tests/slo_workloads: align with ydb-slo-action v2

.github/workflows/slo.yml

@@ -2,275 +2,186 @@ name: SLO
 
 on:
   pull_request:
-    types: [opened, reopened, synchronize]
-    branches:
-      - main
-  workflow_dispatch:
-    inputs:
-      github_issue:
-        description: "GitHub issue number where the SLO results will be reported"
-        required: true
-      baseline_ref:
-        description: "Baseline commit/branch/tag to compare against (leave empty to auto-detect merge-base with main)"
-        required: false
-      slo_workload_duration_seconds:
-        description: "Duration of the SLO workload in seconds"
-        required: false
-        default: "600"
-      slo_workload_read_max_rps:
-        description: "Maximum read RPS for the SLO workload"
-        required: false
-        default: "1000"
-      slo_workload_write_max_rps:
-        description: "Maximum write RPS for the SLO workload"
-        required: false
-        default: "100"
+    types: [opened, reopened, synchronize, labeled]
 
 jobs:
   ydb-slo-action:
+    if: contains(github.event.pull_request.labels.*.name, 'SLO')
+
     name: Run YDB SLO Tests
     runs-on: ubuntu-latest
 
+    permissions:
+      contents: read
+
     strategy:
+      fail-fast: false
       matrix:
-        compiler: [clang, gcc]
-        include:
-          - workload: table
+        sdk:
+          - name: cpp-key-value
+            preset: release-test-clang
+            command: ""
 
     concurrency:
-      group: slo-${{ github.ref }}-${{ matrix.os }}-${{ matrix.workload }}-${{ matrix.compiler }}
+      group: slo-${{ github.ref }}-${{ matrix.sdk.name }}
       cancel-in-progress: true
 
     steps:
       - name: Install dependencies
         run: |
+          set -euxo pipefail
           YQ_VERSION=v4.48.2
           BUILDX_VERSION=0.30.1
           COMPOSE_VERSION=2.40.3
 
-          sudo curl -L https://github.com/mikefarah/yq/releases/download/${YQ_VERSION}/yq_linux_amd64 -o /usr/local/bin/yq && \
-            sudo chmod +x /usr/local/bin/yq
+          sudo curl -fLo /usr/local/bin/yq \
+            "https://github.com/mikefarah/yq/releases/download/${YQ_VERSION}/yq_linux_amd64"
+          sudo chmod +x /usr/local/bin/yq
 
-          echo "Updating Docker plugins..."
           sudo mkdir -p /usr/local/lib/docker/cli-plugins
 
-          echo "Installing Docker Buildx ${BUILDX_VERSION}..."
           sudo curl -fLo /usr/local/lib/docker/cli-plugins/docker-buildx \
             "https://github.com/docker/buildx/releases/download/v${BUILDX_VERSION}/buildx-v${BUILDX_VERSION}.linux-amd64"
           sudo chmod +x /usr/local/lib/docker/cli-plugins/docker-buildx
 
-          echo "Installing Docker Compose ${COMPOSE_VERSION}..."
           sudo curl -fLo /usr/local/lib/docker/cli-plugins/docker-compose \
             "https://github.com/docker/compose/releases/download/v${COMPOSE_VERSION}/docker-compose-linux-x86_64"
           sudo chmod +x /usr/local/lib/docker/cli-plugins/docker-compose
 
-          echo "Installed versions:"
           yq --version
           docker --version
           docker buildx version
           docker compose version
 
-      - name: Checkout current version
+      - name: Checkout current SDK version
         uses: actions/checkout@v5
         with:
-          path: current
+          path: sdk-current
           fetch-depth: 0
           submodules: true
 
       - name: Determine baseline commit
         id: baseline
+        working-directory: sdk-current
         run: |
-          cd current
-          if [[ -n "${{ inputs.baseline_ref }}" ]]; then
-            BASELINE="${{ inputs.baseline_ref }}"
-          else
-            BASELINE=$(git merge-base HEAD origin/main)
-          fi
-          echo "sha=$BASELINE" >> $GITHUB_OUTPUT
+          set -euo pipefail
+          BASELINE=$(git merge-base HEAD origin/main)
+          echo "sha=${BASELINE}" >> "$GITHUB_OUTPUT"
 
