From 8c070851f1a004fbd0079d81aee97707c0f6417f Mon Sep 17 00:00:00 2001
From: Alexey Romanov <alexey.v.romanov@gmail.com>
Date: Mon, 1 Jun 2026 16:03:51 +0300
Subject: [PATCH 1/8] Unify benchmark suites behind matrix config

---
 .../actions/setup-benchmark-runner/action.yml |  25 +-
 .github/workflows/benchmark.yml               | 476 ++++--------------
 benchmarks/count_benchmark_shards.rb          |  33 --
 benchmarks/generate_matrix.rb                 | 216 ++++++++
 4 files changed, 330 insertions(+), 420 deletions(-)
 delete mode 100755 benchmarks/count_benchmark_shards.rb
 create mode 100755 benchmarks/generate_matrix.rb

diff --git a/.github/actions/setup-benchmark-runner/action.yml b/.github/actions/setup-benchmark-runner/action.yml
index 0232838fee..ffc39bd363 100644
--- a/.github/actions/setup-benchmark-runner/action.yml
+++ b/.github/actions/setup-benchmark-runner/action.yml
@@ -2,10 +2,10 @@ name: Setup benchmark runner
 description: Install common tools and build packages needed by benchmark jobs.
 
 inputs:
-  install-k6:
-    description: Whether to install k6 for Rails benchmark jobs.
+  benchmark-tool:
+    description: Benchmark load tool to install. Supported values are k6, vegeta, and none.
     required: false
-    default: 'true'
+    default: k6
 
 runs:
   using: composite
@@ -30,11 +30,28 @@ runs:
       run: sudo apt-get install -y libyaml-dev
 
     - name: Setup k6
-      if: inputs.install-k6 == 'true'
+      if: inputs.benchmark-tool == 'k6'
       uses: grafana/setup-k6-action@v1
       with:
         k6-version: ${{ env.K6_VERSION }}
 
+    - name: Cache Vegeta binary
+      id: cache-vegeta
+      if: inputs.benchmark-tool == 'vegeta'
+      uses: actions/cache@v5
+      with:
+        path: ~/bin/vegeta
+        key: vegeta-${{ runner.os }}-${{ runner.arch }}-${{ env.VEGETA_VERSION }}
+
+    - name: Install Vegeta
+      if: inputs.benchmark-tool == 'vegeta' && steps.cache-vegeta.outputs.cache-hit != 'true'
+      shell: bash
+      run: |
+        echo "Installing Vegeta v${VEGETA_VERSION}"
+        wget -q "https://github.com/tsenart/vegeta/releases/download/v${VEGETA_VERSION}/vegeta_${VEGETA_VERSION}_linux_amd64.tar.gz"
+        tar -xzf "vegeta_${VEGETA_VERSION}_linux_amd64.tar.gz"
+        mv vegeta ~/bin/
+
     - name: Setup Ruby
       uses: ruby/setup-ruby@v1
       with:
diff --git a/.github/workflows/benchmark.yml b/.github/workflows/benchmark.yml
index 3e6e6de70e..afe42294c1 100644
--- a/.github/workflows/benchmark.yml
+++ b/.github/workflows/benchmark.yml
@@ -88,14 +88,14 @@ jobs:
     permissions:
       contents: read
       actions: read
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-24.04
     outputs:
       non_runtime_only: ${{ steps.detect.outputs.non_runtime_only }}
       run_core_benchmarks: ${{ steps.benchmark-outputs.outputs.run_core_benchmarks }}
       run_pro_benchmarks: ${{ steps.benchmark-outputs.outputs.run_pro_benchmarks }}
       run_pro_node_renderer_benchmarks: ${{ steps.benchmark-outputs.outputs.run_pro_node_renderer_benchmarks }}
-      core_benchmark_matrix: ${{ steps.benchmark-matrices.outputs.core_benchmark_matrix }}
-      pro_benchmark_matrix: ${{ steps.benchmark-matrices.outputs.pro_benchmark_matrix }}
+      run_benchmark_suites: ${{ steps.benchmark-matrices.outputs.run_benchmark_suites }}
+      benchmark_matrix: ${{ steps.benchmark-matrices.outputs.benchmark_matrix }}
     steps:
       - uses: actions/checkout@v6
         with:
@@ -147,47 +147,46 @@ jobs:
         # Always runs. GitHub Actions evaluates `strategy.matrix` for downstream jobs at
         # workflow-construction time, before their `if:` gates — so an empty output would
         # crash `fromJSON('')` even when the consuming job is otherwise skipped.
-        # Bump shard totals here when a suite's route set outgrows its current shard count.
+        # Bump shard totals in benchmarks/generate_matrix.rb when a suite's route set
+        # outgrows its current shard count.
         # The script is stdlib-only, so the runner's system Ruby is enough (no Bundler/Rails boot).
+        env:
+          BENCHMARK_APP_VERSION: ${{ github.event.inputs.app_version || 'both' }}
+          BENCHMARK_EVENT_NAME: ${{ github.event_name }}
+          BENCHMARK_NON_RUNTIME_ONLY: ${{ steps.detect.outputs.non_runtime_only }}
+          BENCHMARK_PULL_REQUEST_HEAD_REPO: ${{ github.event.pull_request.head.repo.full_name || '' }}
+          BENCHMARK_PULL_REQUEST_LABELS: ${{ github.event_name == 'pull_request' && toJSON(github.event.pull_request.labels.*.name) || '[]' }}
+          GITHUB_REPOSITORY: ${{ github.repository }}
+          RUN_CORE_BENCHMARKS: ${{ steps.benchmark-outputs.outputs.run_core_benchmarks || 'false' }}
+          RUN_PRO_BENCHMARKS: ${{ steps.benchmark-outputs.outputs.run_pro_benchmarks || 'false' }}
+          RUN_PRO_NODE_RENDERER_BENCHMARKS: ${{ steps.benchmark-outputs.outputs.run_pro_node_renderer_benchmarks || 'false' }}
         run: |
-          CORE_BENCHMARK_MATRIX=$(BENCHMARK_TOTAL_SHARDS=1 BENCHMARK_SUITE_NAME=Core BENCHMARK_SUITE_PREFIX=CORE ruby benchmarks/count_benchmark_shards.rb)
-          PRO_BENCHMARK_MATRIX=$(BENCHMARK_TOTAL_SHARDS=2 BENCHMARK_SUITE_NAME=Pro BENCHMARK_SUITE_PREFIX=PRO ruby benchmarks/count_benchmark_shards.rb)
+          BENCHMARK_MATRIX=$(ruby benchmarks/generate_matrix.rb)
+          export BENCHMARK_MATRIX
+          RUN_BENCHMARK_SUITES=$(ruby -rjson -e 'puts JSON.parse(ENV.fetch("BENCHMARK_MATRIX")).fetch("include").any? { |row| row.fetch("suite_id") != "none" }')
 
           {
-            echo "core_benchmark_matrix<<EOF"
-            echo "$CORE_BENCHMARK_MATRIX"
-            echo "EOF"
-            echo "pro_benchmark_matrix<<EOF"
-            echo "$PRO_BENCHMARK_MATRIX"
+            echo "run_benchmark_suites=$RUN_BENCHMARK_SUITES"
+            echo "benchmark_matrix<<EOF"
+            echo "$BENCHMARK_MATRIX"
             echo "EOF"
           } >> "$GITHUB_OUTPUT"
         shell: bash
 
-  benchmark-core:
+  benchmark:
     needs: detect-changes
     strategy:
       fail-fast: false
-      matrix: ${{ fromJSON(needs.detect-changes.outputs.core_benchmark_matrix) }}
-    # Run on: push to main, workflow_dispatch, or PRs with 'benchmark' / 'benchmark-core' label.
-    # The 'full-ci' label is intentionally excluded — it controls test workflows,
-    # not benchmarks. Use dedicated benchmark labels to trigger perf runs on PRs.
-    # See https://bencher.dev/docs/how-to/github-actions/#pull-requests for the extra pull_request condition
-    name: Core benchmarks (shard ${{ matrix.shard_label }})
-    if: |
-      contains(fromJSON('["both", "core_only"]'), github.event.inputs.app_version || 'both') &&
-      needs.detect-changes.outputs.non_runtime_only != 'true' &&
-      (
-        github.event_name == 'push' ||
-        needs.detect-changes.outputs.run_core_benchmarks == 'true' ||
-        github.event_name == 'workflow_dispatch' ||
-        (
-          github.event_name == 'pull_request' && github.event.pull_request.head.repo.full_name == github.repository &&
-          (
-            contains(github.event.pull_request.labels.*.name, 'benchmark') ||
-            contains(github.event.pull_request.labels.*.name, 'benchmark-core')
-          )
-        )
-      )
+      matrix: ${{ fromJSON(needs.detect-changes.outputs.benchmark_matrix) }}
+    # Which events/labels actually select suites is decided in
+    # benchmarks/generate_matrix.rb (suite_requested_by_event? / suite_selected_by_input?):
+    # push to main, workflow_dispatch, relevant test-impact changes, or same-repo PRs
+    # with suite-specific benchmark labels. The 'full-ci' label is intentionally
+    # excluded — it controls test workflows, not benchmarks. This job just runs
+    # whatever non-empty matrix that produces; the same-repo PR gating follows
+    # https://bencher.dev/docs/how-to/github-actions/#pull-requests.
+    name: ${{ matrix.job_name }}
+    if: needs.detect-changes.outputs.run_benchmark_suites == 'true'
     runs-on: ubuntu-latest
     permissions:
       contents: read
@@ -196,6 +195,7 @@ jobs:
       checks: write
     env:
       SECRET_KEY_BASE: 'dummy-secret-key-for-ci-testing-not-used-in-production'
+      REACT_ON_RAILS_PRO_LICENSE: ${{ secrets.REACT_ON_RAILS_PRO_LICENSE_V2 }}
       BENCHMARK_SHARD_INDEX: ${{ matrix.shard_index }}
       BENCHMARK_TOTAL_SHARDS: ${{ matrix.shard_total }}
       BENCHMARK_SHARD_LABEL: ${{ matrix.shard_label }}
@@ -208,18 +208,25 @@ jobs:
 
       - name: Setup benchmark runner
         uses: ./.github/actions/setup-benchmark-runner
+        with:
+          benchmark-tool: ${{ matrix.benchmark_tool }}
 
-      - name: Install Ruby Gems for Core dummy app
+      - name: Install Ruby Gems for the benchmarked app
         uses: ./.github/actions/setup-bundle
         with:
-          working-directory: react_on_rails/spec/dummy
+          working-directory: ${{ matrix.app_directory }}
           ruby-version: ${{ env.RUBY_VERSION }}
 
-      - name: Prepare Core production assets
+      - name: Generate file-system based entrypoints for Pro
+        if: matrix.generate_packs == 'true'
+        working-directory: ${{ matrix.app_directory }}
+        run: bundle exec rake react_on_rails:generate_packs
+
+      - name: Prepare production assets
+        working-directory: ${{ matrix.app_directory }}
         run: |
           set -e
-          echo "🔨 Building Core production assets..."
-          cd react_on_rails/spec/dummy
+          echo "🔨 Building ${{ matrix.suite_name }} production assets..."
 
           if ! bin/prod-assets; then
             echo "❌ ERROR: Failed to build production assets"
@@ -228,169 +235,42 @@ jobs:
 
           echo "✅ Production assets built successfully"
 
-      - name: Start Core production server
+      - name: Start Pro node renderer
+        if: matrix.server_kind == 'node-renderer'
+        working-directory: ${{ matrix.app_directory }}
+        env:
+          RAILS_ENV: production
+          NODE_ENV: production
+          RENDERER_LOG_LEVEL: error
+          RENDERER_PORT: 3800
         run: |
           set -e
-          echo "🚀 Starting Core production server..."
-          cd react_on_rails/spec/dummy
+          echo "🔧 Pre-seeding node renderer bundle cache..."
+          bundle exec rake react_on_rails_pro:pre_seed_renderer_cache MODE=symlink
 
-          $SERVER_CMD &
-          echo "Server started in background"
+          echo "🚀 Starting ${{ matrix.suite_name }} node renderer..."
+          node renderer/node-renderer.js &
+          echo "Node renderer started in background"
 
