustomCss = "body { color: blue; }"
-		assert.Equal(t, expected, got)
-
-		var stored config.ConfigItem
-		require.NoError(t, testDB.First(&stored, "key = ?", config.XtermjsSettingsKey).Error)
-		var persisted XtermJSSettings
-		require.NoError(t, json.Unmarshal([]byte(stored.Value), &persisted))
-		assert.Equal(t, expected, persisted)
-	})
-
-	t.Run("corrupt JSON is repaired on get", func(t *testing.T) {
-		testDB, _ := setupXtermJSTestDB(t)
-		router := newXtermJSRouter()
-
-		raw := `{"terminalOptions":`
-		require.NoError(t, testDB.Create(&config.ConfigItem{
-			Key:   config.XtermjsSettingsKey,
-			Value: raw,
-		}).Error)
-
-		w := performXtermJSRequest(t, router, http.MethodGet, "/xtermjs", nil)
-		require.Equal(t, http.StatusOK, w.Code)
-
-		got := decodeXtermJSSettings(t, decodeXtermJSResponse(t, w).Data)
-		expected := defaultXtermJSSettings()
-		assert.Equal(t, expected, got)
-
-		var stored config.ConfigItem
-		require.NoError(t, testDB.First(&stored, "key = ?", config.XtermjsSettingsKey).Error)
-		var persisted XtermJSSettings
-		require.NoError(t, json.Unmarshal([]byte(stored.Value), &persisted))
-		assert.Equal(t, expected, persisted)
-	})
-}
-
-func TestXtermJSSettingsNormalizesSeededLegacyConfig(t *testing.T) {
-	testDB, _ := setupXtermJSTestDB(t)
-	router := newXtermJSRouter()
-
-	seeded := `{"terminalOptions":{"cursorBlink":true,"convertEol":true,"fontFamily":"JetBrains Mono","fontSize":0,"macOptionIsMeta":true,"scrollback":1200,"theme":null},"terminalPadding":12,"transparentBackground":true,"customCss":"seeded"}`
-	require.NoError(t, testDB.Create(&config.ConfigItem{
-		Key:   config.XtermjsSettingsKey,
-		Value: seeded,
-	}).Error)
-
-	w := performXtermJSRequest(t, router, http.MethodGet, "/xtermjs", nil)
-	require.Equal(t, http.StatusOK, w.Code)
-
-	got := decodeXtermJSSettings(t, decodeXtermJSResponse(t, w).Data)
-	expected := XtermJSSettings{
-		TerminalOptions: &TerminalOptions{
-			CursorBlink:     ptr(true),
-			ConvertEol:      ptr(true),
-			FontFamily:      "JetBrains Mono",
-			FontSize:        16,
-			MacOptionIsMeta: ptr(true),
-			Scrollback:      ptr(1200),
-			Theme:           nil,
-		},
-		TerminalPadding:       ptr(12),
-		TransparentBackground: true,
-		CustomCss:             "seeded",
-	}
-	assert.Equal(t, expected, got)
-}
-
-func TestXtermJSSettingsSaveFailureReturns500(t *testing.T) {
-	_, sqlDB := setupXtermJSTestDB(t)
-	router := newXtermJSRouter()
-
-	require.NoError(t, sqlDB.Close())
-
-	body := []byte(`{"terminalOptions":{"fontFamily":"JetBrains Mono","fontSize":16,"theme":null}}`)
-	w := performXtermJSRequest(t, router, http.MethodPost, "/xtermjs", body)
-	require.Equal(t, http.StatusInternalServerError, w.Code)
-
-	resp := decodeXtermJSResponse(t, w)
-	assert.Equal(t, "error", resp.Status)
-	assert.Contains(t, resp.Message, "Failed to save settings:")
-}