-          # Try to determine a human-readable ref name for baseline
-          # Check if baseline is on main
-          if git merge-base --is-ancestor $BASELINE origin/main && \
-             [ "$(git rev-parse origin/main)" = "$BASELINE" ]; then
+          if git merge-base --is-ancestor "${BASELINE}" origin/main && \
+             [ "$(git rev-parse origin/main)" = "${BASELINE}" ]; then
             BASELINE_REF="main"
           else
-            # Try to find a branch containing this commit
-            BRANCH=$(git branch -r --contains $BASELINE | grep -v HEAD | head -1 | sed 's/.*\///' || echo "")
-            if [ -n "$BRANCH" ]; then
+            BRANCH=$(git branch -r --contains "${BASELINE}" | grep -v HEAD | head -1 | sed 's|.*/||' || echo "")
+            if [ -n "${BRANCH}" ]; then
               BASELINE_REF="${BRANCH}@${BASELINE:0:7}"
             else
               BASELINE_REF="${BASELINE:0:7}"
             fi
           fi
-          echo "ref=$BASELINE_REF" >> $GITHUB_OUTPUT
+          echo "ref=${BASELINE_REF}" >> "$GITHUB_OUTPUT"
 
-      - name: Checkout baseline version
+      - name: Checkout baseline SDK version
         uses: actions/checkout@v5
         with:
           ref: ${{ steps.baseline.outputs.sha }}
-          path: baseline
+          path: sdk-baseline
           fetch-depth: 1
           submodules: true
 
-      - name: Build Workload Image
-        run: |
-          echo "Cleaning up Docker system before builds..."
-          docker system prune -af --volumes
-          docker builder prune -af
-          df -h
-
-          # Build current version
-          if [ -f "$GITHUB_WORKSPACE/current/tests/slo_workloads/Dockerfile" ]; then
-            echo "Building current app image..."
-            cd "$GITHUB_WORKSPACE/current"
-
-            # Use SLO-specific .dockerignore
-            cp tests/slo_workloads/.dockerignore .dockerignore
-
-            docker build -t ydb-app-current \
-              --build-arg REF="${{ github.head_ref || github.ref_name }}" \
-              --build-arg PRESET=release-test-${{ matrix.compiler }} \
-              -f tests/slo_workloads/Dockerfile .
-
-            # Clean up .dockerignore
-            rm -f .dockerignore
-          else
-            echo "No current app Dockerfile found"
-            exit 1
-          fi
-
-          docker system prune -f --volumes
-          docker builder prune -af
-
-          # Build baseline version
-          if [ -f "$GITHUB_WORKSPACE/baseline/tests/slo_workloads/Dockerfile" ]; then
-            echo "Building baseline app image..."
-            cd "$GITHUB_WORKSPACE/baseline"
-
-            # Use SLO-specific .dockerignore
-            cp tests/slo_workloads/.dockerignore .dockerignore
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
 
-            docker build -t ydb-app-baseline \
-              --build-arg REF="${{ steps.baseline.outputs.ref }}" \
-              --build-arg PRESET=release-test-${{ matrix.compiler }} \
-              -f tests/slo_workloads/Dockerfile .
-
-            # Clean up .dockerignore
-            rm -f .dockerignore
-          else
-            echo "No baseline app Dockerfile found"
-            exit 1
-          fi
-
-          docker system prune -f --volumes
-          docker builder prune -af
+      # Use current's workload harness (Dockerfile, sources, .dockerignore) for
+      # both builds so only the SDK library differs between current and
+      # baseline. Without this the baseline image picks up the harness from
+      # the merge-base commit, which can lag behind the action's contract.
+      # buildx also expects .dockerignore at the context root, not under
+      # tests/, so copy it up in each checkout.
+      - name: Stage workload harness
+        run: |
+          set -euxo pipefail
+          rm -rf sdk-baseline/tests/slo_workloads
+          cp -a sdk-current/tests/slo_workloads sdk-baseline/tests/slo_workloads
+          cp sdk-current/tests/slo_workloads/.dockerignore sdk-current/.dockerignore
+          cp sdk-baseline/tests/slo_workloads/.dockerignore sdk-baseline/.dockerignore
+
+      # `cache-to: type=gha` does NOT export `--mount=type=cache` content, so
+      # ccache state is lost between runs. Persist /root/.ccache via host
+      # directory + cache-dance: actions/cache restores the host dir, the
+      # dance injects it into the BuildKit cache mount before the build and
+      # extracts the updated state afterwards for the next save.
+      - name: Restore ccache
+        id: ccache
+        uses: actions/cache@v4
+        with:
+          path: ccache
+          key: slo-ccache-${{ matrix.sdk.preset }}-${{ github.run_id }}
+          restore-keys: |
+            slo-ccache-${{ matrix.sdk.preset }}-
 
