Refactor SLO workload to use environment variables

.github/workflows/slo-report.yml

@@ -16,42 +16,7 @@ jobs:
       pull-requests: write
     steps:
       - name: Publish YDB SLO Report
-        uses: ydb-platform/ydb-slo-action/report@13c687b7d4b2879da79dd12932dee0ed2b65dd1c
+        uses: ydb-platform/ydb-slo-action/report@v2
         with:
           github_token: ${{ secrets.GITHUB_TOKEN }}
           github_run_id: ${{ github.event.workflow_run.id }}
-
-  remove-slo-label:
-    if: always() && github.event.workflow_run.event == 'pull_request'
-    name: Remove SLO Label
-    needs: ydb-slo-action-report
-    runs-on: ubuntu-latest
-    permissions:
-      pull-requests: write
-    steps:
-      - name: Remove SLO label from PR
-        uses: actions/github-script@v7
-        with:
-          script: |
-            const pullRequests = context.payload.workflow_run.pull_requests;
-            if (pullRequests && pullRequests.length > 0) {
-              for (const pr of pullRequests) {
-                try {
-                  await github.rest.issues.removeLabel({
-                    owner: context.repo.owner,
-                    repo: context.repo.repo,
-                    issue_number: pr.number,
-                    name: 'SLO'
-                  });
-                  console.log(`Removed SLO label from PR #${pr.number}`);
-                } catch (error) {
-                  if (error.status === 404) {
-                    console.log(`SLO label not found on PR #${pr.number}, skipping`);
-                  } else {
-                    throw error;
-                  }
-                }
-              }
-            } else {
-              console.log('No pull requests associated with this workflow run');
-            }

.github/workflows/slo.yml

@@ -42,16 +42,14 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        include:
-          - id: sync-table
-            prefix: table
-            workload: sync-table
-          - id: sync-query
-            prefix: table
-            workload: sync-query
+        sdk:
+          - name: sync-table
+            command: "--read-rps ${{ inputs.slo_workload_read_max_rps || '1000' }} --write-rps ${{ inputs.slo_workload_write_max_rps || '100' }}"
+          - name: sync-query
+            command: "--read-rps ${{ inputs.slo_workload_read_max_rps || '1000' }} --write-rps ${{ inputs.slo_workload_write_max_rps || '100' }}"
 
     concurrency:
-      group: slo-${{ github.ref }}-${{ matrix.workload }}
+      group: slo-${{ github.ref }}-${{ matrix.sdk.name }}
       cancel-in-progress: true
 
     steps:
@@ -141,125 +139,32 @@ jobs:
             -t "ydb-app-baseline" \
             "$GITHUB_WORKSPACE/baseline"
 
-      - name: Initialize YDB SLO
-        id: ydb_slo
-        uses: ydb-platform/ydb-slo-action/init@13c687b7d4b2879da79dd12932dee0ed2b65dd1c
+      - name: Run SLO Tests
+        uses: ydb-platform/ydb-slo-action/init@v2
+        timeout-minutes: 30
         with:
-          github_issue: ${{ github.event.pull_request.number || inputs.github_issue }}
+          github_issue: ${{ github.event.inputs.github_issue }}
           github_token: ${{ secrets.GITHUB_TOKEN }}
-          workload_name: ydb-python-${{ matrix.workload }}
+          workload_name: ${{ matrix.sdk.name }}
+          workload_duration: ${{ inputs.slo_workload_duration_seconds || '600' }}
           workload_current_ref: ${{ github.head_ref || github.ref_name }}
+          workload_current_image: ydb-app-current
+          workload_current_command: ${{ matrix.sdk.command }}
           workload_baseline_ref: ${{ steps.baseline.outputs.ref }}
-
-      - name: Prepare SLO Database
-        run: |
-          docker run --rm \
-            --network ydb_ydb-net \
-            --add-host "ydb:172.28.0.11" \
-            --add-host "ydb:172.28.0.12" \
-            --add-host "ydb:172.28.0.13" \
-            --add-host "ydb:172.28.0.99" \
-            -e "WORKLOAD=${{ matrix.workload }}" \
-            -e "REF=${{ github.head_ref || github.ref_name }}" \
-            ydb-app-current \
-            ${{ matrix.prefix }}-create grpc://ydb:2136 /Root/testdb
-
-      - name: Run SLO Tests (current + baseline in parallel)
-        timeout-minutes: 15
-        env:
-          WORKLOAD: ${{ matrix.workload }}
-          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 }}
-          CURRENT_REF: ${{ github.head_ref || github.ref_name }}
-          BASELINE_REF: ${{ steps.baseline.outputs.ref }}
-        run: |
-          ARGS="${{ matrix.prefix }}-run grpc://ydb:2136 /Root/testdb \
-            --otlp-endpoint http://prometheus:9090/api/v1/otlp/v1/metrics \
-            --report-period 250 \
-            --time ${DURATION} \
-            --read-rps ${READ_RPS} \
-            --write-rps ${WRITE_RPS} \
-            --read-timeout 1000 \
-            --write-timeout 1000"
-
-          echo "Starting current workload (ref=${CURRENT_REF}, workload=${WORKLOAD})..."
-          docker run -d \
-            --name ydb-app-current \
-            --network ydb_ydb-net \
-            --add-host "ydb:172.28.0.11" \
-            --add-host "ydb:172.28.0.12" \
-            --add-host "ydb:172.28.0.13" \
-            --add-host "ydb:172.28.0.99" \
-            -e "REF=${CURRENT_REF}" \
-            -e "WORKLOAD=${WORKLOAD}" \
-            ydb-app-current \
-            $ARGS
-
-          echo "Starting baseline workload (ref=${BASELINE_REF}, workload=${WORKLOAD})..."
-          docker run -d \
-            --name ydb-app-baseline \
-            --network ydb_ydb-net \
-            --add-host "ydb:172.28.0.11" \
-            --add-host "ydb:172.28.0.12" \
-            --add-host "ydb:172.28.0.13" \
-            --add-host "ydb:172.28.0.99" \
-            -e "REF=${BASELINE_REF}" \
-            -e "WORKLOAD=${WORKLOAD}" \
-            ydb-app-baseline \
-            $ARGS
-
-          echo ""
-          echo "==================== INITIAL CURRENT LOGS ===================="
-          docker logs -n 15 ydb-app-current 2>&1 || echo "No current container"
-          echo ""
-          echo "==================== INITIAL BASELINE LOGS ===================="
-          docker logs -n 15 ydb-app-baseline 2>&1 || echo "No baseline container"
-          echo ""
-
-          echo "Waiting for workloads to complete (${DURATION}s)..."
-          sleep ${DURATION}
-
-          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 "1")
-
-          echo "Current exit code: ${CURRENT_EXIT}"
-          echo "Baseline exit code: ${BASELINE_EXIT}"
-
-          echo ""
-          echo "==================== FINAL CURRENT LOGS ===================="
-          docker logs -n 15 ydb-app-current 2>&1 || echo "No current container"
-          echo ""
-          echo "==================== FINAL BASELINE LOGS ===================="
-          docker logs -n 15 ydb-app-baseline 2>&1 || echo "No baseline container"
-          echo ""
-
-          if [[ "${CURRENT_EXIT}" != "0" || "${BASELINE_EXIT}" != "0" ]]; then
-            echo "One or both workloads failed."
-            exit 0
-          fi
-
-          echo "SUCCESS: Workloads completed successfully"
-
-      - if: always()
-        name: Store logs
-        run: |
-          docker logs ydb-app-current > current.log 2>&1 || echo "No current container" > current.log
-          docker logs ydb-app-baseline > baseline.log 2>&1 || echo "No baseline container" > baseline.log
-
-      - if: always()
-        name: Upload logs
-        uses: actions/upload-artifact@v4
+          workload_baseline_image: ydb-app-current
+          workload_baseline_command: ${{ matrix.sdk.command }}
+
+  ydb-slo-action-report:
+    runs-on: ubuntu-latest
+    name: Publish YDB SLO Report
+    needs: ydb-slo-action
+    permissions:
+      checks: write
+      contents: read
+      pull-requests: write
+    steps:
+      - name: Publish YDB SLO Report
+        uses: ydb-platform/ydb-slo-action/report@v2
         with:
-          name: ydb-python-${{ matrix.workload }}-logs
-          path: |
-            ./current.log
-            ./baseline.log
-          retention-days: 1
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+          github_run_id: ${{ github.run_id }}

