perf: optimize search processing

.github/actions/benchmark-capture/action.yml

@@ -0,0 +1,92 @@
+name: "📊 capture benchmark"
+description: "Capture benchmarks for the base and head."
+
+inputs:
+  base-repository:
+    description: "Repository"
+    required: true
+  base-sha:
+    description: "Base commit SHA"
+    required: true
+  head-sha:
+    description: "Head commit SHA"
+    required: true
+  capture-command:
+    description: "Benchmark Capture command"
+    required: true
+  compare-command:
+    description: "Benchmark Comparison command"
+    required: true
+  base-results-path:
+    description: "Relative base results directory"
+    required: true
+  head-results-path:
+    description: "Relative head results directory"
+    required: true
+
+outputs:
+  artifact-id:
+    description: "Results Artifact ID"
+    value: ${{ steps.results.outputs.artifact-id }}
+
+runs:
+  using: composite
+  steps:
+    - name: "📥 check out base"
+      uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+      with:
+        repository: ${{ inputs.base-repository }}
+        ref: ${{ inputs.base-sha }}
+        path: benchmark-base
+        persist-credentials: false
+
+    - name: "⚙️ set up base"
+      shell: bash
+      working-directory: benchmark-base
+      run: |
+        pnpm install --frozen-lockfile
+        pnpm build
+
+    - name: "📂 stage base output"
+      shell: bash
+      run: |
+        mkdir -p dist
+        cp -R benchmark-base/dist/. dist/
+
+    - name: "📊 capture base benchmark"
+      shell: bash
+      env:
+        COMMAND: ${{ inputs.capture-command }}
+      run: bash -eo pipefail -c "$COMMAND"
+
+    - name: "⚙️ set up head"
+      shell: bash
+      run: pnpm build
+
+    - name: "📊 compare benchmark"
+      shell: bash
+      env:
+        COMMAND: ${{ inputs.compare-command }}
+      run: bash -eo pipefail -c "$COMMAND"
+
+    - name: "📂 collect results"
+      shell: bash
+      env:
+        BASE_SHA: ${{ inputs.base-sha }}
+        HEAD_SHA: ${{ inputs.head-sha }}
+        BASE_RESULTS_PATH: ${{ inputs.base-results-path }}
+        HEAD_RESULTS_PATH: ${{ inputs.head-results-path }}
+      run: |
+        mkdir -p benchmark-results/base benchmark-results/head
+        cp "$BASE_RESULTS_PATH"/*.json benchmark-results/base/
+        cp "$HEAD_RESULTS_PATH"/*.json benchmark-results/head/
+        node -e 'require("node:fs").writeFileSync("benchmark-results/measurement.json", JSON.stringify({base: process.env.BASE_SHA, head: process.env.HEAD_SHA}))'
+
+    - name: "📤 upload results"
+      id: results
+      uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
+      with:
+        name: benchmark-results-${{ github.run_attempt }}
+        path: benchmark-results/
+        if-no-files-found: error
+        retention-days: 1

.github/actions/benchmark-report/action.yml

@@ -0,0 +1,41 @@
+name: "📝 report benchmarks"
+description: "Download benchmark results and publish the report"
+
+inputs:
+  artifact-id:
+    description: "Results Artifact ID"
+    required: true
+  base-sha:
+    description: "Base commit SHA"
+    required: true
+  head-sha:
+    description: "Head commit SHA"
+    required: true
+  unit:
+    description: "Mean latency unit label"
+    default: "ms/operation"
+  operations-per-sample:
+    description: "Operations p/sample"
+    default: "1"
+
+runs:
+  using: composite
+  steps:
+    - name: "📥 download results"
+      uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8.0.2
+      with:
+        artifact-ids: ${{ inputs.artifact-id }}
+        path: ${{ runner.temp }}/benchmark-results
+
+    - name: "📝 publish results"
+      uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
+      env:
+        BASE_SHA: ${{ inputs.base-sha }}
+        HEAD_SHA: ${{ inputs.head-sha }}
+        RESULTS: ${{ runner.temp }}/benchmark-results
+        RESULT_UNIT: ${{ inputs.unit }}
+        OPERATIONS_PER_SAMPLE: ${{ inputs.operations-per-sample }}
+      with:
+        script: |
+          const { publish } = await import(`${process.env.GITHUB_WORKSPACE}/.github/scripts/benchmark-report.mjs`);
+          await publish({ github, context, core });