web/api/admin/task.go

@@ -1,154 +0,0 @@
-package admin
-
-import (
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/tasks"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-func GetTasks(c *gin.Context) {
-	dbTasks, err := tasks.GetAllTasks()
-	if err != nil {
-		api.RespondError(c, 500, "Failed to retrieve tasks: "+err.Error())
-		return
-	}
-	var responseTasks []gin.H
-	for _, t := range dbTasks {
-		results, err := tasks.GetTaskResultsByTaskId(t.TaskId)
-		if err != nil {
-			api.RespondError(c, 500, "Failed to retrieve task results: "+err.Error())
-			return
-		}
-
-		var filteredResults []gin.H
-		for _, r := range results {
-			filteredResults = append(filteredResults, gin.H{
-				"client":      r.Client,
-				"result":      r.Result,
-				"exit_code":   r.ExitCode,
-				"finished_at": r.FinishedAt,
-				"created_at":  r.CreatedAt,
-			})
-		}
-
-		responseTasks = append(responseTasks, gin.H{
-			"task_id": t.TaskId,
-			"clients": t.Clients,
-			"command": t.Command,
-			"results": filteredResults,
-		})
-	}
-	api.RespondSuccess(c, responseTasks)
-}
-
-func GetTaskById(c *gin.Context) {
-	taskId := c.Param("task_id")
-	if taskId == "" {
-		api.RespondError(c, 400, "Task ID is required")
-		return
-	}
-	task, err := tasks.GetTaskByTaskId(taskId)
-	if err != nil {
-		api.RespondError(c, 500, "Failed to retrieve task: "+err.Error())
-		return
-	}
-	if task == nil {
-		api.RespondError(c, 404, "Task not found")
-		return
-	}
-	results, err := tasks.GetTaskResultsByTaskId(taskId)
-	if err != nil {
-		api.RespondError(c, 500, "Failed to retrieve task results: "+err.Error())
-		return
-	}
-	var filteredResults []gin.H
-	for _, r := range results {
-		filteredResults = append(filteredResults, gin.H{
-			"client":      r.Client,
-			"result":      r.Result,
-			"exit_code":   r.ExitCode,
-			"finished_at": r.FinishedAt,
-			"created_at":  r.CreatedAt,
-		})
-	}
-	api.RespondSuccess(c, gin.H{
-		"task_id": task.TaskId,
-		"clients": task.Clients,
-		"command": task.Command,
-		"results": filteredResults,
-	})
-}
-
-func GetTasksByClientId(c *gin.Context) {
-	clientId := c.Param("uuid")
-	if clientId == "" {
-		api.RespondError(c, 400, "Client ID is required")
-		return
-	}
-	tasks, err := tasks.GetTasksByClientId(clientId)
-	if err != nil {
-		api.RespondError(c, 500, "Failed to retrieve tasks: "+err.Error())
-		return
-	}
-	if len(tasks) == 0 {
-		api.RespondError(c, 404, "No tasks found for this client")
-		return
-	}
-	api.RespondSuccess(c, tasks)
-}
-
-func GetSpecificTaskResult(c *gin.Context) {
-	taskId := c.Param("task_id")
-	clientId := c.Param("uuid")
-	if taskId == "" || clientId == "" {
-		api.RespondError(c, 400, "Task ID and Client ID are required")
-		return
-	}
-	result, err := tasks.GetSpecificTaskResult(taskId, clientId)
-	if err != nil {
-		api.RespondError(c, 500, "Failed to retrieve task result: "+err.Error())
-		return
-	}
-	if result == nil {
-		api.RespondError(c, 404, "No result found for this task and client")
-		return
-	}
-	api.RespondSuccess(c, result)
-}
-
-// Param: task_id
-func GetTaskResultsByTaskId(c *gin.Context) {
-	taskId := c.Param("task_id")
-	if taskId == "" {
-		api.RespondError(c, 400, "Task ID is required")
-		return
-	}
-	results, err := tasks.GetTaskResultsByTaskId(taskId)
-	if err != nil {
-		api.RespondError(c, 500, "Failed to retrieve task results: "+err.Error())
-		return
-	}
-	if len(results) == 0 {
-		api.RespondError(c, 404, "No results found for this task")
-		return
-	}
-	api.RespondSuccess(c, results)
-}
-
-func GetAllTaskResultByUUID(c *gin.Context) {
-	clientId := c.Param("uuid")
-	if clientId == "" {
-		api.RespondError(c, 400, "Client ID is required")
-		return
-	}
-	results, err := tasks.GetAllTasksResultByUUID(clientId)
-	if err != nil {
-		api.RespondError(c, 500, "Failed to retrieve tasks: "+err.Error())
-		return
-	}
-	if len(results) == 0 {
-		api.RespondError(c, 404, "No tasks found for this client")
-		return
-	}
-	api.RespondSuccess(c, results)
-}

web/api/admin/test/test.go

@@ -1,50 +0,0 @@
-package test
-
-import (
-	"net"
-
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/pkg/config"
-	"github.com/komari-monitor/komari/utils/geoip"
-	"github.com/komari-monitor/komari/utils/messageSender"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-func TestSendMessage(c *gin.Context) {
-	err := messageSender.SendEvent(models.EventMessage{
-		Event:   "Test",
-		Message: "This is a test message from Komari.",
-	})
-	if err != nil {
-		api.RespondError(c, 500, "Failed to send message: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, nil)
-}
-
-func TestGeoIp(c *gin.Context) {
-	ip := c.Query("ip")
-	if ip == "" {
-		if cfIP := c.GetHeader("CF-Connecting-IP"); cfIP != "" {
-			ip = cfIP
-		} else {
-			ip = c.ClientIP()
-		}
-	}
-	cfg, err := config.GetAs[bool](config.GeoIpEnabledKey, false)
-	if err != nil {
-		api.RespondError(c, 500, "Failed to get configuration: "+err.Error())
-		return
-	}
-	if !cfg {
-		api.RespondError(c, 400, "GeoIP is not enabled in the configuration.")
-		return
-	}
-	GeoIpRecord, err := geoip.GetGeoInfo(net.ParseIP(ip))
-	if err != nil {
-		api.RespondError(c, 500, "Failed to get GeoIP record: "+err.Error())
-		return
-	}
-	api.RespondSuccess(c, GeoIpRecord)
-}

web/api/admin/theme.go

@@ -108,6 +108,12 @@ func DeleteTheme(c *gin.Context) {
 		return
 	}
 
+	// 校验主题短名称，防止路径穿越（如 ../）导致删除工作目录外的任意文件
+	if !isValidThemeShort(req.Short) {
+		api.RespondError(c, http.StatusBadRequest, "无效的主题名称")
+		return
+	}
+
 	themeDir := filepath.Join("./data/theme", req.Short)
 
 	// 检查主题是否存在
@@ -135,6 +141,11 @@ func SetTheme(c *gin.Context) {
 
 	// 如果不是default主题，检查主题是否存在
 	if themeName != "default" {
+		// 校验主题名称，防止路径穿越（如 ../）访问工作目录外的文件
+		if !isValidThemeShort(themeName) {
+			api.RespondError(c, http.StatusBadRequest, "无效的主题名称")
+			return
+		}
 		themeDir := filepath.Join("./data/theme", themeName)
 		themeConfigPath := filepath.Join(themeDir, "komari-theme.json")
 
@@ -475,6 +486,12 @@ func UpdateTheme(c *gin.Context) {
 		return
 	}
 
+	// 校验主题短名称，防止路径穿越（如 ../）访问工作目录外的文件
+	if !isValidThemeShort(req.Short) {
+		api.RespondError(c, http.StatusBadRequest, "无效的主题名称")
+		return
+	}
+
 	// 检查主题是否存在
 	themeDir := filepath.Join("./data/theme", req.Short)
 	themeConfigPath := filepath.Join(themeDir, "komari-theme.json")