-          echo "⏳ Waiting for server to be ready..."
+          echo "⏳ Waiting for node renderer to be ready..."
           for i in {1..30}; do
-            if curl -fsS http://localhost:3001 > /dev/null; then
-              echo "✅ Server is ready and responding"
+            if ruby -e "require 'socket'; TCPSocket.new('localhost', 3800).close"; then
+              echo "✅ Node renderer is ready and listening"
               exit 0
             fi
-            echo "  Attempt $i/30: Server not ready yet..."
+            echo "  Attempt $i/30: Node renderer not ready yet..."
             sleep 1
           done
 
-          echo "❌ ERROR: Server failed to start within 30 seconds"
+          echo "❌ ERROR: Node renderer failed to start within 30 seconds"
           exit 1
 
-      - name: Execute Core benchmark suite
-        timeout-minutes: 120
-        run: |
-          set -e
-          echo "🏃 Running Core benchmark suite..."
-
-          if ! $BENCH_CMD benchmarks/bench.rb; then
-            echo "❌ ERROR: Benchmark execution failed"
-            exit 1
-          fi
-
-          echo "Benchmark suite completed successfully"
-
-      - name: Validate Core benchmark results
-        run: |
-          set -e
-          echo "🔍 Validating Core benchmark results..."
-
-          if [ ! -f "bench_results/summary.txt" ]; then
-            echo "❌ ERROR: benchmark summary file not found"
-            exit 1
-          fi
-
-          if [ ! -f "bench_results/benchmark.json" ]; then
-            echo "❌ ERROR: benchmark JSON file not found"
-            exit 1
-          fi
-
-          echo "✅ Benchmark results found"
-          echo ""
-          echo "📊 Summary:"
-          column -t -s $'\t' "bench_results/summary.txt"
-          echo ""
-          echo "Generated files:"
-          ls -lh bench_results/
-
-      - name: Upload Core benchmark results
-        uses: actions/upload-artifact@v7
-        if: always()
-        with:
-          name: benchmark-core-results-${{ github.run_number }}-${{ matrix.shard_slug }}
-          path: bench_results/
-          retention-days: 30
-          if-no-files-found: warn
-
-      - name: Track Core benchmarks with Bencher
-        env:
-          BENCHER_API_TOKEN: ${{ secrets.BENCHER_API_TOKEN }}
-          BENCHER_REPORT_MARKER: ${{ matrix.report_marker }}
-          BENCHMARK_SUITE_NAME: ${{ matrix.suite_name }}
-          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          GITHUB_BASE_SHA: ${{ github.event.pull_request.base.sha || '' }}
-          PR_NUMBER: ${{ github.event.pull_request.number || '' }}
-        run: ruby benchmarks/track_benchmarks.rb
-
-      - name: Core workflow summary
-        if: always()
-        run: |
-          echo "📋 Benchmark Workflow Summary"
-          echo "===================================="
-          echo "Suite: Core"
-          echo "Shard: ${{ matrix.shard_label }}"
-          echo "Status: ${{ job.status }}"
-          echo "Run number: ${{ github.run_number }}"
-          echo "Triggered by: ${{ github.actor }}"
-          echo "Branch: ${{ github.ref_name }}"
-
-  benchmark-pro:
-    needs: detect-changes
-    strategy:
-      fail-fast: false
-      matrix: ${{ fromJSON(needs.detect-changes.outputs.pro_benchmark_matrix) }}
-    # Run on: push to main, workflow_dispatch, or PRs with 'benchmark' / 'benchmark-pro' label.
-    # The 'full-ci' label is intentionally excluded; use the dedicated
-    # benchmark labels to trigger perf runs on PRs.
-    name: Pro benchmarks (shard ${{ matrix.shard_label }})
-    if: |
-      contains(fromJSON('["both", "pro_only", "pro_rails_only"]'), github.event.inputs.app_version || 'both') &&
-      needs.detect-changes.outputs.non_runtime_only != 'true' &&
-      (
-        github.event_name == 'push' ||
-        needs.detect-changes.outputs.run_pro_benchmarks == 'true' ||
-        github.event_name == 'workflow_dispatch' ||
-        (
-          github.event_name == 'pull_request' && github.event.pull_request.head.repo.full_name == github.repository &&
-          (
-            contains(github.event.pull_request.labels.*.name, 'benchmark') ||
-            contains(github.event.pull_request.labels.*.name, 'benchmark-pro')
-          )
-        )
-      )
-    runs-on: ubuntu-latest
-    permissions:
-      contents: read
-      issues: write
-      pull-requests: write
-      checks: write
-    env:
-      SECRET_KEY_BASE: 'dummy-secret-key-for-ci-testing-not-used-in-production'
-      REACT_ON_RAILS_PRO_LICENSE: ${{ secrets.REACT_ON_RAILS_PRO_LICENSE_V2 }}
-      BENCHMARK_SHARD_INDEX: ${{ matrix.shard_index }}
-      BENCHMARK_TOTAL_SHARDS: ${{ matrix.shard_total }}
-      BENCHMARK_SHARD_LABEL: ${{ matrix.shard_label }}
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v6
-        with:
-          persist-credentials: false
-
-      - name: Setup benchmark runner
-        uses: ./.github/actions/setup-benchmark-runner
-
-      - name: Install Ruby Gems for Pro dummy app
-        uses: ./.github/actions/setup-bundle
-        with:
-          working-directory: react_on_rails_pro/spec/dummy
-          ruby-version: ${{ env.RUBY_VERSION }}
-
-      - name: Generate file-system based entrypoints for Pro
-        run: cd react_on_rails_pro/spec/dummy && bundle exec rake react_on_rails:generate_packs
-
-      - name: Prepare Pro production assets
-        run: |
-          set -e
-          echo "🔨 Building Pro production assets..."
-          cd react_on_rails_pro/spec/dummy
-
-          if ! bin/prod-assets; then
-            echo "❌ ERROR: Failed to build production assets"
-            exit 1
-          fi
-
-          echo "✅ Production assets built successfully"
-
-      - name: Start Pro production server
+      - name: Start Rails production server
+        if: matrix.server_kind == 'rails'
+        working-directory: ${{ matrix.app_directory }}
         run: |
           set -e
-          echo "🚀 Starting Pro production server..."
-          cd react_on_rails_pro/spec/dummy
-
+          echo "🚀 Starting ${{ matrix.suite_name }} production server..."
           $SERVER_CMD &
           echo "Server started in background"
 
@@ -407,30 +287,39 @@ jobs:
           echo "❌ ERROR: Server failed to start within 30 seconds"
           exit 1
 
-      - name: Execute Pro benchmark suite
-        timeout-minutes: 120
+      - name: Execute benchmark suite
+        timeout-minutes: ${{ matrix.benchmark_timeout_minutes }}
         run: |
           set -e
-          echo "🏃 Running Pro benchmark suite..."
+          echo "🏃 Running ${{ matrix.suite_name }} benchmark suite..."
 
-          if ! PRO=true $BENCH_CMD benchmarks/bench.rb; then
+          if [ "${{ matrix.pro_env }}" = "true" ]; then
+            if ! PRO=true $BENCH_CMD "${{ matrix.benchmark_script }}"; then
+              echo "❌ ERROR: Benchmark execution failed"
+              exit 1
+            fi
+          elif ! $BENCH_CMD "${{ matrix.benchmark_script }}"; then
             echo "❌ ERROR: Benchmark execution failed"
             exit 1
           fi
 
-          echo "Benchmark suite completed successfully"
+          echo "✅ Benchmark suite completed successfully"
 
-      - name: Validate Pro benchmark results
+      - name: Validate benchmark results
+        env:
+          SUITE_NAME: ${{ matrix.suite_name }}
+          SUMMARY_FILE: ${{ matrix.summary_file }}
+          SUMMARY_TITLE: ${{ matrix.summary_title }}
         run: |
           set -e
-          echo "🔍 Validating Pro benchmark results..."
+          echo "🔍 Validating $SUITE_NAME benchmark results..."
 
-          if [ ! -f "bench_results/summary.txt" ]; then
-            echo "❌ ERROR: Rails benchmark summary file not found"
+          if [ ! -f "$SUMMARY_FILE" ]; then
+            echo "❌ ERROR: benchmark summary file not found"
             exit 1
           fi
-          echo "📊 Rails Benchmark Summary:"
-          column -t -s $'\t' "bench_results/summary.txt"
+          echo "📊 $SUMMARY_TITLE:"
+          column -t -s $'\t' "$SUMMARY_FILE"
           echo ""
 
           if [ ! -f "bench_results/benchmark.json" ]; then
@@ -443,212 +332,33 @@ jobs:
           echo "Generated files:"
           ls -lh bench_results/
 
-      - name: Upload Pro benchmark results
+      - name: Upload benchmark results
         uses: actions/upload-artifact@v7
         if: always()
         with:
-          name: benchmark-pro-results-${{ github.run_number }}-${{ matrix.shard_slug }}
+          name: ${{ matrix.artifact_name_prefix }}-${{ github.run_number }}${{ matrix.artifact_name_suffix }}
           path: bench_results/
           retention-days: 30
           if-no-files-found: warn
 
-      - name: Track Pro benchmarks with Bencher
+      - name: Track benchmarks with Bencher
         env:
           BENCHER_API_TOKEN: ${{ secrets.BENCHER_API_TOKEN }}
           BENCHER_REPORT_MARKER: ${{ matrix.report_marker }}
-          BENCHMARK_SUITE_NAME: ${{ matrix.suite_name }}
+          BENCHMARK_SUITE_NAME: ${{ matrix.bencher_suite_name }}
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           GITHUB_BASE_SHA: ${{ github.event.pull_request.base.sha || '' }}
           PR_NUMBER: ${{ github.event.pull_request.number || '' }}
         run: ruby benchmarks/track_benchmarks.rb
 