.github/scripts/benchmark-report.mjs

@@ -0,0 +1,170 @@
+/* oxlint-disable typescript/no-unsafe-member-access, typescript/no-unsafe-return */
+// Native github-script API objects and parsed JSON have no SDK types here.
+
+import { readFile, readdir } from 'node:fs/promises';
+import path from 'node:path';
+
+export async function publish({ github, context, core }) {
+  const operationsPerSample = Number(process.env.OPERATIONS_PER_SAMPLE ?? '1');
+  const unit = process.env.RESULT_UNIT ?? 'ms/operation';
+
+  const source = {
+    pr: context.payload.pull_request.number,
+    run: context.runId,
+    number: context.runNumber,
+    attempt: Number(process.env.GITHUB_RUN_ATTEMPT),
+    base: process.env.BASE_SHA,
+    head: process.env.HEAD_SHA,
+  };
+
+  const repository = context.repo;
+
+  const skip = (reason) => {
+    core.notice(`Benchmark report skipped: ${reason}`);
+  };
+
+  let rows;
+
+  try {
+    if (!Number.isFinite(operationsPerSample) || operationsPerSample <= 0) {
+      skip('operations per sample must be finite and positive');
+
+      return;
+    }
+
+    const measurement = await readJson(path.join(process.env.RESULTS, 'measurement.json'));
+
+    if (measurement?.base !== source.base || measurement?.head !== source.head) {
+      skip('measurement SHAs do not match the producer base and head');
+
+      return;
+    }
+
+    const baseDirectory = path.join(process.env.RESULTS, 'base');
+    const headDirectory = path.join(process.env.RESULTS, 'head');
+    const baseEntries = await readdir(baseDirectory);
+    const headEntries = await readdir(headDirectory);
+    const baseFiles = baseEntries.toSorted((a, b) => a.localeCompare(b));
+    const headFiles = headEntries.toSorted((a, b) => a.localeCompare(b));
+
+    if (
+      baseFiles.length === 0 ||
+      baseFiles.length !== headFiles.length ||
+      baseFiles.some((file, index) => file !== headFiles[index] || !/^[a-z0-9][a-z0-9.-]{0,95}\.json$/u.test(file))
+    ) {
+      skip('native results must contain matching JSON files with safe workload IDs');
+
+      return;
+    }
+
+    rows = [];
+
+    for (const file of baseFiles) {
+      const baseResult = await readJson(path.join(baseDirectory, file));
+      const headResult = await readJson(path.join(headDirectory, file));
+      const base = Number.isFinite(baseResult?.latency?.mean) ? baseResult.latency.mean / operationsPerSample : Number.NaN;
+      const head = Number.isFinite(headResult?.latency?.mean) ? headResult.latency.mean / operationsPerSample : Number.NaN;
+      const change = (head / base - 1) * 100;
+
+      if (!Number.isFinite(base) || base <= 0 || !Number.isFinite(head) || head <= 0 || !Number.isFinite(change)) {
+        skip('latency means and normalized values must be finite and positive');
+
+        return;
+      }
+
+      rows.push(
+        `| \`${file.slice(0, -5)}\` | ${base.toPrecision(4)} | ${head.toPrecision(4)} | ${change >= 0 ? '+' : ''}${change.toPrecision(3)}% |`,
+      );
+    }
+  } catch {
+    skip('native result JSON is missing, unreadable, or invalid');
+
+    return;
+  }
+
+  const marker = '<!-- benchmark-report -->';
+
+  const comments = await github.paginate(github.rest.issues.listComments, {
+    ...repository,
+    issue_number: source.pr,
+    per_page: 100,
+  });
+
+  const matching = comments.filter(
+    (comment) =>
+      comment.user?.type === 'Bot' && comment.user.login === 'github-actions[bot]' && comment.body?.startsWith(marker),
+  );
+
+  if (matching.length > 1) {
+    skip('multiple bot report comments exist');
+
+    return;
+  }
+
+  const comment = matching[0];
+
+  if (comment) {
+    const previous = comment.body.match(/<!-- benchmark-run:([a-f0-9]{40}):(\d+):(\d+):(\d+) -->/u);
+
+    if (!previous) {
+      skip('the existing report has no trusted run identity');
+
+      return;
+    }
+
+    if (
+      previous[1] === source.head &&
+      (Number(previous[3]) > source.number ||
+        (Number(previous[3]) === source.number &&
+          (Number(previous[2]) !== source.run || Number(previous[4]) >= source.attempt)))
+    ) {
+      skip('an equal or newer report already exists for this head');
+
+      return;
+    }
+  }
+
+  const url = `${context.serverUrl}/${repository.owner}/${repository.repo}`;
+
+  const body = [
+    marker,
+    `<!-- benchmark-run:${source.head}:${source.run}:${source.number}:${source.attempt} -->`,
+    '## Benchmark Comparison',
+    '',
+    `| Workload | Base, ${unit} | PR, ${unit} | Change |`,
+    '| --- | ---: | ---: | ---: |',
+    ...rows,
+    '',
+    `[Base ${source.base}](${url}/commit/${source.base}) · [PR ${source.head}](${url}/commit/${source.head}) · [Run ${source.number}, attempt ${source.attempt}](${url}/actions/runs/${source.run}/attempts/${source.attempt})`,
+  ].join('\n');
+
+  const { data: pr } = await github.rest.pulls.get({
+    ...repository,
+    pull_number: source.pr,
+  });
+
+  if (pr.state !== 'open' || pr.base.sha !== source.base || pr.head.sha !== source.head) {
+    skip('PR closed or base/head changed before publication');
+
+    return;
+  }
+
+  // GitHub has no compare-and-swap for comments. A PR update can still race this final check and write.
+  await (comment
+    ? github.rest.issues.updateComment({
+        ...repository,
+        comment_id: comment.id,
+        body,
+      })
+    : github.rest.issues.createComment({
+        ...repository,
+        issue_number: source.pr,
+        body,
+      }));
+
+  await core.summary.addRaw(body).write();
+}
+
+async function readJson(file) {
+  return JSON.parse(await readFile(file, 'utf8'));
+}