web/api/admin/theme_test.go

@@ -0,0 +1,47 @@
+package admin
+
+import "testing"
+
+// TestIsValidThemeShort_PathTraversal 防止 DeleteTheme/UpdateTheme/SetTheme
+// 的路径穿越漏洞：req.Short 直接进入 filepath.Join("./data/theme", short)，
+// 若 short 含 "../" 会规范化到 ./data/theme 之外，配合 os.RemoveAll 可删除
+// 工作目录外任意目录。isValidThemeShort 必须拒绝所有此类 payload。
+func TestIsValidThemeShort_PathTraversal(t *testing.T) {
+	// 必须被拒绝：路径穿越 / 绝对路径 / 目录分隔符 / 空值 / default
+	deny := []string{
+		"",
+		"default",
+		"..",
+		"../",
+		"./",
+		"../../etc",
+		"../..",
+		"foo/../bar",
+		"/etc/passwd",
+		"foo/bar",
+		"foo\\bar",
+		"a b",
+		"a;b",
+		"a$(id)",
+	}
+	for _, in := range deny {
+		if isValidThemeShort(in) {
+			t.Errorf("isValidThemeShort(%q) = true, want false (路径穿越/非法字符未被拦截)", in)
+		}
+	}
+
+	// 必须被接受：仅字母数字下划线连字符
+	accept := []string{
+		"mytheme",
+		"my-theme",
+		"my_theme",
+		"theme123",
+		"ABC",
+		"a",
+	}
+	for _, in := range accept {
+		if !isValidThemeShort(in) {
+			t.Errorf("isValidThemeShort(%q) = false, want true (合法名称被误拒)", in)
+		}
+	}
+}

web/api/admin/update.go

@@ -0,0 +1,102 @@
+package admin
+
+import (
+	"io"
+	"net/http"
+	"os"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+	"github.com/komari-monitor/komari/database/accounts"
+	"github.com/komari-monitor/komari/database/auditlog"
+	"github.com/komari-monitor/komari/utils/geoip"
+	"github.com/komari-monitor/komari/web/api"
+)
+
+// update.go
+// 文件/二进制/敏感操作类的更新接口，保留为 REST handler（不走 RPC 桥）。
+// 由原 web/api/admin/update/ 子包合并而来。
+
+func UpdateUser(c *gin.Context) {
+	var req struct {
+		Uuid     string  `json:"uuid" binding:"required"`
+		Name     *string `json:"username"`
+		Password *string `json:"password"`
+		SsoType  *string `json:"sso_type"`
+		TwoFa    string  `json:"2fa_code"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		api.RespondError(c, 400, "Invalid or missing request body: "+err.Error())
+		return
+	}
+	if req.Password == nil && req.Name == nil {
+		api.RespondError(c, 400, "At least one field (username or password) must be provided")
+		return
+	}
+	if req.Name != nil && len(*req.Name) < 3 {
+		api.RespondError(c, 400, "Username must be at least 3 characters long")
+		return
+	}
+	if req.Password != nil && len(*req.Password) < 6 {
+		api.RespondError(c, 400, "Password must be at least 6 characters long")
+		return
+	}
+	if req.Password != nil {
+		c.Set("2fa_code", req.TwoFa)
+		if err := api.VerifySensitive2FA(c); err != nil {
+			api.RespondError(c, 401, err.Error())
+			return
+		}
+	}
+	if err := accounts.UpdateUser(req.Uuid, req.Name, req.Password, req.SsoType); err != nil {
+		api.RespondError(c, 500, "Failed to update user: "+err.Error())
+		return
+	}
+	uuid, _ := c.Get("uuid")
+	auditlog.Log(c.ClientIP(), uuid.(string), "User updated", "warn")
+	api.RespondSuccess(c, gin.H{"uuid": req.Uuid})
+}
+
+func UpdateMmdbGeoIP(c *gin.Context) {
+	if err := geoip.UpdateDatabase(); err != nil {
+		api.RespondError(c, 500, "Failed to update GeoIP database "+err.Error())
+		return
+	}
+	uuid, _ := c.Get("uuid")
+	auditlog.Log(c.ClientIP(), uuid.(string), "GeoIP database updated", "info")
+	api.RespondSuccess(c, nil)
+}
+
+func UploadFavicon(c *gin.Context) {
+	c.Request.Body = http.MaxBytesReader(c.Writer, c.Request.Body, 5<<20) // 5MB
+	data, err := io.ReadAll(c.Request.Body)
+	if err != nil {
+		if strings.Contains(err.Error(), "request body too large") {
+			api.RespondError(c, http.StatusRequestEntityTooLarge, "File too large. Maximum size is 5MB")
+		} else {
+			api.RespondError(c, http.StatusBadRequest, err.Error())
+		}
+		return
+	}
+	if err := os.WriteFile("./data/favicon.ico", data, 0644); err != nil {
+		api.RespondError(c, http.StatusInternalServerError, "Failed to save favicon: "+err.Error())
+		return
+	}
+	uuid, _ := c.Get("uuid")
+	auditlog.Log(c.ClientIP(), uuid.(string), "Favicon uploaded", "info")
+	api.RespondSuccess(c, nil)
+}
+
+func DeleteFavicon(c *gin.Context) {
+	if err := os.Remove("./data/favicon.ico"); err != nil {
+		if os.IsNotExist(err) {
+			api.RespondError(c, http.StatusNotFound, "Favicon not found")
+		} else {
+			api.RespondError(c, http.StatusInternalServerError, "Failed to delete favicon: "+err.Error())
+		}
+		return
+	}
+	uuid, _ := c.Get("uuid")
+	auditlog.Log(c.ClientIP(), uuid.(string), "Favicon deleted", "info")
+	api.RespondSuccess(c, nil)
+}