-      - name: Pro workflow summary
+      - name: Benchmark workflow summary
         if: always()
         run: |
           echo "📋 Benchmark Workflow Summary"
           echo "===================================="
-          echo "Suite: Pro"
+          echo "Suite: ${{ matrix.suite_name }}"
           echo "Shard: ${{ matrix.shard_label }}"
           echo "Status: ${{ job.status }}"
           echo "Run number: ${{ github.run_number }}"
           echo "Triggered by: ${{ github.actor }}"
           echo "Branch: ${{ github.ref_name }}"
-
-  benchmark-pro-node-renderer:
-    needs: detect-changes
-    # Run on: push to main, workflow_dispatch, or PRs with 'benchmark', 'benchmark-pro',
-    # or 'benchmark-pro-node-renderer' label.
-    # The 'full-ci' label is intentionally excluded; use the dedicated
-    # benchmark labels to trigger perf runs on PRs.
-    if: |
-      contains(fromJSON('["both", "pro_only", "pro_node_renderer_only"]'), github.event.inputs.app_version || 'both') &&
-      needs.detect-changes.outputs.non_runtime_only != 'true' &&
-      (
-        github.event_name == 'push' ||
-        needs.detect-changes.outputs.run_pro_node_renderer_benchmarks == 'true' ||
-        github.event_name == 'workflow_dispatch' ||
-        (
-          github.event_name == 'pull_request' && github.event.pull_request.head.repo.full_name == github.repository &&
-          (
-            contains(github.event.pull_request.labels.*.name, 'benchmark') ||
-            contains(github.event.pull_request.labels.*.name, 'benchmark-pro') ||
-            contains(github.event.pull_request.labels.*.name, 'benchmark-pro-node-renderer')
-          )
-        )
-      )
-    runs-on: ubuntu-latest
-    permissions:
-      contents: read
-      issues: write
-      pull-requests: write
-      checks: write
-    env:
-      SECRET_KEY_BASE: 'dummy-secret-key-for-ci-testing-not-used-in-production'
-      REACT_ON_RAILS_PRO_LICENSE: ${{ secrets.REACT_ON_RAILS_PRO_LICENSE_V2 }}
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v6
-        with:
-          persist-credentials: false
-
-      - name: Setup benchmark runner
-        uses: ./.github/actions/setup-benchmark-runner
-        with:
-          install-k6: false
-
-      - name: Cache Vegeta binary
-        id: cache-vegeta
-        uses: actions/cache@v5
-        with:
-          path: ~/bin/vegeta
-          key: vegeta-${{ runner.os }}-${{ runner.arch }}-${{ env.VEGETA_VERSION }}
-
-      - name: Install Vegeta
-        if: steps.cache-vegeta.outputs.cache-hit != 'true'
-        run: |
-          echo "Installing Vegeta v${VEGETA_VERSION}"
-          wget -q "https://github.com/tsenart/vegeta/releases/download/v${VEGETA_VERSION}/vegeta_${VEGETA_VERSION}_linux_amd64.tar.gz"
-          tar -xzf "vegeta_${VEGETA_VERSION}_linux_amd64.tar.gz"
-          mv vegeta ~/bin/
-
-      - name: Install Ruby Gems for Pro dummy app
-        uses: ./.github/actions/setup-bundle
-        with:
-          working-directory: react_on_rails_pro/spec/dummy
-          ruby-version: ${{ env.RUBY_VERSION }}
-
-      - name: Generate file-system based entrypoints for Pro
-        run: cd react_on_rails_pro/spec/dummy && bundle exec rake react_on_rails:generate_packs
-
-      - name: Prepare Pro production assets
-        run: |
-          set -e
-          echo "🔨 Building Pro production assets..."
-          cd react_on_rails_pro/spec/dummy
-
-          if ! bin/prod-assets; then
-            echo "❌ ERROR: Failed to build production assets"
-            exit 1
-          fi
-
-          echo "✅ Production assets built successfully"
-
-      - name: Pre-seed Node Renderer bundle cache
-        # Stages the just-built bundles into `.node-renderer-bundles/<hash>-production/`
-        # so bench-node-renderer.rb finds them at startup instead of waiting for the
-        # first SSR request to upload them.
-        run: |
-          set -e
-          echo "🔧 Pre-seeding node renderer bundle cache..."
-          cd react_on_rails_pro/spec/dummy
-          RAILS_ENV=production NODE_ENV=production \
-            bundle exec rake react_on_rails_pro:pre_seed_renderer_cache MODE=symlink
-
-      - name: Start Pro node renderer
-        run: |
-          set -e
-          echo "🚀 Starting Pro node renderer..."
-          cd react_on_rails_pro/spec/dummy
-
-          RENDERER_LOG_LEVEL=error RENDERER_PORT=3800 RAILS_ENV=production NODE_ENV=production node renderer/node-renderer.js &
-          echo "Node renderer started in background"
-
-          echo "⏳ Waiting for node renderer to be ready..."
-          for i in {1..30}; do
-            if ruby -e "require 'socket'; TCPSocket.new('localhost', 3800).close"; then
-              echo "✅ Node renderer is ready and listening"
-              exit 0
-            fi
-            echo "  Attempt $i/30: Node renderer not ready yet..."
-            sleep 1
-          done
-
-          echo "❌ ERROR: Node renderer failed to start within 30 seconds"
-          exit 1
-
-      - name: Execute Pro Node Renderer benchmark suite
-        timeout-minutes: 30
-        run: |
-          set -e
-          echo "🏃 Running Pro Node Renderer benchmark suite..."
-
-          if ! $BENCH_CMD benchmarks/bench-node-renderer.rb; then
-            echo "❌ ERROR: Node Renderer benchmark execution failed"
-            exit 1
-          fi
-
-          echo "✅ Node Renderer benchmark suite completed successfully"
-
-      - name: Validate Pro Node Renderer benchmark results
-        run: |
-          set -e
-          echo "🔍 Validating Pro Node Renderer benchmark results..."
-
-          if [ ! -f "bench_results/node_renderer_summary.txt" ]; then
-            echo "❌ ERROR: Node Renderer benchmark summary file not found"
-            exit 1
-          fi
-          echo "📊 Node Renderer Benchmark Summary:"
-          column -t -s $'\t' "bench_results/node_renderer_summary.txt"
-          echo ""
-
-          if [ ! -f "bench_results/benchmark.json" ]; then
-            echo "❌ ERROR: benchmark JSON file not found"
-            exit 1
-          fi
-
-          echo "✅ Benchmark results validated"
-          echo ""
-          echo "Generated files:"
-          ls -lh bench_results/
-
-      - name: Upload Pro Node Renderer benchmark results
-        uses: actions/upload-artifact@v7
-        if: always()
-        with:
-          name: benchmark-pro-node-renderer-results-${{ github.run_number }}
-          path: bench_results/
-          retention-days: 30
-          if-no-files-found: warn
-
-      - name: Track Pro Node Renderer benchmarks with Bencher
-        env:
-          BENCHER_API_TOKEN: ${{ secrets.BENCHER_API_TOKEN }}
-          BENCHER_REPORT_MARKER: '<!-- BENCHER_REPORT_PRO_NODE_RENDERER -->'
-          BENCHMARK_SUITE_NAME: Pro Node Renderer
-          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          GITHUB_BASE_SHA: ${{ github.event.pull_request.base.sha || '' }}
-          PR_NUMBER: ${{ github.event.pull_request.number || '' }}
-        run: ruby benchmarks/track_benchmarks.rb
-
-      - name: Pro Node Renderer workflow summary
-        if: always()
-        run: |
-          echo "📋 Benchmark Workflow Summary"
-          echo "===================================="
-          echo "Suite: Pro Node Renderer"
-          echo "Status: ${{ job.status }}"
-          echo "Run number: ${{ github.run_number }}"
-          echo "Triggered by: ${{ github.actor }}"
-          echo "Branch: ${{ github.ref_name }}"
diff --git a/benchmarks/count_benchmark_shards.rb b/benchmarks/count_benchmark_shards.rb
deleted file mode 100755
index 0ccd0bd3ae..0000000000
--- a/benchmarks/count_benchmark_shards.rb
+++ /dev/null
@@ -1,33 +0,0 @@
-#!/usr/bin/env ruby
-# frozen_string_literal: true
-
-# Emits the GitHub Actions matrix `include:` payload for one benchmark suite.
-# Stdlib-only so the workflow can invoke it with the runner's system Ruby — no
-# Bundler, no Rails boot.
-
-require "json"
-
-SHARD_TOTAL = Integer(ENV.fetch("BENCHMARK_TOTAL_SHARDS"))
-SUITE_NAME = ENV.fetch("BENCHMARK_SUITE_NAME")
-SUITE_PREFIX = ENV.fetch("BENCHMARK_SUITE_PREFIX")
-
-raise "BENCHMARK_TOTAL_SHARDS must be positive (got #{SHARD_TOTAL})" unless SHARD_TOTAL.positive?
-
-matrix = {
-  include: Array.new(SHARD_TOTAL) do |index|
-    shard_number = index + 1
-    shard_label = "#{shard_number}/#{SHARD_TOTAL}"
-    shard_slug = "shard-#{shard_number}-of-#{SHARD_TOTAL}"
-
-    {
-      shard_index: index,
-      shard_total: SHARD_TOTAL,
-      shard_label: shard_label,
-      shard_slug: shard_slug,
-      suite_name: "#{SUITE_NAME} (shard #{shard_label})",
-      report_marker: "<!-- BENCHER_REPORT_#{SUITE_PREFIX}_#{shard_slug.upcase.tr('-', '_')} -->"
-    }
-  end
-}
-
-puts JSON.generate(matrix)
diff --git a/benchmarks/generate_matrix.rb b/benchmarks/generate_matrix.rb
new file mode 100755
index 0000000000..c6edfa39e6
--- /dev/null
+++ b/benchmarks/generate_matrix.rb
@@ -0,0 +1,216 @@
+#!/usr/bin/env ruby
+# frozen_string_literal: true
+
+# Emits the GitHub Actions matrix `include:` payload for benchmark suites.
+# Stdlib-only so the workflow can invoke it with the runner's system Ruby — no
+# Bundler, no Rails boot.
+
+require "json"
+
+SUITES = [
+  {
+    id: "core",
+    suite_name: "Core",
+    suite_prefix: "CORE",
+    shard_total: 1,
+    app_versions: %w[both core_only],
+    run_output: "RUN_CORE_BENCHMARKS",
+    labels: %w[benchmark benchmark-core],
+    app_directory: "react_on_rails/spec/dummy",
+    artifact_name: "benchmark-core-results",
+    benchmark_tool: "k6",
+    benchmark_script: "benchmarks/bench.rb",
+    benchmark_timeout_minutes: 120,
+    pro_env: false,
+    generate_packs: false,
+    server_kind: "rails",
+    summary_file: "bench_results/summary.txt",
+    summary_title: "Summary"
+  },
+  {
+    id: "pro",
+    suite_name: "Pro",
+    suite_prefix: "PRO",
+    shard_total: 2,
+    app_versions: %w[both pro_only pro_rails_only],
+    run_output: "RUN_PRO_BENCHMARKS",
+    labels: %w[benchmark benchmark-pro],
+    app_directory: "react_on_rails_pro/spec/dummy",
+    artifact_name: "benchmark-pro-results",
+    benchmark_tool: "k6",
+    benchmark_script: "benchmarks/bench.rb",
+    benchmark_timeout_minutes: 120,
+    pro_env: true,
+    generate_packs: true,
+    server_kind: "rails",
+    summary_file: "bench_results/summary.txt",
+    summary_title: "Rails Benchmark Summary"
+  },
+  {
+    id: "pro-node-renderer",
+    suite_name: "Pro Node Renderer",
+    suite_prefix: "PRO_NODE_RENDERER",
+    shard_total: 1,
+    app_versions: %w[both pro_only pro_node_renderer_only],
+    run_output: "RUN_PRO_NODE_RENDERER_BENCHMARKS",
+    labels: %w[benchmark benchmark-pro benchmark-pro-node-renderer],
+    app_directory: "react_on_rails_pro/spec/dummy",
+    artifact_name: "benchmark-pro-node-renderer-results",
+    benchmark_tool: "vegeta",
+    report_marker: "<!-- BENCHER_REPORT_PRO_NODE_RENDERER -->",
+    benchmark_script: "benchmarks/bench-node-renderer.rb",
+    benchmark_timeout_minutes: 30,
+    pro_env: false,
+    generate_packs: true,
+    server_kind: "node-renderer",
+    summary_file: "bench_results/node_renderer_summary.txt",
+    summary_title: "Node Renderer Benchmark Summary"
+  }
+].freeze
+
+EMPTY_MATRIX_ROW = {
+  suite_id: "none",
+  suite_name: "No benchmark suite",
+  job_name: "Benchmark suites skipped",
+  shard_index: 0,
+  shard_total: 1,
+  shard_label: "1/1",
+  shard_slug: "skipped",
+  bencher_suite_name: "No benchmark suite",
+  report_marker: "<!-- BENCHER_REPORT_SKIPPED -->",
+  app_directory: ".",
+  artifact_name_prefix: "benchmark-skipped",
+  artifact_name_suffix: "",
+  benchmark_tool: "none",
+  benchmark_script: "",
+  benchmark_timeout_minutes: 1,
+  pro_env: false,
+  generate_packs: false,
+  server_kind: "none",
+  summary_file: "",
+  summary_title: ""
+}.freeze
+
+# Every app_version any suite accepts. The workflow_dispatch input is constrained
+# to these, but validating guards against a typo or future rename silently
+# selecting no suites (which would skip benchmarks while CI stays green).
+VALID_APP_VERSIONS = SUITES.flat_map { |suite| suite.fetch(:app_versions) }.uniq.freeze
+
+def truthy_env?(name)
+  ENV.fetch(name, "false") == "true"
+end
+
+def pull_request_labels
+  raw_labels = ENV.fetch("BENCHMARK_PULL_REQUEST_LABELS", "[]")
+  return [] if raw_labels.empty?
+
+  parsed = JSON.parse(raw_labels)
+  parsed.is_a?(Array) ? parsed : []
+rescue JSON::ParserError => e
+  raise "BENCHMARK_PULL_REQUEST_LABELS must be JSON array: #{e.message}"
+end
+
+def suite_requested_by_event?(suite, labels)
+  event_name = ENV.fetch("BENCHMARK_EVENT_NAME")
+
+  return true if event_name == "push"
+  return true if event_name == "workflow_dispatch"
+  return false unless event_name == "pull_request"
+
+  # Fork PRs never run: only honor the change-detection (run_output) or label
+  # triggers when the PR's head repo is this repo. GITHUB_REPOSITORY is always set
+  # by Actions, so a blank/forked head repo can't match it.
+  return false unless ENV.fetch("BENCHMARK_PULL_REQUEST_HEAD_REPO", "") == ENV.fetch("GITHUB_REPOSITORY")
+
+  truthy_env?(suite.fetch(:run_output)) || suite.fetch(:labels).intersect?(labels)
+end
+
+def suite_selected_by_input?(suite)
+  app_version = ENV.fetch("BENCHMARK_APP_VERSION", "both")
+  unless VALID_APP_VERSIONS.include?(app_version)
+    raise "BENCHMARK_APP_VERSION must be one of #{VALID_APP_VERSIONS.join(', ')} (got #{app_version.inspect})"
+  end
+
+  suite.fetch(:app_versions).include?(app_version)
+end
+
+def suite_enabled?(suite, labels)
+  suite_selected_by_input?(suite) && suite_requested_by_event?(suite, labels)
+end
+
+def suite_rows(suite)
+  shard_total = suite.fetch(:shard_total)
+  raise "#{suite.fetch(:id)} shard_total must be positive (got #{shard_total})" unless shard_total.positive?
+
+  Array.new(shard_total) do |index|
+    shard_number = index + 1
+    shard_label = "#{shard_number}/#{shard_total}"
+    shard_slug = "shard-#{shard_number}-of-#{shard_total}"
+    suite_name = suite.fetch(:suite_name)
+    suite_prefix = suite.fetch(:suite_prefix)
+
+    suite_row(suite, suite_name, suite_prefix,
+              shard_index: index,
+              shard_total: shard_total,
+              shard_label: shard_label,
+              shard_slug: shard_slug)
+  end
+end
+
+def suite_row(suite, suite_name, suite_prefix, shard_index:, shard_total:, shard_label:, shard_slug:)
+  {
+    suite_id: suite.fetch(:id),
+    suite_name: suite_name,
+    job_name: benchmark_job_name(suite_name, shard_total, shard_label),
+    shard_index: shard_index,
+    shard_total: shard_total,
+    shard_label: shard_label,
+    shard_slug: shard_slug,
+    bencher_suite_name: bencher_suite_name(suite_name, shard_total, shard_label),
+    report_marker: report_marker(suite, suite_prefix, shard_slug),
+    app_directory: suite.fetch(:app_directory),
+    artifact_name_prefix: suite.fetch(:artifact_name),
+    artifact_name_suffix: artifact_name_suffix(shard_total, shard_slug),
+    benchmark_tool: suite.fetch(:benchmark_tool),
+    benchmark_script: suite.fetch(:benchmark_script),
+    benchmark_timeout_minutes: suite.fetch(:benchmark_timeout_minutes),
+    pro_env: suite.fetch(:pro_env).to_s,
+    generate_packs: suite.fetch(:generate_packs).to_s,
+    server_kind: suite.fetch(:server_kind),
+    summary_file: suite.fetch(:summary_file),
+    summary_title: suite.fetch(:summary_title)
+  }
+end
+
+def benchmark_job_name(suite_name, shard_total, shard_label)
+  return "#{suite_name} benchmarks" if shard_total == 1
+
+  "#{suite_name} benchmarks (shard #{shard_label})"
+end
+
+def bencher_suite_name(suite_name, shard_total, shard_label)
+  return suite_name if shard_total == 1
+
+  "#{suite_name} (shard #{shard_label})"
+end
+
+def artifact_name_suffix(shard_total, shard_slug)
+  return "" if shard_total == 1
+
+  "-#{shard_slug}"
+end
+
+def report_marker(suite, suite_prefix, shard_slug)
+  suite[:report_marker] || "<!-- BENCHER_REPORT_#{suite_prefix}_#{shard_slug.upcase.tr('-', '_')} -->"
+end
+
+labels = pull_request_labels
+rows = if truthy_env?("BENCHMARK_NON_RUNTIME_ONLY")
+         []
+       else
+         SUITES.select { |suite| suite_enabled?(suite, labels) }.flat_map { |suite| suite_rows(suite) }
+       end
+
+matrix = { include: rows.empty? ? [EMPTY_MATRIX_ROW] : rows }
+
+puts JSON.generate(matrix)