-          echo "Final disk space after builds:"
-          df -h
+      - name: Inject ccache into BuildKit
+        uses: reproducible-containers/buildkit-cache-dance@v3.1.2
+        with:
+          cache-map: |
+            {
+              "ccache": "/root/.ccache"
+            }
+          # Always extract so newly-compiled TUs from this run are saved by
+          # actions/cache (key uses ${{ github.run_id }}, so each run gets
+          # its own snapshot). Without extraction the cache stays frozen at
+          # whatever was first persisted.
+          skip-extraction: false
+
+      # A clean build of the SLO image takes ~30 min because the Dockerfile
+      # rebuilds the full C++ toolchain + abseil/protobuf/grpc from source.
+      # The GHA cache lets subsequent runs reuse every layer up to the SDK
+      # source COPY, so only the actual workload link step reruns (~3 min).
+      - name: Build current workload image
+        uses: docker/build-push-action@v6
+        with:
+          context: sdk-current
+          file: sdk-current/tests/slo_workloads/Dockerfile
+          platforms: linux/amd64
+          tags: ydb-app-current
+          load: true
+          build-args: PRESET=${{ matrix.sdk.preset }}
+          cache-from: type=gha,scope=slo-${{ matrix.sdk.preset }}
+          cache-to: type=gha,mode=max,scope=slo-${{ matrix.sdk.preset }}
+
+      - name: Build baseline workload image
+        id: baseline-build
+        continue-on-error: true
+        uses: docker/build-push-action@v6
+        with:
+          context: sdk-baseline
+          file: sdk-baseline/tests/slo_workloads/Dockerfile
+          platforms: linux/amd64
+          tags: ydb-app-baseline
+          load: true
+          build-args: PRESET=${{ matrix.sdk.preset }}
+          cache-from: type=gha,scope=slo-${{ matrix.sdk.preset }}
+
+      # If the historical commit lacks the SLO Dockerfile or can't compile,
+      # reuse the current image so the SLO run is still comparable against
+      # itself rather than failing outright.
+      - name: Fall back to current image for baseline
+        if: steps.baseline-build.outcome == 'failure'
+        run: |
+          echo "Baseline build failed; reusing current image as baseline."
+          docker tag ydb-app-current ydb-app-baseline
 