web/api/admin/update/favicon.go

@@ -1,46 +0,0 @@
-package update
-
-import (
-	"io"
-	"net/http"
-	"os"
-	"strings"
-
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/auditlog"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-func UploadFavicon(c *gin.Context) {
-	c.Request.Body = http.MaxBytesReader(c.Writer, c.Request.Body, 5<<20) // 5MB
-	data, err := io.ReadAll(c.Request.Body)
-	if err != nil {
-		if strings.Contains(err.Error(), "request body too large") {
-			api.RespondError(c, http.StatusRequestEntityTooLarge, "File too large. Maximum size is 5MB")
-		} else {
-			api.RespondError(c, http.StatusBadRequest, err.Error())
-		}
-		return
-	}
-	if err := os.WriteFile("./data/favicon.ico", data, 0644); err != nil {
-		api.RespondError(c, http.StatusInternalServerError, "Failed to save favicon: "+err.Error())
-		return
-	}
-	uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), uuid.(string), "Favicon uploaded", "info")
-	api.RespondSuccess(c, nil)
-}
-
-func DeleteFavicon(c *gin.Context) {
-	if err := os.Remove("./data/favicon.ico"); err != nil {
-		if os.IsNotExist(err) {
-			api.RespondError(c, http.StatusNotFound, "Favicon not found")
-		} else {
-			api.RespondError(c, http.StatusInternalServerError, "Failed to delete favicon: "+err.Error())
-		}
-		return
-	}
-	uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), uuid.(string), "Favicon deleted", "info")
-	api.RespondSuccess(c, nil)
-}

web/api/admin/update/geoip.go

@@ -1,18 +0,0 @@
-package update
-
-import (
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/auditlog"
-	"github.com/komari-monitor/komari/utils/geoip"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-func UpdateMmdbGeoIP(c *gin.Context) {
-	if err := geoip.UpdateDatabase(); err != nil {
-		api.RespondError(c, 500, "Failed to update GeoIP database "+err.Error())
-		return
-	}
-	uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), uuid.(string), "GeoIP database updated", "info")
-	api.RespondSuccess(c, nil)
-}

web/api/admin/update/user.go

@@ -1,48 +0,0 @@
-package update
-
-import (
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/accounts"
-	"github.com/komari-monitor/komari/database/auditlog"
-	"github.com/komari-monitor/komari/web/api"
-)
-
-func UpdateUser(c *gin.Context) {
-	var req struct {
-		Uuid     string  `json:"uuid" binding:"required"`
-		Name     *string `json:"username"`
-		Password *string `json:"password"`
-		SsoType  *string `json:"sso_type"`
-		TwoFa    string  `json:"2fa_code"`
-	}
-	if err := c.ShouldBindJSON(&req); err != nil {
-		api.RespondError(c, 400, "Invalid or missing request body: "+err.Error())
-		return
-	}
-	if req.Password == nil && req.Name == nil {
-		api.RespondError(c, 400, "At least one field (username or password) must be provided")
-		return
-	}
-	if req.Name != nil && len(*req.Name) < 3 {
-		api.RespondError(c, 400, "Username must be at least 3 characters long")
-		return
-	}
-	if req.Password != nil && len(*req.Password) < 6 {
-		api.RespondError(c, 400, "Password must be at least 6 characters long")
-		return
-	}
-	if req.Password != nil {
-		c.Set("2fa_code", req.TwoFa)
-		if err := api.VerifySensitive2FA(c); err != nil {
-			api.RespondError(c, 401, err.Error())
-			return
-		}
-	}
-	if err := accounts.UpdateUser(req.Uuid, req.Name, req.Password, req.SsoType); err != nil {
-		api.RespondError(c, 500, "Failed to update user: "+err.Error())
-		return
-	}
-	uuid, _ := c.Get("uuid")
-	auditlog.Log(c.ClientIP(), uuid.(string), "User updated", "warn")
-	api.RespondSuccess(c, gin.H{"uuid": req.Uuid})
-}

web/api/client/ingest.go

@@ -0,0 +1,49 @@
+package client
+
+import (
+	"time"
+
+	"github.com/komari-monitor/komari/database/models"
+	"github.com/komari-monitor/komari/database/tasks"
+	v1 "github.com/komari-monitor/komari/protocol/v1"
+	agent_runtime "github.com/komari-monitor/komari/web/agent"
+)
+
+// ingest.go
+// agent 上报数据的传输无关处理逻辑。v1 (REST/WS) 与 v2 (JSON-RPC) 两套上报入口
+// 经过各自的协议解析后，统一调用这里的函数落库并更新运行时状态，消除重复。
+
+// ingestReport 保存一次负载上报并刷新运行时状态。
+// protocolVersion 标记上报所用协议（1 或 2），用于运行时区分客户端能力。
+// markPresence 为 true 时按 POST 上报会话刷新在线状态（WS 连接自行管理在线状态，应传 false）。
+func ingestReport(uuid string, report v1.Report, protocolVersion int, markPresence bool) error {
+	report.UUID = uuid
+	savedReport, err := SaveClientReport(uuid, report)
+	if err != nil {
+		return err
+	}
+	agent_runtime.SetLatestReport(uuid, &savedReport)
+	agent_runtime.SetClientProtocolVersion(uuid, protocolVersion)
+	if markPresence {
+		refreshPostPresence(uuid)
+	}
+	return nil
+}
+
+// ingestBasicInfo 保存客户端基础信息。fallbackIP 在上报未携带 IP 时用作兜底。
+func ingestBasicInfo(uuid string, info map[string]interface{}, fallbackIP string) error {
+	if info == nil {
+		info = map[string]interface{}{}
+	}
+	return saveClientBasicInfo(info, uuid, fallbackIP)
+}
+
+// ingestPingResult 保存一条 ping 探测结果。
+func ingestPingResult(uuid string, taskID uint, value int, finishedAt time.Time) {
+	tasks.SavePingRecord(models.PingRecord{
+		Client: uuid,
+		TaskId: taskID,
+		Value:  value,
+		Time:   models.FromTime(finishedAt),
+	})
+}