From 0043ddfc19694f9fcbb01f5c5f2beb0496c8ecd0 Mon Sep 17 00:00:00 2001
From: Alexey Romanov <alexey.v.romanov@gmail.com>
Date: Mon, 1 Jun 2026 16:45:12 +0300
Subject: [PATCH 2/8] Inline setup-benchmark-runner

---
 .../actions/setup-benchmark-runner/action.yml | 142 ------------------
 .github/workflows/benchmark.yml               | 128 +++++++++++++++-
 2 files changed, 123 insertions(+), 147 deletions(-)
 delete mode 100644 .github/actions/setup-benchmark-runner/action.yml

diff --git a/.github/actions/setup-benchmark-runner/action.yml b/.github/actions/setup-benchmark-runner/action.yml
deleted file mode 100644
index ffc39bd363..0000000000
--- a/.github/actions/setup-benchmark-runner/action.yml
+++ /dev/null
@@ -1,142 +0,0 @@
-name: Setup benchmark runner
-description: Install common tools and build packages needed by benchmark jobs.
-
-inputs:
-  benchmark-tool:
-    description: Benchmark load tool to install. Supported values are k6, vegeta, and none.
-    required: false
-    default: k6
-
-runs:
-  using: composite
-  steps:
-    - name: Install Bencher CLI
-      # Pinned: alert-detection heuristics grep stderr for "Alert[s]",
-      # "threshold violation", and "boundary violation". Those strings
-      # are not a documented API contract, so upgrading the CLI requires
-      # re-verifying that the expected phrases still appear in alert output.
-      uses: bencherdev/bencher@v0.6.2
-
-    - name: Add tools directory to PATH
-      shell: bash
-      run: |
-        mkdir -p ~/bin
-        echo "$HOME/bin" >> "$GITHUB_PATH"
-
-    - name: Install libyaml-dev
-      # Must precede any gem install/load: psych links libyaml dynamically. On warm caches the
-      # missing-lib failure is silent; cold runs hit it.
-      shell: bash
-      run: sudo apt-get install -y libyaml-dev
-
-    - name: Setup k6
-      if: inputs.benchmark-tool == 'k6'
-      uses: grafana/setup-k6-action@v1
-      with:
-        k6-version: ${{ env.K6_VERSION }}
-
-    - name: Cache Vegeta binary
-      id: cache-vegeta
-      if: inputs.benchmark-tool == 'vegeta'
-      uses: actions/cache@v5
-      with:
-        path: ~/bin/vegeta
-        key: vegeta-${{ runner.os }}-${{ runner.arch }}-${{ env.VEGETA_VERSION }}
-
-    - name: Install Vegeta
-      if: inputs.benchmark-tool == 'vegeta' && steps.cache-vegeta.outputs.cache-hit != 'true'
-      shell: bash
-      run: |
-        echo "Installing Vegeta v${VEGETA_VERSION}"
-        wget -q "https://github.com/tsenart/vegeta/releases/download/v${VEGETA_VERSION}/vegeta_${VEGETA_VERSION}_linux_amd64.tar.gz"
-        tar -xzf "vegeta_${VEGETA_VERSION}_linux_amd64.tar.gz"
-        mv vegeta ~/bin/
-
-    - name: Setup Ruby
-      uses: ruby/setup-ruby@v1
-      with:
-        ruby-version: ${{ env.RUBY_VERSION }}
-
-    - name: Get gem home directory
-      shell: bash
-      run: echo "GEM_HOME_PATH=$(gem env home)" >> "$GITHUB_ENV"
-
-    - name: Cache foreman gem
-      id: cache-foreman
-      uses: actions/cache@v5
-      with:
-        path: ${{ env.GEM_HOME_PATH }}
-        key: foreman-gem-${{ runner.os }}-ruby-${{ env.RUBY_VERSION }}
-
-    - name: Install foreman
-      if: steps.cache-foreman.outputs.cache-hit != 'true'
-      shell: bash
-      run: gem install foreman
-
-    - name: Setup pnpm
-      uses: pnpm/action-setup@v6
-      with:
-        cache: true
-        cache_dependency_path: '**/pnpm-lock.yaml'
-        run_install: false
-
-    - name: Setup Node
-      uses: actions/setup-node@v6
-      with:
-        node-version: '22'
-
-    - name: Print system information
-      shell: bash
-      run: |
-        echo "Linux release: "
-        cat /etc/issue
-        echo "Current user: "
-        whoami
-        echo "Current directory: "
-        pwd
-        echo "Ruby version: "
-        ruby -v
-        echo "Node version: "
-        node -v
-        echo "Pnpm version: "
-        pnpm --version
-        echo "Bundler version: "
-        bundle --version
-
-    - name: Configure CPU pinning and process priority
-      shell: bash
-      run: |
-        NPROC=$(nproc)
-        echo "Available CPUs: $NPROC"
-
-        if [ "$NPROC" -le 1 ]; then
-          SERVER_CMD="bin/prod"
-          BENCH_CMD="ruby"
-        else
-          SERVER_CMD="nice -n -20 taskset -c 1-$((NPROC-1)) bin/prod"
-          BENCH_CMD="nice -n -10 taskset -c 0 ruby"
-        fi
-
-        echo "SERVER_CMD=$SERVER_CMD" >> "$GITHUB_ENV"
-        echo "BENCH_CMD=$BENCH_CMD" >> "$GITHUB_ENV"
-
-        if [ -z "$WEB_CONCURRENCY" ]; then
-          # Clamp to >=1 so single-CPU runners don't end up with WEB_CONCURRENCY=0 (Puma
-          # would start in single-mode, which the benchmarks aren't tuned for).
-          WEB_CONCURRENCY=$((NPROC > 1 ? NPROC - 1 : 1))
-          echo "WEB_CONCURRENCY=$WEB_CONCURRENCY" >> "$GITHUB_ENV"
-          echo "WEB_CONCURRENCY (auto): $WEB_CONCURRENCY"
-        else
-          echo "WEB_CONCURRENCY (from input): $WEB_CONCURRENCY"
-        fi
-
-        echo "SERVER_CMD: $SERVER_CMD"
-        echo "BENCH_CMD: $BENCH_CMD"
-
-    - name: Install Node modules with pnpm
-      shell: bash
-      run: pnpm install --frozen-lockfile
-
-    - name: Build workspace packages
-      shell: bash
-      run: pnpm run build
diff --git a/.github/workflows/benchmark.yml b/.github/workflows/benchmark.yml
index afe42294c1..7824f8cd4e 100644
--- a/.github/workflows/benchmark.yml
+++ b/.github/workflows/benchmark.yml
@@ -78,7 +78,7 @@ env:
   REQUEST_TIMEOUT: ${{ github.event.inputs.request_timeout }}
   CONNECTIONS: ${{ github.event.inputs.connections }}
   MAX_CONNECTIONS: ${{ github.event.inputs.connections }}
-  # WEB_CONCURRENCY default is set dynamically to NPROC-1 in "Configure CPU pinning" step
+  # WEB_CONCURRENCY default is set dynamically to NPROC-1 in "Configure benchmark commands" step
   WEB_CONCURRENCY: ${{ github.event.inputs.web_concurrency }}
   RAILS_MAX_THREADS: ${{ github.event.inputs.rails_threads || 3 }}
   RAILS_MIN_THREADS: ${{ github.event.inputs.rails_threads || 3 }}
@@ -187,7 +187,7 @@ jobs:
     # https://bencher.dev/docs/how-to/github-actions/#pull-requests.
     name: ${{ matrix.job_name }}
     if: needs.detect-changes.outputs.run_benchmark_suites == 'true'
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-24.04
     permissions:
       contents: read
       issues: write
@@ -206,10 +206,128 @@ jobs:
         with:
           persist-credentials: false
 