-      - name: Initialize YDB SLO
-        uses: ydb-platform/ydb-slo-action/init@main
+      - name: Run SLO Tests
+        uses: ydb-platform/ydb-slo-action/init@v2
+        timeout-minutes: 30
         with:
-          github_issue: ${{ github.event.inputs.github_issue }}
+          github_issue: ${{ github.event.pull_request.number }}
           github_token: ${{ secrets.GITHUB_TOKEN }}
-          workload_name: ${{ matrix.workload }}-${{ matrix.compiler }}
+          workload_name: ${{ matrix.sdk.name }}
+          workload_duration: "600"
           workload_current_ref: ${{ github.head_ref || github.ref_name }}
+          workload_current_image: ydb-app-current
+          workload_current_command: ${{ matrix.sdk.command }} --read-rps 1000 --write-rps 100
           workload_baseline_ref: ${{ steps.baseline.outputs.ref }}
-
-      - name: Prepare SLO Database
-        run: |
-          echo "Preparing SLO database..."
-          docker run --rm --network ydb_ydb-net \
-            --add-host "ydb:172.28.0.11" \
-            --add-host "ydb:172.28.0.12" \
-            --add-host "ydb:172.28.0.13" \
-            --add-host "ydb:172.28.0.99" \
-            ydb-app-current --connection-string grpc://ydb:2136/?database=/Root/testdb create --dont-push
-
-      - name: Run SLO Tests (parallel)
-        timeout-minutes: 15
-        run: |
-          DURATION=${{ inputs.slo_workload_duration_seconds || 600 }}
-          READ_RPS=${{ inputs.slo_workload_read_max_rps || 1000 }}
-          WRITE_RPS=${{ inputs.slo_workload_write_max_rps || 100 }}
-
-          ARGS="--connection-string grpc://ydb:2136/?database=/Root/testdb run \
-            --metrics-push-url http://prometheus:9090/api/v1/otlp/v1/metrics \
-            --time $DURATION \
-            --read-rps $READ_RPS \
-            --write-rps $WRITE_RPS \
-            --read-timeout 100 \
-            --write-timeout 100"
-
-          echo "Starting ydb-app-current..."
-          docker run -d \
-            --name ydb-app-current \
-            --network ydb_ydb-net \
-            --add-host "ydb:172.28.0.11" \
-            --add-host "ydb:172.28.0.12" \
-            --add-host "ydb:172.28.0.13" \
-            --add-host "ydb:172.28.0.99" \
-            ydb-app-current $ARGS
-
-          echo "Starting ydb-app-baseline..."
-          docker run -d \
-            --name ydb-app-baseline \
-            --network ydb_ydb-net \
-            --add-host "ydb:172.28.0.11" \
-            --add-host "ydb:172.28.0.12" \
-            --add-host "ydb:172.28.0.13" \
-            --add-host "ydb:172.28.0.99" \
-            ydb-app-baseline $ARGS
-
-          # Show initial logs
-          echo ""
-          echo "==================== INITIAL CURRENT LOGS ===================="
-          docker logs -n 15 ydb-app-current 2>&1 || echo "No current container"
-          echo ""
-          echo "==================== INITIAL BASELINE LOGS ===================="
-          docker logs -n 15 ydb-app-baseline 2>&1 || echo "No baseline container"
-          echo ""
-
-          # Wait for workloads to complete
-          echo "Waiting for workloads to complete (${DURATION}s)..."
-          sleep ${DURATION}
-
-          # Stop containers after workload duration and wait for graceful shutdown
-          echo "Stopping containers after ${DURATION}s..."
-          docker stop --timeout=30 ydb-app-current ydb-app-baseline 2>&1 || true
-
-          # Force kill if still running
-          docker kill ydb-app-current ydb-app-baseline 2>&1 || true
-
-          # Check exit codes
-          CURRENT_EXIT=$(docker inspect ydb-app-current --format='{{.State.ExitCode}}' 2>/dev/null || echo "1")
-          BASELINE_EXIT=$(docker inspect ydb-app-baseline --format='{{.State.ExitCode}}' 2>/dev/null || echo "0")
-
-          echo "Current container exit code: $CURRENT_EXIT"
-          echo "Baseline container exit code: $BASELINE_EXIT"
-
-          # Show final logs
-          echo ""
-          echo "==================== FINAL CURRENT LOGS ===================="
-          docker logs -n 15 ydb-app-current 2>&1 || echo "No current container"
-          echo ""
-          echo "==================== FINAL BASELINE LOGS ===================="
-          docker logs -n 15 ydb-app-baseline 2>&1 || echo "No baseline container"
-          echo ""
-
-          echo "SUCCESS: Workloads completed successfully"
-
-      - if: always()
-        name: Store logs
-        run: |
-          docker logs ydb-app-current > current.log 2>&1 || echo "No current container"
-          docker logs ydb-app-baseline > baseline.log 2>&1 || echo "No baseline container"
-
-      - if: always()
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.workload }}-${{ matrix.compiler }}-slo-cpp-sdk-logs
-          path: |
-            ./current.log
-            ./baseline.log
-          retention-days: 1
+          workload_baseline_image: ydb-app-baseline
+          workload_baseline_command: ${{ matrix.sdk.command }} --read-rps 1000 --write-rps 100