pyproject.toml

@@ -1,3 +1,9 @@
+[tool.ty.environment]
+extra-paths = ["tests/slo/src"]
+
+[tool.ruff]
+line-length = 120
+
 [tool.black]
 line-length = 120
 

tests/slo/Dockerfile

@@ -1,13 +1,21 @@
 # syntax=docker/dockerfile:1
 
 # This image packages the Python SLO workload runner.
-# It expects to be run with arguments like:
-#   docker run --rm <image> table-run <endpoint> <db> --otlp-endpoint http://prometheus:9090/api/v1/otlp/v1/metrics ...
+#
+# Connection and workload identity are configured via environment variables:
+#   YDB_ENDPOINT          grpc://ydb:2136
+#   YDB_DATABASE          /Root/testdb
+#   WORKLOAD_DURATION     600
+#   WORKLOAD_NAME         sync-query | sync-table | topic
+#   WORKLOAD_REF          <branch/ref label for metrics>
+#   OTEL_EXPORTER_OTLP_METRICS_ENDPOINT  http://ydb-prometheus:9090/api/v1/otlp/v1/metrics
+#
+# Additional tuning flags (read/write RPS, timeouts, thread counts) are passed via
+# the Docker CMD, e.g.:
+#   docker run --rm --env-file ... <image> --read-rps 1000 --write-rps 100
 #
 # Notes:
 # - OpenTelemetry 1.39.x requires Python >= 3.9.
-# - The entrypoint is `python ./tests/slo/src`, i.e. it runs the `__main__.py`
-#   from that directory (same as `python tests/slo/src ...` in CI).
 
 FROM python:3.11-slim AS build
 
@@ -36,3 +44,5 @@ COPY --from=build /opt/venv /opt/venv
 COPY --from=build /src/tests/slo/src /app/tests/slo/src
 
 ENTRYPOINT ["python", "./tests/slo/src"]
+
+CMD ["--read-rps", "1000", "--write-rps", "100"]

tests/slo/requirements.txt

@@ -1,6 +1,6 @@
 requests==2.33.0
 aiolimiter==1.1.0
-quantile-estimator==0.1.2
+hdrhistogram==0.10.3
 
 # OpenTelemetry (OTLP/HTTP exporter)
 # NOTE: OpenTelemetry 1.39.1 requires Python >= 3.9.

tests/slo/src/__main__.py

@@ -2,8 +2,7 @@
 import logging
 
 from options import parse_options
-from root_runner import run_from_args
-
+from root_runner import run_all
 
 if __name__ == "__main__":
     args = parse_options()
@@ -12,4 +11,4 @@
     log_level = logging.DEBUG if args.debug else logging.INFO
     logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)-8s %(message)s")
 
-    run_from_args(args)
+    run_all(args)