-      - name: Setup benchmark runner
-        uses: ./.github/actions/setup-benchmark-runner
+      - name: Install Bencher CLI
+        # Pinned: alert-detection heuristics grep stderr for "Alert[s]",
+        # "threshold violation", and "boundary violation". Those strings
+        # are not a documented API contract, so upgrading the CLI requires
+        # re-verifying that the expected phrases still appear in alert output.
+        uses: bencherdev/bencher@v0.6.2
+
+      - name: Add tools directory to PATH
+        run: |
+          mkdir -p ~/bin
+          echo "$HOME/bin" >> "$GITHUB_PATH"
+
+      - name: Install libyaml-dev
+        # Must precede any gem install/load: psych links libyaml dynamically. On warm caches the
+        # missing-lib failure is silent; cold runs hit it.
+        run: sudo apt-get install -y libyaml-dev
+
+      - name: Setup k6
+        if: matrix.benchmark_tool == 'k6'
+        uses: grafana/setup-k6-action@v1
+        with:
+          k6-version: ${{ env.K6_VERSION }}
+
+      - name: Cache Vegeta binary
+        id: cache-vegeta
+        if: matrix.benchmark_tool == 'vegeta'
+        uses: actions/cache@v5
+        with:
+          path: ~/bin/vegeta
+          key: vegeta-${{ runner.os }}-${{ runner.arch }}-${{ env.VEGETA_VERSION }}
+
+      - name: Install Vegeta
+        if: matrix.benchmark_tool == 'vegeta' && steps.cache-vegeta.outputs.cache-hit != 'true'
+        run: |
+          echo "Installing Vegeta v${VEGETA_VERSION}"
+          wget -q "https://github.com/tsenart/vegeta/releases/download/v${VEGETA_VERSION}/vegeta_${VEGETA_VERSION}_linux_amd64.tar.gz"
+          tar -xzf "vegeta_${VEGETA_VERSION}_linux_amd64.tar.gz"
+          mv vegeta ~/bin/
+
+      - name: Setup Ruby
+        uses: ruby/setup-ruby@v1
+        with:
+          ruby-version: ${{ env.RUBY_VERSION }}
+
+      - name: Get gem home directory
+        id: gem-home
+        run: echo "path=$(gem env home)" >> "$GITHUB_OUTPUT"
+
+      - name: Cache foreman gem
+        id: cache-foreman
+        uses: actions/cache@v5
         with:
-          benchmark-tool: ${{ matrix.benchmark_tool }}
+          path: ${{ steps.gem-home.outputs.path }}
+          key: foreman-gem-${{ runner.os }}-${{ runner.arch }}-ruby-${{ env.RUBY_VERSION }}
+
+      - name: Install foreman
+        if: steps.cache-foreman.outputs.cache-hit != 'true'
+        run: gem install foreman
+
+      - name: Setup pnpm
+        uses: pnpm/action-setup@v6
+        with:
+          cache: true
+          cache_dependency_path: '**/pnpm-lock.yaml'
+          run_install: false
+
+      - name: Setup Node
+        uses: actions/setup-node@v6
+        with:
+          node-version: '22'
+
+      - name: Print system information
+        run: |
+          echo "Linux release: "
+          cat /etc/issue
+          echo "Current user: "
+          whoami
+          echo "Current directory: "
+          pwd
+          echo "Ruby version: "
+          ruby -v
+          echo "Node version: "
+          node -v
+          echo "Pnpm version: "
+          pnpm --version
+          echo "Bundler version: "
+          bundle --version
+
+      - name: Configure benchmark commands
+        run: |
+          NPROC=$(nproc)
+          echo "Available CPUs: $NPROC"
+
+          if [ "$NPROC" -le 1 ]; then
+            SERVER_CMD="bin/prod"
+            BENCH_CMD="ruby"
+          else
+            SERVER_CMD="taskset -c 1-$((NPROC-1)) bin/prod"
+            BENCH_CMD="taskset -c 0 ruby"
+          fi
+
+          echo "SERVER_CMD=$SERVER_CMD" >> "$GITHUB_ENV"
+          echo "BENCH_CMD=$BENCH_CMD" >> "$GITHUB_ENV"
+
+          if [ -z "$WEB_CONCURRENCY" ]; then
+            # Clamp to >=1 so single-CPU runners don't end up with WEB_CONCURRENCY=0 (Puma
+            # would start in single-mode, which the benchmarks aren't tuned for).
+            WEB_CONCURRENCY=$((NPROC > 1 ? NPROC - 1 : 1))
+            echo "WEB_CONCURRENCY=$WEB_CONCURRENCY" >> "$GITHUB_ENV"
+            echo "WEB_CONCURRENCY (auto): $WEB_CONCURRENCY"
+          else
+            echo "WEB_CONCURRENCY (from input): $WEB_CONCURRENCY"
+          fi
+
+          echo "SERVER_CMD: $SERVER_CMD"
+          echo "BENCH_CMD: $BENCH_CMD"
+
+      - name: Install Node modules with pnpm
+        run: pnpm install --frozen-lockfile
+
+      - name: Build workspace packages
+        run: pnpm run build
 
       - name: Install Ruby Gems for the benchmarked app
         uses: ./.github/actions/setup-bundle

From 4bf329b799814949ce918546b9147afe913872b2 Mon Sep 17 00:00:00 2001
From: Alexey Romanov <alexey.v.romanov@gmail.com>
Date: Mon, 1 Jun 2026 17:36:35 +0300
Subject: [PATCH 3/8] Avoid sharding small manual benchmark route sets

---
 .github/workflows/benchmark.yml |  1 +
 benchmarks/generate_matrix.rb   | 18 +++++++++++++++++-
 2 files changed, 18 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/benchmark.yml b/.github/workflows/benchmark.yml
index 7824f8cd4e..0a83e16ee1 100644
--- a/.github/workflows/benchmark.yml
+++ b/.github/workflows/benchmark.yml
@@ -153,6 +153,7 @@ jobs:
         env:
           BENCHMARK_APP_VERSION: ${{ github.event.inputs.app_version || 'both' }}
           BENCHMARK_EVENT_NAME: ${{ github.event_name }}
+          BENCHMARK_ROUTES: ${{ github.event.inputs.routes || '' }}
           BENCHMARK_NON_RUNTIME_ONLY: ${{ steps.detect.outputs.non_runtime_only }}
           BENCHMARK_PULL_REQUEST_HEAD_REPO: ${{ github.event.pull_request.head.repo.full_name || '' }}
           BENCHMARK_PULL_REQUEST_LABELS: ${{ github.event_name == 'pull_request' && toJSON(github.event.pull_request.labels.*.name) || '[]' }}
diff --git a/benchmarks/generate_matrix.rb b/benchmarks/generate_matrix.rb
index c6edfa39e6..d515f833f5 100755
--- a/benchmarks/generate_matrix.rb
+++ b/benchmarks/generate_matrix.rb
@@ -7,6 +7,8 @@
 
 require "json"
 
+MIN_ROUTES_FOR_SHARDING = 10
+
 SUITES = [
   {
     id: "core",
@@ -138,8 +140,22 @@ def suite_enabled?(suite, labels)
   suite_selected_by_input?(suite) && suite_requested_by_event?(suite, labels)
 end
 
+def explicit_routes
+  ENV.fetch("BENCHMARK_ROUTES", "").split(",").map(&:strip).reject(&:empty?)
+end
+
+def shard_total_for_suite(suite)
+  configured_shard_total = suite.fetch(:shard_total)
+  explicit_route_count = explicit_routes.length
+
+  return configured_shard_total if explicit_route_count.zero?
+  return 1 if explicit_route_count < MIN_ROUTES_FOR_SHARDING
+
+  [configured_shard_total, explicit_route_count].min
+end
+
 def suite_rows(suite)
-  shard_total = suite.fetch(:shard_total)
+  shard_total = shard_total_for_suite(suite)
   raise "#{suite.fetch(:id)} shard_total must be positive (got #{shard_total})" unless shard_total.positive?
 
   Array.new(shard_total) do |index|

From 7da22f20651cd634cd19e5f4f9ee8fed5568331e Mon Sep 17 00:00:00 2001
From: Alexey Romanov <alexey.v.romanov@gmail.com>
Date: Mon, 1 Jun 2026 17:47:38 +0300
Subject: [PATCH 4/8] Remove unused benchmark shard slug from matrix

---
 benchmarks/generate_matrix.rb | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/benchmarks/generate_matrix.rb b/benchmarks/generate_matrix.rb
index d515f833f5..42b0754fee 100755
--- a/benchmarks/generate_matrix.rb
+++ b/benchmarks/generate_matrix.rb
@@ -77,7 +77,6 @@
   shard_index: 0,
   shard_total: 1,
   shard_label: "1/1",
-  shard_slug: "skipped",
   bencher_suite_name: "No benchmark suite",
   report_marker: "<!-- BENCHER_REPORT_SKIPPED -->",
   app_directory: ".",
@@ -165,15 +164,16 @@ def suite_rows(suite)
     suite_name = suite.fetch(:suite_name)
     suite_prefix = suite.fetch(:suite_prefix)
 
-    suite_row(suite, suite_name, suite_prefix,
+    suite_row(suite, suite_name,
               shard_index: index,
               shard_total: shard_total,
               shard_label: shard_label,
-              shard_slug: shard_slug)
+              report_marker: report_marker(suite, suite_prefix, shard_slug),
+              artifact_name_suffix: artifact_name_suffix(shard_total, shard_slug))
   end
 end
 
-def suite_row(suite, suite_name, suite_prefix, shard_index:, shard_total:, shard_label:, shard_slug:)
+def suite_row(suite, suite_name, shard_index:, shard_total:, shard_label:, report_marker:, artifact_name_suffix:)
   {
     suite_id: suite.fetch(:id),
     suite_name: suite_name,
@@ -181,12 +181,11 @@ def suite_row(suite, suite_name, suite_prefix, shard_index:, shard_total:, shard
     shard_index: shard_index,
     shard_total: shard_total,
     shard_label: shard_label,
-    shard_slug: shard_slug,
     bencher_suite_name: bencher_suite_name(suite_name, shard_total, shard_label),
-    report_marker: report_marker(suite, suite_prefix, shard_slug),
+    report_marker: report_marker,
     app_directory: suite.fetch(:app_directory),
     artifact_name_prefix: suite.fetch(:artifact_name),
-    artifact_name_suffix: artifact_name_suffix(shard_total, shard_slug),
+    artifact_name_suffix: artifact_name_suffix,
     benchmark_tool: suite.fetch(:benchmark_tool),
     benchmark_script: suite.fetch(:benchmark_script),
     benchmark_timeout_minutes: suite.fetch(:benchmark_timeout_minutes),

From 532aff5566f4065fe50afb6e6d0217eef3b0b0d2 Mon Sep 17 00:00:00 2001
From: Alexey Romanov <alexey.v.romanov@gmail.com>
Date: Mon, 1 Jun 2026 18:09:40 +0300
Subject: [PATCH 5/8] Add tests

---
 .github/workflows/benchmark.yml               |  29 +++
 .github/workflows/lint-js-and-ruby.yml        |   6 +
 benchmarks/generate_matrix.rb                 |  22 +-
 benchmarks/spec/benchmark_matrix_spec.rb      | 190 ++++++++++++++++++
 .../spec}/benchmark_routes_spec.rb            |   2 +-
 benchmarks/spec/spec_helper.rb                |  39 ++++
 script/ci-changes-detector                    |  37 +++-
 script/ci-changes-detector-test.bash          |  52 +++++
 8 files changed, 365 insertions(+), 12 deletions(-)
 create mode 100644 benchmarks/spec/benchmark_matrix_spec.rb
 rename {react_on_rails/spec/react_on_rails => benchmarks/spec}/benchmark_routes_spec.rb (98%)
 create mode 100644 benchmarks/spec/spec_helper.rb

