parakeet_gpu_tests.yml

name: Parakeet GPU Tests (Smoke + DER + WER + Concurrency)

on:
  workflow_dispatch:
    inputs:
      image_tag:
        description: 'Docker image tag to test (e.g. abc1234)'
        required: true
      der_report_only:
        description: 'DER benchmark report-only mode (true = no fail gate)'
        required: false
        default: 'false'
      wer_report_only:
        description: 'WER gate report-only mode (true = no fail gate)'
        required: false
        default: 'false'
  schedule:
    # Nightly at 03:00 UTC — tests latest dev image
    - cron: '0 3 * * *'

env:
  SERVICE: parakeet
  REGION: us-central1
  NAMESPACE: dev-omi-backend
  JOB_NAME: parakeet-gpu-test-${{ github.run_id }}

jobs:
  gpu-tests:
    environment: development
    permissions:
      contents: 'read'
      id-token: 'write'

    runs-on: ubuntu-latest
    timeout-minutes: 35

    steps:
      - name: Checkout
        uses: actions/checkout@v7

      - name: Google Auth
        id: auth
        uses: 'google-github-actions/auth@v3'
        with:
          credentials_json: ${{ secrets.GCP_CREDENTIALS }}

      - name: Set up gcloud
        uses: google-github-actions/setup-gcloud@v3

      - name: Get GKE credentials
        uses: google-github-actions/get-gke-credentials@v3
        with:
          cluster_name: dev-omi-gke
          location: ${{ env.REGION }}
          project_id: based-hardware-dev

      - name: Resolve image tag
        id: tag
        run: |
          if [[ -n "${{ github.event.inputs.image_tag }}" ]]; then
            echo "tag=${{ github.event.inputs.image_tag }}" >> "$GITHUB_OUTPUT"
          else
            # Nightly: use latest dev image tag from running deployment
            TAG=$(kubectl -n ${{ env.NAMESPACE }} get deploy dev-omi-parakeet \
              -o jsonpath='{.spec.template.spec.containers[0].image}' | cut -d: -f2)
            echo "tag=$TAG" >> "$GITHUB_OUTPUT"
          fi
          echo "Using image tag: $(cat $GITHUB_OUTPUT | grep tag | cut -d= -f2)"

      - name: Clean up previous test job
        run: |
          kubectl -n ${{ env.NAMESPACE }} delete job ${{ env.JOB_NAME }} --ignore-not-found=true
          sleep 5

      - name: Create GPU test job
        run: |
          cat <<'JOBEOF' | envsubst '$JOB_NAME $NAMESPACE $IMAGE_TAG $DER_REPORT_ONLY $WER_REPORT_ONLY' | kubectl apply -f -
          apiVersion: batch/v1
          kind: Job
          metadata:
            name: ${JOB_NAME}
            namespace: ${NAMESPACE}
          spec:
            activeDeadlineSeconds: 1800
            backoffLimit: 0
            template:
              spec:
                restartPolicy: Never
                containers:
                  - name: gpu-test
                    image: gcr.io/based-hardware-dev/parakeet:${IMAGE_TAG}
                    command:
                      - bash
                      - -c
                      - |
                        set -e
                        echo "=== Parakeet GPU Test Suite ==="
                        echo "Image: gcr.io/based-hardware-dev/parakeet:${IMAGE_TAG}"
                        echo ""

                        echo "--- Phase 1: Container Smoke Tests (CPU) ---"
                        python -m pytest tests/container/test_parakeet_smoke.py -v \
                          -k "not TestGPUInference" \
                          --tb=short 2>&1
                        echo ""

                        echo "--- Phase 2: Contract Tests ---"
                        python -m pytest tests/container/test_parakeet_contract.py -v \
                          --tb=short 2>&1
                        echo ""

                        echo "--- Phase 3: GPU Smoke Tests ---"
                        python -m pytest tests/container/test_parakeet_smoke.py -v \
                          -k "TestGPUInference" \
                          --tb=short 2>&1
                        echo ""

                        echo "--- Phase 4: DER Regression Gate (synthetic) ---"
                        python -m pytest tests/container/test_parakeet_der_gate.py -v -s \
                          --tb=short 2>&1
                        echo ""