web/api/client/ping.go

@@ -1,55 +0,0 @@
-package client
-
-import (
-	"net/http"
-	"time"
-
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/database/tasks"
-)
-
-func GetPingTasks(c *gin.Context) {
-	uuid, _ := c.Get("client_uuid")
-	clientUUID, _ := uuid.(string)
-	if clientUUID == "" {
-		c.JSON(http.StatusBadRequest, gin.H{"error": "client_uuid not found"})
-		return
-	}
-
-	result := tasks.GetPingTasksByClient(clientUUID)
-	c.JSON(http.StatusOK, result)
-}
-
-func UploadPingResult(c *gin.Context) {
-	uuid, _ := c.Get("client_uuid")
-	clientUUID, _ := uuid.(string)
-	if clientUUID == "" {
-		c.JSON(http.StatusBadRequest, gin.H{"error": "client_uuid not found"})
-		return
-	}
-
-	var req struct {
-		TaskID     uint      `json:"task_id" binding:"required"`
-		Value      int       `json:"value"`
-		PingType   string    `json:"ping_type"`
-		FinishedAt time.Time `json:"finished_at" binding:"required"`
-	}
-	if err := c.ShouldBindJSON(&req); err != nil {
-		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request: " + err.Error()})
-		return
-	}
-
-	record := models.PingRecord{
-		Client: clientUUID,
-		TaskId: req.TaskID,
-		Value:  req.Value,
-		Time:   models.FromTime(req.FinishedAt),
-	}
-	if err := tasks.SavePingRecord(record); err != nil {
-		c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to save ping result: " + err.Error()})
-		return
-	}
-
-	c.JSON(http.StatusOK, gin.H{"status": "success"})
-}

web/api/client/report.go

@@ -14,21 +14,19 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/gorilla/websocket"
 	"github.com/komari-monitor/komari/database/clients"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/database/tasks"
 	v1 "github.com/komari-monitor/komari/protocol/v1"
 	"github.com/komari-monitor/komari/utils/notifier"
 	agent_runtime "github.com/komari-monitor/komari/web/agent"
 	"github.com/komari-monitor/komari/web/api"
 	"github.com/komari-monitor/komari/web/connection"
 	report_cache "github.com/komari-monitor/komari/web/report"
-	"github.com/patrickmn/go-cache"
 )
 
 const (
 	// 如果超过这个时间没有收到任何消息，则认为连接已死
 	// 因为目前server没有存agent的信息上报间隔。只有写一个默认的
-	readWait = 11 * time.Second
+	readWait        = 11 * time.Second
+	postPresenceTTL = 35 * time.Second
 )
 
 // postPresenceEntry 保存单个客户端的 POST 上报会话状态
@@ -55,16 +53,16 @@ func refreshPostPresence(uuid string) {
 		entry.timer.Stop()
 		// 重新创建 AfterFunc 以在闭包中捕获新的 generation
 		gen := entry.generation
-		entry.timer = time.AfterFunc(readWait, func() {
+		entry.timer = time.AfterFunc(postPresenceTTL, func() {
 			postPresenceExpired(uuid, entry.connID, gen)
 		})
-		agent_runtime.KeepAlivePresence(uuid, entry.connID, readWait)
+		agent_runtime.KeepAlivePresence(uuid, entry.connID, postPresenceTTL)
 		return
 	}
 
 	// 新 POST 会话：生成 connID，标记在线，启动超时定时器
 	connID := time.Now().UnixNano()
-	agent_runtime.KeepAlivePresence(uuid, connID, readWait)
+	agent_runtime.KeepAlivePresence(uuid, connID, postPresenceTTL)
 	agent_runtime.SetClientProtocolVersion(uuid, 1)
 	go notifier.OnlineNotification(uuid, connID)
 
@@ -75,7 +73,7 @@ func refreshPostPresence(uuid string) {
 		generation: defaultGeneration,
 	}
 
-	entry.timer = time.AfterFunc(readWait, func() {
+	entry.timer = time.AfterFunc(postPresenceTTL, func() {
 		postPresenceExpired(uuid, connID, defaultGeneration)
 	})
 
@@ -120,8 +118,6 @@ func UploadReport(c *gin.Context) {
 		c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid request body"})
 		return
 	}
-	report.UpdatedAt = time.Now()
-
 	// 优先使用 body 中的 UUID，若为空则从中间件注入的上下文中获取
 	uuid := report.UUID
 	if uuid == "" {
@@ -133,19 +129,12 @@ func UploadReport(c *gin.Context) {
 		c.JSON(http.StatusBadRequest, gin.H{"error": "UUID is required"})
 		return
 	}
-	report.UUID = uuid
 
-	err = SaveClientReport(uuid, report)
-	if err != nil {
+	// POST 上报：落库、更新运行时状态并刷新在线状态
+	if err := ingestReport(uuid, report, 1, true); err != nil {
 		c.JSON(http.StatusInternalServerError, gin.H{"error": fmt.Sprintf("%v", err)})
 		return
 	}
-	// Update report with method and token
-
-	agent_runtime.SetLatestReport(uuid, &report)
-
-	// POST 上报后标记节点在线，超时未收到新 POST 则触发离线
-	refreshPostPresence(uuid)
 
 	c.Request.Body = io.NopCloser(bytes.NewBuffer(bodyBytes)) // Restore the body for further use
 	c.JSON(200, gin.H{"status": "success"})
@@ -250,13 +239,11 @@ func processMessage(conn *connection.SafeConn, message []byte, uuid string) {
 			conn.WriteJSON(gin.H{"status": "error", "error": "Invalid report format"})
 			return
 		}
-		report.UpdatedAt = time.Now()
-		err = SaveClientReport(uuid, report)
-		if err != nil {
+		// WS 连接自行管理在线状态，无需刷新 POST presence
+		if err := ingestReport(uuid, report, 1, false); err != nil {
 			conn.WriteJSON(gin.H{"status": "error", "error": fmt.Sprintf("%v", err)})
 			return
 		}
-		agent_runtime.SetLatestReport(uuid, &report)
 	case "ping_result":
 		var reqBody struct {
 			PingTaskID uint      `json:"task_id"`
@@ -269,29 +256,16 @@ func processMessage(conn *connection.SafeConn, message []byte, uuid string) {
 			conn.WriteJSON(gin.H{"status": "error", "error": "Invalid ping result format"})
 			return
 		}
-		pingResult := models.PingRecord{
-			Client: uuid,
-			TaskId: reqBody.PingTaskID,
-			Value:  reqBody.PingResult,
-			Time:   models.FromTime(reqBody.FinishedAt),
-		}
-		tasks.SavePingRecord(pingResult)
+		ingestPingResult(uuid, reqBody.PingTaskID, reqBody.PingResult, reqBody.FinishedAt)
 	default:
 		log.Printf("Unknown message type: %s", msgType.Type)
 		conn.WriteJSON(gin.H{"status": "error", "error": "Unknown message type"})
 	}
 }
 