diff --git a/.github/workflows/benchmark.yml b/.github/workflows/benchmark.yml
index 0a83e16ee1..58b6cd38d0 100644
--- a/.github/workflows/benchmark.yml
+++ b/.github/workflows/benchmark.yml
@@ -173,6 +173,35 @@ jobs:
             echo "EOF"
           } >> "$GITHUB_OUTPUT"
         shell: bash
+      # Only pay the Bundler setup when benchmarks will run or the scripts
+      # themselves changed; unrelated PRs keep this job stdlib-only and fast.
+      # TODO: when fixing #2214 this should switch to the new root Gemfile.
+      - name: Install libyaml-dev
+        # Must precede any gem install/load: psych links libyaml dynamically. On warm caches the
+        # missing-lib failure is silent; cold runs hit it.
+        if: steps.benchmark-matrices.outputs.run_benchmark_suites == 'true' || steps.detect.outputs.benchmarks_changed == 'true'
+        run: sudo apt-get install -y libyaml-dev
+      - name: Setup Ruby
+        if: steps.benchmark-matrices.outputs.run_benchmark_suites == 'true' || steps.detect.outputs.benchmarks_changed == 'true'
+        uses: ruby/setup-ruby@v1
+        with:
+          ruby-version: ${{ env.RUBY_VERSION }}
+      # The benchmark specs only need rspec, which the root Gemfile provides (it
+      # delegates to react_on_rails/Gemfile). The setup-bundle action can't manage
+      # it because there is no committed root Gemfile.lock yet, so install manually.
+      # TODO(#2214): once a root Gemfile.lock is committed, replace this step with
+      # the setup-bundle action (working-directory: ".") and delete it.
+      - name: Install gems for benchmark specs
+        if: steps.benchmark-matrices.outputs.run_benchmark_suites == 'true' || steps.detect.outputs.benchmarks_changed == 'true'
+        run: bundle install --jobs 4 --retry 3
+      # Validate the benchmark scripts (matrix generation, route discovery) before
+      # any benchmark runs: if this logic is wrong, the matrix and routes feeding
+      # the run are wrong, so the results would be invalid. Running it here (rather
+      # than a separate job) reuses this job's checkout/runner and still gates the
+      # benchmark job, which needs detect-changes to succeed.
+      - name: Run benchmark script specs
+        if: steps.benchmark-matrices.outputs.run_benchmark_suites == 'true' || steps.detect.outputs.benchmarks_changed == 'true'
+        run: bundle exec rspec benchmarks/spec
 
   benchmark:
     needs: detect-changes
diff --git a/.github/workflows/lint-js-and-ruby.yml b/.github/workflows/lint-js-and-ruby.yml
index 92f22d7540..b63eb7b8a6 100644
--- a/.github/workflows/lint-js-and-ruby.yml
+++ b/.github/workflows/lint-js-and-ruby.yml
@@ -165,6 +165,12 @@ jobs:
         env:
           BUNDLE_GEMFILE: ${{ github.workspace }}/react_on_rails/Gemfile
         run: cd react_on_rails && bundle exec rubocop
+      # The benchmarks/ scripts live outside the gem, so the package rubocop run
+      # above doesn't see them. Lint them from the repo root (root .rubocop.yml).
+      - name: Lint benchmark scripts
+        env:
+          BUNDLE_GEMFILE: ${{ github.workspace }}/react_on_rails/Gemfile
+        run: bundle exec rubocop benchmarks
       - name: Validate RBS type signatures
         env:
           BUNDLE_GEMFILE: ${{ github.workspace }}/react_on_rails/Gemfile
diff --git a/benchmarks/generate_matrix.rb b/benchmarks/generate_matrix.rb
index 42b0754fee..4c5991c198 100755
--- a/benchmarks/generate_matrix.rb
+++ b/benchmarks/generate_matrix.rb
@@ -219,13 +219,17 @@ def report_marker(suite, suite_prefix, shard_slug)
   suite[:report_marker] || "<!-- BENCHER_REPORT_#{suite_prefix}_#{shard_slug.upcase.tr('-', '_')} -->"
 end
 
-labels = pull_request_labels
-rows = if truthy_env?("BENCHMARK_NON_RUNTIME_ONLY")
-         []
-       else
-         SUITES.select { |suite| suite_enabled?(suite, labels) }.flat_map { |suite| suite_rows(suite) }
-       end
-
-matrix = { include: rows.empty? ? [EMPTY_MATRIX_ROW] : rows }
+def build_matrix
+  labels = pull_request_labels
+  rows = if truthy_env?("BENCHMARK_NON_RUNTIME_ONLY")
+           []
+         else
+           SUITES.select { |suite| suite_enabled?(suite, labels) }.flat_map { |suite| suite_rows(suite) }
+         end
+
+  { include: rows.empty? ? [EMPTY_MATRIX_ROW] : rows }
+end
 
-puts JSON.generate(matrix)
+# Only emit when run as a script; `require`-ing the file (e.g. from specs) just
+# loads the helpers above without printing.
+puts JSON.generate(build_matrix) if __FILE__ == $PROGRAM_NAME
diff --git a/benchmarks/spec/benchmark_matrix_spec.rb b/benchmarks/spec/benchmark_matrix_spec.rb
new file mode 100644
index 0000000000..65f6a8ac99
--- /dev/null
+++ b/benchmarks/spec/benchmark_matrix_spec.rb
@@ -0,0 +1,190 @@
+# frozen_string_literal: true
+
+require_relative "spec_helper"
+require_relative "../generate_matrix"
+
+# These specs pin the gating/sharding/naming logic in generate_matrix.rb that
+# decides which benchmark jobs run and how Bencher attributes their results. The
+# script reads everything from ENV, so each example sets the env the workflow
+# would pass (see .github/workflows/benchmark.yml "Set benchmark matrices") and
+# asserts on the resulting matrix `include:` rows.
+RSpec.describe "benchmark matrix generation" do
+  include BenchmarkEnvHelper
+
+  # Returns the matrix `include:` rows produced for the given ENV.
+  def rows_for(env)
+    with_env(env) { build_matrix }.fetch(:include)
+  end
+
+  def suite_ids_for(env)
+    rows_for(env).map { |row| row.fetch(:suite_id) }
+  end
+
+  describe "the empty/skipped placeholder" do
+    it "emits a single 'none' row when no suite is selected (fork PR)" do
+      rows = rows_for(
+        "BENCHMARK_EVENT_NAME" => "pull_request",
+        "BENCHMARK_PULL_REQUEST_HEAD_REPO" => "contributor/react_on_rails",
+        "GITHUB_REPOSITORY" => "shakacode/react_on_rails",
+        "BENCHMARK_PULL_REQUEST_LABELS" => '["benchmark-pro"]'
+      )
+
+      expect(rows.size).to eq(1)
+      expect(rows.first).to include(suite_id: "none", benchmark_tool: "none")
+    end
+
+    it "forces the placeholder for non-runtime-only changes regardless of other inputs" do
+      expect(suite_ids_for(
+               "BENCHMARK_EVENT_NAME" => "push",
+               "BENCHMARK_NON_RUNTIME_ONLY" => "true"
+             )).to eq(["none"])
+    end
+  end
+
+  describe "event gating" do
+    it "runs every suite on push (app_version 'both')" do
+      expect(suite_ids_for(
+               "BENCHMARK_EVENT_NAME" => "push",
+               "BENCHMARK_APP_VERSION" => "both"
+             )).to eq(%w[core pro pro pro-node-renderer])
+    end
+
+    it "honors a suite-specific run_output even without labels or push" do
+      # RUN_PRO_BENCHMARKS gates only the Pro suite; core/node-renderer stay off.
+      expect(suite_ids_for(
+               "BENCHMARK_EVENT_NAME" => "pull_request",
+               "BENCHMARK_PULL_REQUEST_HEAD_REPO" => "shakacode/react_on_rails",
+               "GITHUB_REPOSITORY" => "shakacode/react_on_rails",
+               "RUN_PRO_BENCHMARKS" => "true"
+             )).to eq(%w[pro pro])
+    end
+
+    it "selects suites by label intersection on a same-repo PR" do
+      # benchmark-pro is shared by Pro and Pro-Node-Renderer, but not Core.
+      expect(suite_ids_for(
+               "BENCHMARK_EVENT_NAME" => "pull_request",
+               "BENCHMARK_PULL_REQUEST_HEAD_REPO" => "shakacode/react_on_rails",
+               "GITHUB_REPOSITORY" => "shakacode/react_on_rails",
+               "BENCHMARK_PULL_REQUEST_LABELS" => '["benchmark-pro"]'
+             )).to eq(%w[pro pro pro-node-renderer])
+    end
+
+    it "ignores PR labels from a fork (fork-safety guard)" do
+      expect(suite_ids_for(
+               "BENCHMARK_EVENT_NAME" => "pull_request",
+               "BENCHMARK_PULL_REQUEST_HEAD_REPO" => "contributor/react_on_rails",
+               "GITHUB_REPOSITORY" => "shakacode/react_on_rails",
+               "BENCHMARK_PULL_REQUEST_LABELS" => '["benchmark"]'
+             )).to eq(["none"])
+    end
+  end
+
+  describe "app_version input filtering" do
+    it "restricts a workflow_dispatch to the node-renderer suite for pro_node_renderer_only" do
+      expect(suite_ids_for(
+               "BENCHMARK_EVENT_NAME" => "workflow_dispatch",
+               "BENCHMARK_APP_VERSION" => "pro_node_renderer_only"
+             )).to eq(%w[pro-node-renderer])
+    end
+
+    it "restricts to Core only for core_only" do
+      expect(suite_ids_for(
+               "BENCHMARK_EVENT_NAME" => "push",
+               "BENCHMARK_APP_VERSION" => "core_only"
+             )).to eq(%w[core])
+    end
+  end
+
+  describe "sharding and naming" do
+    it "expands the 2-shard Pro suite into per-shard rows" do
+      # pro_rails_only is exclusive to the Pro suite; pro_only would also pull in
+      # Pro-Node-Renderer (its app_versions include pro_only).
+      rows = rows_for(
+        "BENCHMARK_EVENT_NAME" => "push",
+        "BENCHMARK_APP_VERSION" => "pro_rails_only"
+      )
+
+      expect(rows.map { |row| row.fetch(:shard_label) }).to eq(%w[1/2 2/2])
+      expect(rows.map { |row| row.fetch(:job_name) }).to eq(
+        ["Pro benchmarks (shard 1/2)", "Pro benchmarks (shard 2/2)"]
+      )
+      expect(rows.map { |row| row.fetch(:bencher_suite_name) }).to eq(
+        ["Pro (shard 1/2)", "Pro (shard 2/2)"]
+      )
+      expect(rows.map { |row| row.fetch(:artifact_name_suffix) }).to eq(
+        %w[-shard-1-of-2 -shard-2-of-2]
+      )
+      expect(rows.map { |row| row.fetch(:report_marker) }).to eq(
+        ["<!-- BENCHER_REPORT_PRO_SHARD_1_OF_2 -->", "<!-- BENCHER_REPORT_PRO_SHARD_2_OF_2 -->"]
+      )
+    end
+
+    it "drops the shard suffix and plural naming for a single-shard suite" do
+      row = rows_for(
+        "BENCHMARK_EVENT_NAME" => "push",
+        "BENCHMARK_APP_VERSION" => "core_only"
+      ).fetch(0)
+
+      expect(row).to include(
+        job_name: "Core benchmarks",
+        bencher_suite_name: "Core",
+        artifact_name_suffix: "",
+        report_marker: "<!-- BENCHER_REPORT_CORE_SHARD_1_OF_1 -->"
+      )
+    end
+
+    it "uses the explicit report_marker override when a suite defines one" do
+      row = rows_for(
+        "BENCHMARK_EVENT_NAME" => "push",
+        "BENCHMARK_APP_VERSION" => "pro_node_renderer_only"
+      ).fetch(0)
+
+      expect(row.fetch(:report_marker)).to eq("<!-- BENCHER_REPORT_PRO_NODE_RENDERER -->")
+    end
+  end
+
+  describe "explicit-route sharding (BENCHMARK_ROUTES, workflow_dispatch)" do
+    # Isolate the Pro suite (configured for 2 shards) via pro_rails_only, then vary
+    # the manual route list to drive shard_total_for_suite. Fewer than 10 routes
+    # isn't worth parallelizing, so it collapses to one shard.
+    def pro_shard_labels(routes)
+      rows_for(
+        "BENCHMARK_EVENT_NAME" => "workflow_dispatch",
+        "BENCHMARK_APP_VERSION" => "pro_rails_only",
+        "BENCHMARK_ROUTES" => routes
+      ).map { |row| row.fetch(:shard_label) }
+    end
+
+    it "collapses to a single shard when fewer than 10 routes are given" do
+      expect(pro_shard_labels("/,/hello,/world")).to eq(%w[1/1])
+    end
+
+    it "keeps the configured shard count when at least 10 routes are given" do
+      routes = (1..10).map { |n| "/r#{n}" }.join(",")
+      expect(pro_shard_labels(routes)).to eq(%w[1/2 2/2])
+    end
+
+    it "ignores blank/whitespace entries and falls back to the configured count" do
+      expect(pro_shard_labels(" , ,  ")).to eq(%w[1/2 2/2])
+    end
+  end
+
+  describe "input validation" do
+    it "raises a descriptive error for malformed label JSON" do
+      expect do
+        rows_for(
+          "BENCHMARK_EVENT_NAME" => "pull_request",
+          "BENCHMARK_PULL_REQUEST_LABELS" => "not-json"
+        )
+      end.to raise_error(/BENCHMARK_PULL_REQUEST_LABELS must be JSON array/)
+    end
+
+    it "raises when a suite would produce a non-positive shard count" do
+      with_env({}) do
+        expect do
+          suite_rows({ id: "fake", shard_total: 0 })
+        end.to raise_error(/fake shard_total must be positive \(got 0\)/)
+      end
+    end
+  end
+end
diff --git a/react_on_rails/spec/react_on_rails/benchmark_routes_spec.rb b/benchmarks/spec/benchmark_routes_spec.rb
similarity index 98%
rename from react_on_rails/spec/react_on_rails/benchmark_routes_spec.rb
rename to benchmarks/spec/benchmark_routes_spec.rb
index 761951b7b6..db2f356ec0 100644
--- a/react_on_rails/spec/react_on_rails/benchmark_routes_spec.rb
+++ b/benchmarks/spec/benchmark_routes_spec.rb
@@ -2,7 +2,7 @@
 
 require_relative "spec_helper"
 require "tmpdir"
-require_relative "../../../benchmarks/lib/benchmark_routes"
+require_relative "../lib/benchmark_routes"
 