.github/workflows/benchmark.yml

@@ -15,6 +15,10 @@ jobs:
     name: "📊 compare benchmarks"
     runs-on: ubuntu-24.04
     timeout-minutes: 15
+    outputs:
+      artifact-id: ${{ steps.results.outputs.artifact-id }}
+      base: ${{ github.event.pull_request.base.sha }}
+      head: ${{ github.event.pull_request.head.sha }}
 
     steps:
       - name: "📥 check out head"
@@ -27,32 +31,46 @@ jobs:
       - name: "⚙️ set up node and pnpm"
         uses: devlsh/tools/github/setup@5a43d10d4535a7cf1368acd9313e70d417ac6235
 
-      - name: "📥 check out base"
+      - name: "📊 compare benchmarks"
+        id: results
+        uses: ./.github/actions/benchmark-capture
+        with:
+          base-repository: ${{ github.event.pull_request.base.repo.full_name }}
+          base-sha: ${{ github.event.pull_request.base.sha }}
+          head-sha: ${{ github.event.pull_request.head.sha }}
+          capture-command: pnpm bench:full --mode capture --reporter=default
+          compare-command: pnpm bench:full --mode compare --reporter=default --reporter=github-actions
+          base-results-path: .vitest/benchmarks/full
+          head-results-path: .vitest/benchmarks/full/current
+
+  report:
+    name: "📝 report benchmarks"
+    needs: benchmark
+    if: >-
+      github.event.pull_request.head.repo.full_name == github.repository &&
+      needs.benchmark.outputs.artifact-id != '' &&
+      needs.benchmark.outputs.base != '' &&
+      needs.benchmark.outputs.head != ''
+    runs-on: ubuntu-24.04
+    permissions:
+      contents: read
+      issues: write
+    steps:
+      - name: "📥 check out publisher"
         uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
-          repository: ${{ github.event.pull_request.base.repo.full_name }}
-          ref: ${{ github.event.pull_request.base.sha }}
-          path: benchmark-base
+          ref: ${{ github.event.pull_request.head.sha }}
           persist-credentials: false
+          sparse-checkout: |
+            .github/actions/benchmark-report
+            .github/scripts/benchmark-report.mjs
+          sparse-checkout-cone-mode: false
 
-      - name: "📥 install base dependencies"
-        working-directory: benchmark-base
-        run: pnpm install --frozen-lockfile
-
-      - name: "📦 build base"
-        working-directory: benchmark-base
-        run: pnpm build
-
-      - name: "📂 stage base output"
-        run: |
-          mkdir -p dist
-          cp -R benchmark-base/dist/. dist/
-
-      - name: "📊 capture base benchmark"
-        run: pnpm bench:full --mode capture --reporter=default
-
-      - name: "📦 build head"
-        run: pnpm build
-
-      - name: "📊 compare benchmark"
-        run: pnpm bench:full --mode compare --reporter=default --reporter=github-actions
+      - name: "📝 report benchmarks"
+        uses: ./.github/actions/benchmark-report
+        with:
+          artifact-id: ${{ needs.benchmark.outputs.artifact-id }}
+          base-sha: ${{ needs.benchmark.outputs.base }}
+          head-sha: ${{ needs.benchmark.outputs.head }}
+          unit: ms/search
+          operations-per-sample: "4"