-func SaveClientReport(uuid string, report v1.Report) error {
-	reports, _ := report_cache.Records.Get(uuid)
-	if reports == nil {
-		reports = []v1.Report{}
-	}
+func SaveClientReport(uuid string, report v1.Report) (v1.Report, error) {
 	if report.CPU.Usage < 0.01 {
 		report.CPU.Usage = 0.01
 	}
-	reports = append(reports.([]v1.Report), report)
-	report_cache.Records.Set(uuid, reports, cache.DefaultExpiration)
-
-	return nil
+	return report_cache.AppendClientReport(uuid, report)
 }

web/api/client/report_v2.go

@@ -13,8 +13,6 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/gorilla/websocket"
 	"github.com/komari-monitor/komari/database/clients"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/database/tasks"
 	v2 "github.com/komari-monitor/komari/protocol/v2"
 	"github.com/komari-monitor/komari/utils/notifier"
 	agent_runtime "github.com/komari-monitor/komari/web/agent"
@@ -53,15 +51,9 @@ func handleV2RPC(uuid string, req v2.Request, allowWait bool) v2.Response {
 		if err := bindV2Params(req.Params, &params); err != nil {
 			return v2.Error(req.ID, -32602, "invalid report params", err.Error())
 		}
-		report := params.Report
-		report.UUID = uuid
-		report.UpdatedAt = time.Now()
-		if err := SaveClientReport(uuid, report); err != nil {
+		if err := ingestReport(uuid, params.Report, 2, true); err != nil {
 			return v2.Error(req.ID, -32000, "failed to save report", err.Error())
 		}
-		agent_runtime.SetLatestReport(uuid, &report)
-		refreshPostPresence(uuid)
-		agent_runtime.SetClientProtocolVersion(uuid, 2)
 		return v2.Success(req.ID, gin.H{
 			"status": "success",
 			"events": agent_runtime.TakeV2Events(uuid, params.AckEventIDs, 8),
@@ -71,11 +63,7 @@ func handleV2RPC(uuid string, req v2.Request, allowWait bool) v2.Response {
 		if err := bindV2Params(req.Params, &params); err != nil {
 			return v2.Error(req.ID, -32602, "invalid basic info params", err.Error())
 		}
-		if params.Info == nil {
-			params.Info = map[string]interface{}{}
-		}
-		params.Info["uuid"] = uuid
-		if err := clients.SaveClientInfo(params.Info); err != nil {
+		if err := ingestBasicInfo(uuid, params.Info, ""); err != nil {
 			return v2.Error(req.ID, -32000, "failed to save basic info", err.Error())
 		}
 		return v2.Success(req.ID, gin.H{"status": "success"})
@@ -90,12 +78,7 @@ func handleV2RPC(uuid string, req v2.Request, allowWait bool) v2.Response {
 				finishedAt = t
 			}
 		}
-		tasks.SavePingRecord(models.PingRecord{
-			Client: uuid,
-			TaskId: params.TaskID,
-			Value:  params.Value,
-			Time:   models.FromTime(finishedAt),
-		})
+		ingestPingResult(uuid, params.TaskID, params.Value, finishedAt)
 		return v2.Success(req.ID, gin.H{"status": "success"})
 	case v2.MethodAgentPull:
 		var params v2.PullParams
@@ -168,6 +151,9 @@ func WebSocketV2RPC(c *gin.Context) {
 		agent_runtime.DeleteClientConditionally(uuid, conn)
 		notifierOffline(uuid, conn.ID)
 	}()
+	if !pushQueuedV2Events(conn, uuid) {
+		return
+	}
 
 	for {
 		conn.SetReadDeadline(time.Now().Add(readWait))
@@ -194,6 +180,25 @@ func WebSocketV2RPC(c *gin.Context) {
 	}
 }
 
+func pushQueuedV2Events(conn *connection.SafeConn, uuid string) bool {
+	events := agent_runtime.TakeV2Events(uuid, nil, 0)
+	if len(events) == 0 {
+		return true
+	}
+	ackIDs := make([]string, 0, len(events))
+	for _, event := range events {
+		payload := v2.Request{JSONRPC: v2.Version, Method: event.Method, Params: event.Params}
+		if err := conn.WriteJSON(payload); err != nil {
+			agent_runtime.AckV2Events(uuid, ackIDs)
+			log.Printf("failed to push queued v2 event %s to client %s: %v", event.ID, uuid, err)
+			return false
+		}
+		ackIDs = append(ackIDs, event.ID)
+	}
+	agent_runtime.AckV2Events(uuid, ackIDs)
+	return true
+}
+
 func clientUUIDFromContext(c *gin.Context) (string, bool) {
 	if v, ok := c.Get("client_uuid"); ok {
 		if uuid, ok := v.(string); ok && uuid != "" {

web/api/client/task.go

@@ -1,36 +0,0 @@
-package client
-
-import (
-	"time"
-
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/clients"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/database/tasks"
-)
-
-func TaskResult(c *gin.Context) {
-	token := c.Query("token")
-	clientId, _ := clients.GetClientUUIDByToken(token)
-	if clientId == "" {
-		c.JSON(400, gin.H{"status": "error", "message": "Invalid or missing token"})
-		return
-	}
-	var req struct {
-		TaskId     string    `json:"task_id" binding:"required"`
-		Result     string    `json:"result"`
-		ExitCode   int       `json:"exit_code"`
-		FinishedAt time.Time `json:"finished_at" binding:"required"`
-	}
-	if err := c.ShouldBindJSON(&req); err != nil {
-		c.JSON(400, gin.H{"status": "error", "message": "Invalid request"})
-		return
-	}
-
-	if err := tasks.SaveTaskResult(req.TaskId, clientId, req.Result, req.ExitCode, models.FromTime(req.FinishedAt)); err != nil {
-		c.JSON(500, gin.H{"status": "error", "message": "Failed to update task result: " + err.Error()})
-		return
-	}
-
-	c.JSON(200, gin.H{"status": "success", "message": "Task result updated successfully"})
-}