 RSpec.describe "benchmark route discovery helpers" do
   describe "#benchmark_routes_from_rails_routes_output" do
diff --git a/benchmarks/spec/spec_helper.rb b/benchmarks/spec/spec_helper.rb
new file mode 100644
index 0000000000..cd18f90e2d
--- /dev/null
+++ b/benchmarks/spec/spec_helper.rb
@@ -0,0 +1,39 @@
+# frozen_string_literal: true
+
+# Lightweight spec_helper for the stdlib-only scripts under benchmarks/.
+#
+# Unlike react_on_rails/spec/react_on_rails/spec_helper.rb, this deliberately
+# does NOT load Rails or the react_on_rails gem: the benchmark scripts are plain
+# Ruby (require "json" and friends), so the runner only needs rspec. Keep it that
+# way — pulling in Rails here would defeat the point of testing these in
+# isolation and slow the suite down.
+
+require "json"
+
+# Mixin for specs that drive scripts which read all of their input from ENV
+# (e.g. generate_matrix.rb). `with_env` swaps in the given vars, clears every
+# other benchmark gating key so the host/CI environment can't leak in, and
+# always restores the original ENV afterwards.
+module BenchmarkEnvHelper
+  GATING_ENV_KEYS = %w[
+    BENCHMARK_EVENT_NAME
+    BENCHMARK_APP_VERSION
+    BENCHMARK_NON_RUNTIME_ONLY
+    BENCHMARK_PULL_REQUEST_LABELS
+    BENCHMARK_PULL_REQUEST_HEAD_REPO
+    BENCHMARK_ROUTES
+    GITHUB_REPOSITORY
+    RUN_CORE_BENCHMARKS
+    RUN_PRO_BENCHMARKS
+    RUN_PRO_NODE_RENDERER_BENCHMARKS
+  ].freeze
+
+  def with_env(vars)
+    snapshot = ENV.to_h
+    GATING_ENV_KEYS.each { |key| ENV.delete(key) }
+    vars.each { |key, value| ENV[key.to_s] = value.to_s }
+    yield
+  ensure
+    ENV.replace(snapshot)
+  end
+end
diff --git a/script/ci-changes-detector b/script/ci-changes-detector
index 4d15ea8bc7..14bc81dd9e 100755
--- a/script/ci-changes-detector
+++ b/script/ci-changes-detector
@@ -58,8 +58,11 @@ if [ -z "${CI:-}" ] && [ "$CURRENT_REF" = "HEAD" ]; then
 fi
 
 if [ -z "$CHANGED_FILES" ]; then
+  # Don't exit here: with no files the loop below sets nothing, DOCS_ONLY stays
+  # true, and the docs-only output block emits the full flag set (non_runtime_only
+  # =true, every run_* false, benchmarks_changed=false). Exiting would instead
+  # leave every output blank.
   echo "No changes detected between $BASE_REF and $CURRENT_REF"
-  exit 0
 fi
 
 # Initialize flags
@@ -68,6 +71,7 @@ LINT_CONFIG_CHANGED=false
 PRO_LINT_CONFIG_CHANGED=false
 RUBY_CORE_CHANGED=false
 RSPEC_CHANGED=false
+BENCHMARKS_CHANGED=false
 SPEC_DUMMY_CHANGED=false
 JS_CHANGED=false
 GENERATORS_CHANGED=false
@@ -241,6 +245,9 @@ source_file_comment_only_change() {
 
 # Analyze each changed file
 while IFS= read -r file; do
+  # A here-string yields a single empty line for empty input, so skip blank lines —
+  # otherwise an empty diff would fall into the catch-all below and trigger every job.
+  [ -z "$file" ] && continue
   # *.md and docs/* intentionally overlap (docs/foo.md matches both); *.md wins
   # because it comes first. Several later branches (generators, Ruby core,
   # specs) also pair top-level and **/* globs for clarity even though * matches
@@ -282,6 +289,19 @@ while IFS= read -r file; do
       fi
       ;;
 
+    # Benchmark scripts (stdlib-only Ruby under benchmarks/) and their specs.
+    # These only need rubocop here; their specs run in the benchmark workflow's
+    # script-validation step (not the gem rspec suite), and they don't need the
+    # full dummy/integration/Pro suite the catch-all would otherwise trigger.
+    benchmarks/*|benchmarks/**/*)
+      DOCS_ONLY=false
+      if source_file_comment_only_change "$file"; then
+        SOURCE_COMMENT_ONLY_CHANGED=true
+      else
+        BENCHMARKS_CHANGED=true
+      fi
+      ;;
+
     # JavaScript/TypeScript source (including tests, scripts, and package files)
     package.json|pnpm-lock.yaml|packages/react-on-rails/src/*|packages/react-on-rails/src/**/*|packages/react-on-rails/tests/*|packages/react-on-rails/tests/**/*|packages/react-on-rails/scripts/*|packages/react-on-rails/scripts/**/*|packages/react-on-rails/package.json|packages/react-on-rails/tsconfig.json|.github/workflows/package-js-tests.yml)
       DOCS_ONLY=false
@@ -396,6 +416,7 @@ elif [ "$SOURCE_COMMENT_ONLY_CHANGED" = true ] || [ "$PRO_SOURCE_COMMENT_ONLY_CH
     [ "$PRO_LINT_CONFIG_CHANGED" = false ] &&
     [ "$RUBY_CORE_CHANGED" = false ] &&
     [ "$RSPEC_CHANGED" = false ] &&
+    [ "$BENCHMARKS_CHANGED" = false ] &&
     [ "$SPEC_DUMMY_CHANGED" = false ] &&
     [ "$JS_CHANGED" = false ] &&
     [ "$GENERATORS_CHANGED" = false ] &&
@@ -430,6 +451,7 @@ if [ "$DOCS_ONLY" = true ]; then
     "run_pro_tests:false"
     "run_pro_dummy_tests:false"
     "run_pro_node_renderer_tests:false"
+    "benchmarks_changed:false"
   )
 
   # Output to GITHUB_OUTPUT if in GitHub Actions
@@ -468,6 +490,7 @@ echo "Changed file categories:"
 [ "$RUBY_CORE_CHANGED" = true ] && echo -e "${YELLOW}  • Ruby core source code${NC}"
 [ "$JS_CHANGED" = true ] && echo -e "${YELLOW}  • JavaScript/TypeScript code${NC}"
 [ "$RSPEC_CHANGED" = true ] && echo -e "${YELLOW}  • RSpec tests${NC}"
+[ "$BENCHMARKS_CHANGED" = true ] && echo -e "${YELLOW}  • Benchmark scripts${NC}"
 [ "$SPEC_DUMMY_CHANGED" = true ] && echo -e "${YELLOW}  • Dummy app${NC}"
 [ "$GENERATORS_CHANGED" = true ] && echo -e "${YELLOW}  • Generators${NC}"
 [ "$PRO_JS_CHANGED" = true ] && echo -e "${YELLOW}  • React on Rails Pro JavaScript/TypeScript${NC}"
@@ -508,6 +531,14 @@ if [ "$RUBY_CORE_CHANGED" = true ] || [ "$RSPEC_CHANGED" = true ] || [ "$GENERAT
   RUN_RUBY_TESTS=true
 fi
 
+# Benchmark scripts live outside the gem: they're linted by rubocop and covered
+# by their own specs in the benchmark workflow (not the gem rspec suite), so they
+# only need lint here. The benchmarks_changed output drives the benchmark
+# workflow's script-validation step.
+if [ "$BENCHMARKS_CHANGED" = true ]; then
+  RUN_LINT=true
+fi
+
 if [ "$JS_CHANGED" = true ]; then
   RUN_JS_TESTS=true
 fi
@@ -588,6 +619,7 @@ if [ -n "${GITHUB_OUTPUT:-}" ]; then
     echo "run_pro_tests=$RUN_PRO_TESTS"
     echo "run_pro_dummy_tests=$RUN_PRO_DUMMY_TESTS"
     echo "run_pro_node_renderer_tests=$RUN_PRO_NODE_RENDERER_TESTS"
+    echo "benchmarks_changed=$BENCHMARKS_CHANGED"
   } >> "$GITHUB_OUTPUT"
 fi
 
@@ -606,7 +638,8 @@ if [ "${CI_JSON_OUTPUT:-}" = "1" ]; then
   "run_pro_lint": $RUN_PRO_LINT,
   "run_pro_tests": $RUN_PRO_TESTS,
   "run_pro_dummy_tests": $RUN_PRO_DUMMY_TESTS,
-  "run_pro_node_renderer_tests": $RUN_PRO_NODE_RENDERER_TESTS
+  "run_pro_node_renderer_tests": $RUN_PRO_NODE_RENDERER_TESTS,
+  "benchmarks_changed": $BENCHMARKS_CHANGED
 }
 EOF
 fi
diff --git a/script/ci-changes-detector-test.bash b/script/ci-changes-detector-test.bash
index daea7a54c4..3c6edf87bc 100644
--- a/script/ci-changes-detector-test.bash
+++ b/script/ci-changes-detector-test.bash
@@ -96,6 +96,7 @@ setup_repo() {
   mkdir -p react_on_rails/spec/react_on_rails
   mkdir -p react_on_rails/app/helpers
   mkdir -p react_on_rails_pro/app/controllers/react_on_rails_pro/rolling_deploy
+  mkdir -p benchmarks/lib
   cat > docs/guide.md <<'DOC'
 # Guide
 DOC
@@ -138,6 +139,13 @@ module ReactOnRails
     end
   end
 end
+RUBY
+  cat > benchmarks/lib/sample.rb <<'RUBY'
+module BenchmarkSample
+  def self.call
+    "ok"
+  end
+end
 RUBY
   cat > react_on_rails_pro/app/controllers/react_on_rails_pro/rolling_deploy/bundles_controller.rb <<'RUBY'
 module ReactOnRailsPro
@@ -553,6 +561,48 @@ test_core_js_changes_request_e2e() {
   assert_contains "$out" '"run_e2e_tests": true' "core js output"
 }
 