                        echo "--- Phase 5: Real-speech DER Benchmark (VoxConverse) ---"
                        echo "Starting uvicorn for benchmark..."
                        uvicorn main:app --host 127.0.0.1 --port 8080 --loop uvloop &
                        SERVER_PID=$!
                        for i in $(seq 1 60); do
                          if python -c "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8080/health')" 2>/dev/null; then
                            echo "Server ready after ${i}s"
                            break
                          fi
                          sleep 1
                        done

                        set +e
                        DER_REPORT_ONLY=${DER_REPORT_ONLY:-false} \
                        DER_RESULTS_PATH=/tmp/parakeet-gpu-artifacts/der-benchmark-results.json \
                        python -m pytest tests/container/test_parakeet_der_benchmark.py -v -s \
                          --tb=short 2>&1
                        BENCH_EXIT=$?
                        echo "--- DER Diagnostics Artifact ---"
                        if [ -f /tmp/parakeet-gpu-artifacts/der-benchmark-results.json ]; then
                          echo "DER_DIAGNOSTICS_JSON_BEGIN"
                          cat /tmp/parakeet-gpu-artifacts/der-benchmark-results.json
                          echo "DER_DIAGNOSTICS_JSON_END"
                        else
                          echo "DER diagnostics artifact was not produced"
                        fi
                        set -e

                        if [ $BENCH_EXIT -ne 0 ]; then
                          echo "DER benchmark failed (exit=$BENCH_EXIT)"
                          kill $SERVER_PID 2>/dev/null || true
                          wait $SERVER_PID 2>/dev/null || true
                          exit $BENCH_EXIT
                        fi
                        echo ""

                        echo "--- Phase 6: Concurrency Stress Test ---"
                        set +e
                        CONCURRENCY_LEVEL=8 CONCURRENCY_ROUNDS=3 \
                        python -m pytest tests/container/test_parakeet_concurrency.py -v -s \
                          --tb=short 2>&1
                        CONC_EXIT=$?
                        set -e

                        if [ $CONC_EXIT -ne 0 ]; then
                          echo "Concurrency test failed (exit=$CONC_EXIT)"
                          kill $SERVER_PID 2>/dev/null || true
                          wait $SERVER_PID 2>/dev/null || true
                          exit $CONC_EXIT
                        fi
                        echo ""

                        echo "--- Phase 7: WER Regression Gate (LibriSpeech) ---"
                        set +e
                        WER_REPORT_ONLY=${WER_REPORT_ONLY:-false} \
                        python -m pytest tests/container/test_parakeet_wer_gate.py -v -s \
                          --tb=short 2>&1
                        WER_EXIT=$?
                        set -e

                        if [ $WER_EXIT -ne 0 ]; then
                          echo "WER gate failed (exit=$WER_EXIT)"
                          kill $SERVER_PID 2>/dev/null || true
                          wait $SERVER_PID 2>/dev/null || true
                          exit $WER_EXIT
                        fi
                        echo ""

                        echo "--- Phase 8: High-Concurrency Stress Test ---"
                        set +e
                        HIGH_CONCURRENCY_LEVEL=16 HIGH_CONCURRENCY_ROUNDS=5 \
                        python -m pytest tests/container/test_parakeet_high_concurrency.py -v -s \
                          --tb=short 2>&1
                        HCONC_EXIT=$?
                        set -e

                        kill $SERVER_PID 2>/dev/null || true
                        wait $SERVER_PID 2>/dev/null || true

                        if [ $HCONC_EXIT -ne 0 ]; then
                          echo "High-concurrency test failed (exit=$HCONC_EXIT)"
                          exit $HCONC_EXIT
                        fi
                        echo ""