web/api/client/uploadBasicInfo.go

@@ -22,6 +22,65 @@ func getClientIPType(ip net.IP) int {
 	}
 }
 
+func saveClientBasicInfo(info map[string]interface{}, uuid string, fallbackIP string) error {
+	info["uuid"] = uuid
+	applyFallbackClientIP(info, fallbackIP)
+	appendClientRegionFromGeoIP(info)
+	return clients.SaveClientInfo(info)
+}
+
+func applyFallbackClientIP(info map[string]interface{}, fallbackIP string) {
+	if hasClientIP(info) {
+		return
+	}
+	ip := net.ParseIP(fallbackIP)
+
+	switch getClientIPType(ip) {
+	case 0:
+		info["ipv4"] = fallbackIP
+	case 1:
+		info["ipv6"] = fallbackIP
+	}
+}
+
+func hasClientIP(info map[string]interface{}) bool {
+	if ipv4, ok := info["ipv4"].(string); ok && ipv4 != "" {
+		return true
+	}
+	if ipv6, ok := info["ipv6"].(string); ok && ipv6 != "" {
+		return true
+	}
+	return false
+}
+
+func appendClientRegionFromGeoIP(info map[string]interface{}) {
+	cfg, err := config.GetAs[bool](config.GeoIpEnabledKey)
+	if err != nil || !cfg {
+		return
+	}
+
+	for _, key := range []string{"ipv4", "ipv6"} {
+		ipStr, ok := info[key].(string)
+		if !ok || ipStr == "" {
+			continue
+		}
+		ip := net.ParseIP(ipStr)
+		if ip == nil {
+			continue
+		}
+		record, _ := geoip.GetGeoInfo(ip)
+		if record == nil {
+			continue
+		}
+		region := geoip.GetRegionUnicodeEmoji(record.ISOCode)
+		if region == "" {
+			continue
+		}
+		info["region"] = region
+		return
+	}
+}
+
 func UploadBasicInfo(c *gin.Context) {
 	var cbi = map[string]interface{}{}
 	if err := c.ShouldBindJSON(&cbi); err != nil {
@@ -36,47 +95,7 @@ func UploadBasicInfo(c *gin.Context) {
 		return
 	}
 
-	cbi["uuid"] = uuid
-	// 如果没有传入 IP 信息，尝试从请求中获取
-	if (func() bool {
-		if v4, ok := cbi["ipv4"].(string); !ok || v4 == "" {
-			if v6, ok := cbi["ipv6"].(string); !ok || v6 == "" {
-				return true
-			}
-		}
-		return false
-	})() {
-		ipStr := c.ClientIP()
-		ip := net.ParseIP(ipStr)
-		ipType := getClientIPType(ip)
-
-		switch ipType {
-		case 0:
-			cbi["ipv4"] = ipStr
-		case 1:
-			cbi["ipv6"] = ipStr
-		default:
-			break
-		}
-	}
-
-	if cfg, err := config.GetAs[bool](config.GeoIpEnabledKey); err == nil && cfg {
-		if ipv4, ok := cbi["ipv4"].(string); ok && ipv4 != "" {
-			ip4 := net.ParseIP(ipv4)
-			ip4_record, _ := geoip.GetGeoInfo(ip4)
-			if ip4_record != nil {
-				cbi["region"] = geoip.GetRegionUnicodeEmoji(ip4_record.ISOCode)
-			}
-		} else if ipv6, ok := cbi["ipv6"].(string); ok && ipv6 != "" {
-			ip6 := net.ParseIP(ipv6)
-			ip6_record, _ := geoip.GetGeoInfo(ip6)
-			if ip6_record != nil {
-				cbi["region"] = geoip.GetRegionUnicodeEmoji(ip6_record.ISOCode)
-			}
-		}
-	}
-
-	if err := clients.SaveClientInfo(cbi); err != nil {
+	if err := ingestBasicInfo(uuid, cbi, c.ClientIP()); err != nil {
 		c.JSON(500, gin.H{"status": "error", "error": err})
 		return
 	}