+test_benchmark_source_change_lints_and_flags_benchmarks_only() {
+  setup_repo
+  write_file_change "benchmarks/generate_matrix.rb"
+
+  local out
+  out="$(detector_output)"
+  assert_contains "$out" '"benchmarks_changed": true' "benchmark source output"
+  assert_contains "$out" '"run_lint": true' "benchmark source output"
+  # Benchmark scripts have no coverage in the gem suite, dummy app, or Pro stack.
+  assert_contains "$out" '"run_ruby_tests": false' "benchmark source output"
+  assert_contains "$out" '"run_dummy_tests": false' "benchmark source output"
+  assert_contains "$out" '"run_e2e_tests": false' "benchmark source output"
+  assert_contains "$out" '"run_pro_tests": false' "benchmark source output"
+}
+
+test_benchmark_comment_only_change_is_non_runtime_but_keeps_lint() {
+  setup_repo
+  perl -0pi -e 's/module BenchmarkSample/# Describe the sample.\nmodule BenchmarkSample/' \
+    benchmarks/lib/sample.rb
+  commit_change "benchmark comment"
+
+  local out
+  out="$(detector_output)"
+  assert_contains "$out" '"non_runtime_only": true' "benchmark comment output"
+  assert_contains "$out" '"run_lint": true' "benchmark comment output"
+  assert_contains "$out" '"benchmarks_changed": false' "benchmark comment output"
+  assert_contains "$out" '"run_ruby_tests": false' "benchmark comment output"
+}
+
+test_empty_diff_skips_everything() {
+  setup_repo
+  git commit --allow-empty -m "no file changes" >/dev/null
+
+  local out
+  out="$(detector_output)"
+  assert_contains "$out" '"non_runtime_only": true' "empty diff output"
+  assert_contains "$out" '"run_lint": false' "empty diff output"
+  assert_contains "$out" '"run_ruby_tests": false' "empty diff output"
+  assert_contains "$out" '"benchmarks_changed": false' "empty diff output"
+}
+
+run_test test_empty_diff_skips_everything
 run_test test_docs_changes_are_non_runtime_only
 run_test test_ruby_comment_only_change_skips_heavy_tests_but_keeps_lint
 run_test test_ruby_block_comment_only_change_skips_heavy_tests_but_keeps_lint
@@ -584,6 +634,8 @@ run_test test_generator_only_changes_do_not_request_e2e
 run_test test_dummy_app_changes_request_e2e
 run_test test_core_ruby_changes_request_e2e
 run_test test_core_js_changes_request_e2e
+run_test test_benchmark_source_change_lints_and_flags_benchmarks_only
+run_test test_benchmark_comment_only_change_is_non_runtime_but_keeps_lint
 
 echo
 echo "CI changes detector tests: $TESTS_RUN run, $TESTS_FAILED failed"

From 3e1ee839448947f72ee12e1acf7391a034179f68 Mon Sep 17 00:00:00 2001
From: Alexey Romanov <alexey.v.romanov@gmail.com>
Date: Mon, 1 Jun 2026 23:39:00 +0300
Subject: [PATCH 6/8] Drop dead matrix job outputs and harden workflow step
 inputs

- Remove the unused run_*_benchmarks detect-changes outputs; after the matrix
  unification no downstream job consumes them (only step-local env does).
- Route matrix/context values through env: in the Execute and Summary steps so
  the shell treats them as data, not code (GitHub Actions injection hardening).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 .github/workflows/benchmark.yml | 39 ++++++++++++++++++++++-----------
 1 file changed, 26 insertions(+), 13 deletions(-)

diff --git a/.github/workflows/benchmark.yml b/.github/workflows/benchmark.yml
index 58b6cd38d0..3017359d44 100644
--- a/.github/workflows/benchmark.yml
+++ b/.github/workflows/benchmark.yml
@@ -91,9 +91,6 @@ jobs:
     runs-on: ubuntu-24.04
     outputs:
       non_runtime_only: ${{ steps.detect.outputs.non_runtime_only }}
-      run_core_benchmarks: ${{ steps.benchmark-outputs.outputs.run_core_benchmarks }}
-      run_pro_benchmarks: ${{ steps.benchmark-outputs.outputs.run_pro_benchmarks }}
-      run_pro_node_renderer_benchmarks: ${{ steps.benchmark-outputs.outputs.run_pro_node_renderer_benchmarks }}
       run_benchmark_suites: ${{ steps.benchmark-matrices.outputs.run_benchmark_suites }}
       benchmark_matrix: ${{ steps.benchmark-matrices.outputs.benchmark_matrix }}
     steps:
@@ -437,16 +434,23 @@ jobs:
 
       - name: Execute benchmark suite
         timeout-minutes: ${{ matrix.benchmark_timeout_minutes }}
+        # Route matrix values through env so the shell treats them as data, not code
+        # (GitHub Actions injection hardening), even though they come from our own
+        # generate_matrix.rb.
+        env:
+          SUITE_NAME: ${{ matrix.suite_name }}
+          PRO_ENV: ${{ matrix.pro_env }}
+          BENCHMARK_SCRIPT: ${{ matrix.benchmark_script }}
         run: |
           set -e
-          echo "🏃 Running ${{ matrix.suite_name }} benchmark suite..."
+          echo "🏃 Running $SUITE_NAME benchmark suite..."
 
-          if [ "${{ matrix.pro_env }}" = "true" ]; then
-            if ! PRO=true $BENCH_CMD "${{ matrix.benchmark_script }}"; then
+          if [ "$PRO_ENV" = "true" ]; then
+            if ! PRO=true $BENCH_CMD "$BENCHMARK_SCRIPT"; then
               echo "❌ ERROR: Benchmark execution failed"
               exit 1
             fi
-          elif ! $BENCH_CMD "${{ matrix.benchmark_script }}"; then
+          elif ! $BENCH_CMD "$BENCHMARK_SCRIPT"; then
             echo "❌ ERROR: Benchmark execution failed"
             exit 1
           fi
@@ -501,12 +505,21 @@ jobs:
 
       - name: Benchmark workflow summary
         if: always()
+        # Route context/matrix values through env so the shell treats them as data,
+        # not code (GitHub Actions injection hardening; ref_name can be attacker-set).
+        env:
+          SUITE_NAME: ${{ matrix.suite_name }}
+          SHARD_LABEL: ${{ matrix.shard_label }}
+          JOB_STATUS: ${{ job.status }}
+          RUN_NUMBER: ${{ github.run_number }}
+          ACTOR: ${{ github.actor }}
+          REF_NAME: ${{ github.ref_name }}
         run: |
           echo "📋 Benchmark Workflow Summary"
           echo "===================================="
-          echo "Suite: ${{ matrix.suite_name }}"
-          echo "Shard: ${{ matrix.shard_label }}"
-          echo "Status: ${{ job.status }}"
-          echo "Run number: ${{ github.run_number }}"
-          echo "Triggered by: ${{ github.actor }}"
-          echo "Branch: ${{ github.ref_name }}"
+          echo "Suite: $SUITE_NAME"
+          echo "Shard: $SHARD_LABEL"
+          echo "Status: $JOB_STATUS"
+          echo "Run number: $RUN_NUMBER"
+          echo "Triggered by: $ACTOR"
+          echo "Branch: $REF_NAME"

From fbaf37ab3e01f2536b32c4f5ebcc344302cd48d7 Mon Sep 17 00:00:00 2001
From: Alexey Romanov <alexey.v.romanov@gmail.com>
Date: Tue, 2 Jun 2026 01:30:26 +0300
Subject: [PATCH 7/8] Generate the skipped benchmark matrix row from a real
 suite

Build the no-suite-selected placeholder via suite_row (overriding only
suite_id and a friendly job_name) instead of hand-maintaining a parallel
EMPTY_MATRIX_ROW literal. This guarantees the placeholder shares a real
row's exact key set (the matrix include: requires it, and the workflow
reads matrix.<field> on the skipped path too) and emits the same string
pro_env/generate_packs values real rows produce.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 benchmarks/generate_matrix.rb            | 32 +++++++-----------------
 benchmarks/spec/benchmark_matrix_spec.rb |  4 ++-
 2 files changed, 12 insertions(+), 24 deletions(-)

diff --git a/benchmarks/generate_matrix.rb b/benchmarks/generate_matrix.rb
index 4c5991c198..15daf829c0 100755
--- a/benchmarks/generate_matrix.rb
+++ b/benchmarks/generate_matrix.rb
@@ -70,28 +70,6 @@
   }
 ].freeze
 
-EMPTY_MATRIX_ROW = {
-  suite_id: "none",
-  suite_name: "No benchmark suite",
-  job_name: "Benchmark suites skipped",
-  shard_index: 0,
-  shard_total: 1,
-  shard_label: "1/1",
-  bencher_suite_name: "No benchmark suite",
-  report_marker: "<!-- BENCHER_REPORT_SKIPPED -->",
-  app_directory: ".",
-  artifact_name_prefix: "benchmark-skipped",
-  artifact_name_suffix: "",
-  benchmark_tool: "none",
-  benchmark_script: "",
-  benchmark_timeout_minutes: 1,
-  pro_env: false,
-  generate_packs: false,
-  server_kind: "none",
-  summary_file: "",
-  summary_title: ""
-}.freeze
-
 # Every app_version any suite accepts. The workflow_dispatch input is constrained
 # to these, but validating guards against a typo or future rename silently
 # selecting no suites (which would skip benchmarks while CI stays green).
@@ -219,6 +197,14 @@ def report_marker(suite, suite_prefix, shard_slug)
   suite[:report_marker] || "<!-- BENCHER_REPORT_#{suite_prefix}_#{shard_slug.upcase.tr('-', '_')} -->"
 end
 
+# No suite selected: emit one row so the matrix `include:` stays non-empty and all
+# rows share keys. Built from a real suite (any will do — its job is gated off via
+# run_benchmark_suites) with the "none" sentinel the gate keys on; the friendlier
+# job_name is all that's user-visible, as the skipped job's label.
+def empty_matrix_row
+  suite_rows(SUITES.first).first.merge(suite_id: "none", job_name: "Benchmark suites skipped")
+end
+
 def build_matrix
   labels = pull_request_labels
   rows = if truthy_env?("BENCHMARK_NON_RUNTIME_ONLY")
@@ -227,7 +213,7 @@ def build_matrix
            SUITES.select { |suite| suite_enabled?(suite, labels) }.flat_map { |suite| suite_rows(suite) }
          end
 
-  { include: rows.empty? ? [EMPTY_MATRIX_ROW] : rows }
+  { include: rows.empty? ? [empty_matrix_row] : rows }
 end
 
 # Only emit when run as a script; `require`-ing the file (e.g. from specs) just
diff --git a/benchmarks/spec/benchmark_matrix_spec.rb b/benchmarks/spec/benchmark_matrix_spec.rb
index 65f6a8ac99..95ac2a61b3 100644
--- a/benchmarks/spec/benchmark_matrix_spec.rb
+++ b/benchmarks/spec/benchmark_matrix_spec.rb
@@ -30,7 +30,9 @@ def suite_ids_for(env)
       )
 
       expect(rows.size).to eq(1)
-      expect(rows.first).to include(suite_id: "none", benchmark_tool: "none")
+      # suite_id "none" is the only load-bearing field (it drives
+      # run_benchmark_suites); the placeholder is otherwise a real row's shape.
+      expect(rows.first).to include(suite_id: "none")
     end
 
     it "forces the placeholder for non-runtime-only changes regardless of other inputs" do

From 85cae02fe05b2735468f9c885c77219cd9cb4bbb Mon Sep 17 00:00:00 2001
From: Alexey Romanov <alexey.v.romanov@gmail.com>
Date: Tue, 2 Jun 2026 15:23:56 +0300
Subject: [PATCH 8/8] Make Vegeta download depend on arch

---
 .github/workflows/benchmark.yml | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/benchmark.yml b/.github/workflows/benchmark.yml
index 3017359d44..db7e35e7ff 100644
--- a/.github/workflows/benchmark.yml
+++ b/.github/workflows/benchmark.yml
@@ -267,9 +267,10 @@ jobs:
       - name: Install Vegeta
         if: matrix.benchmark_tool == 'vegeta' && steps.cache-vegeta.outputs.cache-hit != 'true'
         run: |
-          echo "Installing Vegeta v${VEGETA_VERSION}"
-          wget -q "https://github.com/tsenart/vegeta/releases/download/v${VEGETA_VERSION}/vegeta_${VEGETA_VERSION}_linux_amd64.tar.gz"
-          tar -xzf "vegeta_${VEGETA_VERSION}_linux_amd64.tar.gz"
+          ARCH=$(dpkg --print-architecture)
+          echo "Installing Vegeta v${VEGETA_VERSION} (${ARCH})"
+          wget -q "https://github.com/tsenart/vegeta/releases/download/v${VEGETA_VERSION}/vegeta_${VEGETA_VERSION}_linux_${ARCH}.tar.gz"
+          tar -xzf "vegeta_${VEGETA_VERSION}_linux_${ARCH}.tar.gz"
           mv vegeta ~/bin/
 
       - name: Setup Ruby