                        echo "=== ALL TESTS PASSED ==="
                    env:
                      - name: HUGGINGFACE_TOKEN
                        valueFrom:
                          secretKeyRef:
                            name: dev-omi-backend-secrets
                            key: HUGGINGFACE_TOKEN
                      - name: PARAKEET_MODEL
                        value: "nvidia/parakeet-tdt-0.6b-v3"
                      - name: PARAKEET_INFERENCE_MODE
                        value: "nemo"
                      - name: NVIDIA_VISIBLE_DEVICES
                        value: "all"
                      - name: DER_REPORT_ONLY
                        value: "${DER_REPORT_ONLY}"
                      - name: WER_REPORT_ONLY
                        value: "${WER_REPORT_ONLY}"
                    resources:
                      requests:
                        cpu: 1
                        memory: 8Gi
                        nvidia.com/gpu: 1
                      limits:
                        cpu: 2
                        memory: 20Gi
                        nvidia.com/gpu: 1
                affinity:
                  nodeAffinity:
                    requiredDuringSchedulingIgnoredDuringExecution:
                      nodeSelectorTerms:
                        - matchExpressions:
                            - key: service
                              operator: In
                              values:
                                - parakeet
                            - key: env
                              operator: In
                              values:
                                - dev
          JOBEOF
        env:
          IMAGE_TAG: ${{ steps.tag.outputs.tag }}
          DER_REPORT_ONLY: ${{ github.event.inputs.der_report_only || 'false' }}
          WER_REPORT_ONLY: ${{ github.event.inputs.wer_report_only || 'false' }}

      - name: Wait for job completion
        run: |
          echo "Waiting for GPU test job to complete (timeout: 30 min)..."
          kubectl -n ${{ env.NAMESPACE }} wait --for=condition=complete \
            --timeout=1800s job/${{ env.JOB_NAME }} 2>&1 || true

          # Check job status
          STATUS=$(kubectl -n ${{ env.NAMESPACE }} get job ${{ env.JOB_NAME }} \
            -o jsonpath='{.status.conditions[0].type}')
          echo "Job status: $STATUS"

      - name: Collect test logs
        if: always()
        run: |
          mkdir -p artifacts/parakeet
          echo "=== GPU Test Job Logs ==="
          kubectl -n ${{ env.NAMESPACE }} logs job/${{ env.JOB_NAME }} 2>&1 | tee artifacts/parakeet/gpu-test.log || true
          echo ""
          echo "=== Job/Pod Details ==="
          kubectl -n ${{ env.NAMESPACE }} describe job/${{ env.JOB_NAME }} 2>&1 || true
          kubectl -n ${{ env.NAMESPACE }} get events --field-selector involvedObject.name=${{ env.JOB_NAME }} 2>&1 || true

      - name: Extract DER diagnostics artifact
        if: always()
        run: |
          python - <<'PY'
          from pathlib import Path

          log_path = Path("artifacts/parakeet/gpu-test.log")
          output_path = Path("artifacts/parakeet/der-benchmark-results.json")
          text = log_path.read_text() if log_path.exists() else ""
          start = "DER_DIAGNOSTICS_JSON_BEGIN"
          end = "DER_DIAGNOSTICS_JSON_END"
          start_index = text.find(start)
          end_index = text.find(end, start_index + len(start)) if start_index != -1 else -1
          if start_index == -1 or end_index == -1:
              print("DER diagnostics JSON markers not found in order; skipping artifact extraction")
              raise SystemExit(0)
          json_text = text[start_index + len(start) : end_index].strip()
          output_path.write_text(json_text + "\n")
          print(f"Wrote {output_path}")
          PY

      - name: Upload DER diagnostics artifact
        if: always()
        uses: actions/upload-artifact@v7
        with:
          name: parakeet-der-diagnostics-${{ github.run_id }}-${{ github.run_attempt }}
          path: artifacts/parakeet/der-benchmark-results.json
          if-no-files-found: ignore
          retention-days: 30

      - name: Check job result
        run: |
          SUCCEEDED=$(kubectl -n ${{ env.NAMESPACE }} get job ${{ env.JOB_NAME }} \
            -o jsonpath='{.status.succeeded}')
          FAILED=$(kubectl -n ${{ env.NAMESPACE }} get job ${{ env.JOB_NAME }} \
            -o jsonpath='{.status.failed}')

          if [[ "$SUCCEEDED" == "1" ]]; then
            echo "✅ All GPU tests passed"
          else
            echo "❌ GPU tests failed (succeeded=$SUCCEEDED, failed=$FAILED)"
            # Print failure details
            kubectl -n ${{ env.NAMESPACE }} logs job/${{ env.JOB_NAME }} --tail=50 2>&1 || true
            exit 1
          fi

      - name: Clean up test job
        if: always()
        run: |
          kubectl -n ${{ env.NAMESPACE }} delete job ${{ env.JOB_NAME }} --ignore-not-found=true