web/api/client/uploadBasicInfo_test.go

@@ -0,0 +1,86 @@
+package client
+
+import (
+	"net"
+	"testing"
+	"time"
+
+	"github.com/komari-monitor/komari/cmd/flags"
+	"github.com/komari-monitor/komari/database/dbcore"
+	"github.com/komari-monitor/komari/database/models"
+	"github.com/komari-monitor/komari/pkg/config"
+	v2 "github.com/komari-monitor/komari/protocol/v2"
+	"github.com/komari-monitor/komari/utils/geoip"
+)
+
+type staticGeoIPProvider struct {
+	name string
+	iso  string
+}
+
+func (p staticGeoIPProvider) Name() string {
+	return p.name
+}
+
+func (p staticGeoIPProvider) GetGeoInfo(ip net.IP) (*geoip.GeoInfo, error) {
+	return &geoip.GeoInfo{ISOCode: p.iso, Name: p.iso}, nil
+}
+
+func (p staticGeoIPProvider) UpdateDatabase() error {
+	return nil
+}
+
+func (p staticGeoIPProvider) Close() error {
+	return nil
+}
+
+func TestV2BasicInfoFillsRegionFromGeoIP(t *testing.T) {
+	flags.DatabaseType = "sqlite"
+	flags.DatabaseFile = "file:v2_basic_info_geoip?mode=memory&cache=shared"
+
+	db := dbcore.GetDBInstance()
+	if err := config.Set(config.GeoIpEnabledKey, true); err != nil {
+		t.Fatalf("enable geoip: %v", err)
+	}
+
+	oldProvider := geoip.CurrentProvider
+	geoip.CurrentProvider = staticGeoIPProvider{name: t.Name(), iso: "SG"}
+	t.Cleanup(func() {
+		geoip.CurrentProvider = oldProvider
+	})
+
+	clientUUID := "client-v2-geoip"
+	now := models.FromTime(time.Now())
+	if err := db.Create(&models.Client{
+		UUID:      clientUUID,
+		Token:     "token-v2-geoip",
+		Name:      "client_v2_geoip",
+		CreatedAt: now,
+		UpdatedAt: now,
+	}).Error; err != nil {
+		t.Fatalf("create client: %v", err)
+	}
+
+	resp := handleV2RPC(clientUUID, v2.Request{
+		JSONRPC: v2.Version,
+		Method:  v2.MethodAgentBasicInfo,
+		Params: map[string]interface{}{
+			"info": map[string]interface{}{
+				"ipv4": "8.8.8.8",
+			},
+		},
+		ID: "basic-info",
+	}, false)
+	if resp.Error != nil {
+		t.Fatalf("v2 basic info failed: %+v", resp.Error)
+	}
+
+	var got models.Client
+	if err := db.First(&got, "uuid = ?", clientUUID).Error; err != nil {
+		t.Fatalf("load client: %v", err)
+	}
+	want := geoip.GetRegionUnicodeEmoji("SG")
+	if got.Region != want {
+		t.Fatalf("expected GeoIP region to be saved, got %q", got.Region)
+	}
+}

web/api/principal.go

@@ -0,0 +1,56 @@
+package api
+
+// principal.go
+// 统一身份识别。IdentifyPrincipal 是全局唯一的主体识别入口,替代此前散落在
+// IdentityMiddleware / transport.detectPermissionGroup / transport.buildContextMeta
+// 三处的重复逻辑。识别优先级:API Key > Session(用户) > Client Token(agent) > 匿名。
+
+import (
+	"github.com/gin-gonic/gin"
+	"github.com/komari-monitor/komari/database/accounts"
+	"github.com/komari-monitor/komari/pkg/rpc"
+)
+
+// principalContextKey 是 principal 在 gin.Context 中的存储键。
+const principalContextKey = "principal"
+
+// IdentifyPrincipal 识别当前请求的调用主体。不写入任何状态,可安全多次调用。
+// 识别优先级与历史 IdentityMiddleware 一致:API Key > Session > Client Token > 匿名。
+func IdentifyPrincipal(c *gin.Context) *rpc.Principal {
+	// 1. API Key(Authorization: Bearer <key>)
+	if isApiKeyValid(c.GetHeader("Authorization")) {
+		return rpc.NewAPIKeyPrincipal()
+	}
+
+	// 2. Session(管理员用户)
+	if session, err := c.Cookie("session_token"); err == nil && session != "" {
+		if uuid, err := accounts.GetSession(session); err == nil && uuid != "" {
+			return rpc.NewUserPrincipal(uuid)
+		}
+	}
+
+	// 3. Client Token(agent 客户端)
+	if token := extractClientToken(c); token != "" {
+		if uuid, err := checkTokenAndGetUUID(token); err == nil && uuid != "" {
+			return rpc.NewAgentPrincipal(uuid)
+		}
+	}
+
+	// 4. 匿名访客
+	return rpc.NewAnonymousPrincipal()
+}
+
+// SetPrincipal 将已识别的主体写入 gin.Context,供下游(RPC 边界等)复用。
+func SetPrincipal(c *gin.Context, p *rpc.Principal) {
+	c.Set(principalContextKey, p)
+}
+
+// GetPrincipal 读取已识别的主体;若中间件未运行则返回 nil。
+func GetPrincipal(c *gin.Context) *rpc.Principal {
+	if v, ok := c.Get(principalContextKey); ok {
+		if p, ok := v.(*rpc.Principal); ok {
+			return p
+		}
+	}
+	return nil
+}

web/api/public/latestRecord.go

@@ -1,46 +0,0 @@
-package public
-
-import (
-	"github.com/gin-gonic/gin"
-	"github.com/komari-monitor/komari/database/accounts"
-	"github.com/komari-monitor/komari/database/dbcore"
-	"github.com/komari-monitor/komari/database/models"
-	"github.com/komari-monitor/komari/web/api"
-	report_cache "github.com/komari-monitor/komari/web/report"
-)
-
-func GetClientRecentRecords(c *gin.Context) {
-	uuid := c.Param("uuid")
-
-	if uuid == "" {
-		api.RespondError(c, 400, "UUID is required")
-		return
-	}
-
-	// 登录状态检查
-	isLogin := false
-	session, _ := c.Cookie("session_token")
-	_, err := accounts.GetUserBySession(session)
-	if err == nil {
-		isLogin = true
-	}
-
-	// 仅在未登录时需要 Hidden 信息做过滤
-	hiddenMap := map[string]bool{}
-	if !isLogin {
-		var hiddenClients []models.Client
-		db := dbcore.GetDBInstance()
-		_ = db.Select("uuid").Where("hidden = ?", true).Find(&hiddenClients).Error
-		for _, cli := range hiddenClients {
-			hiddenMap[cli.UUID] = true
-		}
-
-		if hiddenMap[uuid] {
-			api.RespondError(c, 400, "UUID is required") //防止未登录用户获取隐藏客户端数据
-			return
-		}
-	}
-
-	records, _ := report_cache.Records.Get(uuid)
-	api.RespondSuccess(c, records)
-}

web/api/public/login.go

@@ -28,7 +28,7 @@ func setSessionCookie(c *gin.Context, value string, maxAge int) {
 		Value:    value,
 		Path:     "/",
 		MaxAge:   maxAge,
-		Secure:   utils.GetScheme(c) == "https" || utils.CurrentVersion != "dev",
+		Secure:   utils.GetScheme(c) == "https",
 		HttpOnly: true,
 		SameSite: http.SameSiteLaxMode,
 	})