step-duration-regression.yml

name: Step Duration Regression

# Fails CI if the action's setup or post step exceeds STEP_MAX_SECONDS.
# Catches regressions of the dangling-Http2Session class of bug that
# silently added ~30s per step.

on:
  workflow_dispatch:
  pull_request:
  push:
    branches: [main]
  schedule:
    - cron: "17 6 * * *"

env:
  STEP_MAX_SECONDS: "5"
  SETUP_STEP_NAME: "Setup Docker Builder under test"

jobs:
  exercise-action:
    name: Run build with setup-docker-builder
    runs-on: blacksmith
    steps:
      - name: Checkout
        uses: actions/checkout@v6

      - name: ${{ env.SETUP_STEP_NAME }}
        uses: ./
        with:
          buildx-version: v0.23.0

      - name: Trivial build
        run: |
          cat > Dockerfile.test <<'EOF'
          FROM alpine:3.20
          RUN echo "step-duration-regression"
          EOF
          docker buildx build -f Dockerfile.test --load -t setup-docker-builder-regression:local .

  validate-step-durations:
    name: Validate step durations
    needs: exercise-action
    if: always()
    runs-on: blacksmith
    permissions:
      actions: read
      contents: read
    steps:
      - name: Assert setup and post steps stay under threshold
        env:
          GH_TOKEN: ${{ github.token }}
          RUN_ID: ${{ github.run_id }}
          REPO: ${{ github.repository }}
          MAX_SECONDS: ${{ env.STEP_MAX_SECONDS }}
          SETUP_STEP_NAME: ${{ env.SETUP_STEP_NAME }}
          TARGET_JOB_NAME: "Run build with setup-docker-builder"
        run: |
          set -euo pipefail

          echo "Fetching steps for run ${RUN_ID} in ${REPO}..."
          jobs_json=$(gh api \
            -H "Accept: application/vnd.github+json" \
            "repos/${REPO}/actions/runs/${RUN_ID}/jobs?per_page=100")

          setup_name="${SETUP_STEP_NAME}"
          post_name="Post ${SETUP_STEP_NAME}"

          # Returns step duration in seconds, or "MISSING".
          step_duration_seconds() {
            local job="$1"
            local step="$2"

            local line
            line=$(echo "$jobs_json" | jq -r \
              --arg job "$job" --arg step "$step" '
                .jobs[] | select(.name == $job) | .steps[] | select(.name == $step) |
                "\(.started_at) \(.completed_at)"
              ')

            local started completed
            started=$(echo "$line" | awk '{print $1}')
            completed=$(echo "$line" | awk '{print $2}')

            if [[ -z "${started:-}" || -z "${completed:-}" \
                  || "${started}" == "null" || "${completed}" == "null" ]]; then
              echo "MISSING"
              return
            fi

            local s c
            s=$(date -u -d "$started" +%s)
            c=$(date -u -d "$completed" +%s)
            echo $(( c - s ))
          }

          echo ""
          echo "=== ${TARGET_JOB_NAME} steps ==="
          echo "$jobs_json" | jq -r --arg job "$TARGET_JOB_NAME" '
            .jobs[] | select(.name == $job) | .steps[] |
            "  step=\"\(.name)\" started=\(.started_at) completed=\(.completed_at) conclusion=\(.conclusion)"
          '

          setup_duration=$(step_duration_seconds "$TARGET_JOB_NAME" "$setup_name")
          post_duration=$(step_duration_seconds  "$TARGET_JOB_NAME" "$post_name")

          echo ""
          echo "Setup step (\"${setup_name}\"): ${setup_duration}s"
          echo "Post  step (\"${post_name}\"):  ${post_duration}s"
          echo "Threshold: ${MAX_SECONDS}s"

          {
            echo "## Step Durations"
            echo ""
            echo "| Step | Duration | Threshold |"
            echo "|---|---:|---:|"
            echo "| Setup (\`${setup_name}\`) | **${setup_duration}s** | ${MAX_SECONDS}s |"
            echo "| Post (\`${post_name}\`)   | **${post_duration}s**  | ${MAX_SECONDS}s |"
          } >> "$GITHUB_STEP_SUMMARY"

          fail=0

          assert_under_threshold() {
            local label="$1" value="$2"
            if [[ "$value" == "MISSING" ]]; then
              echo "::error::Could not find ${label} step in job \"${TARGET_JOB_NAME}\". Did the previous job fail before the step ran?"
              fail=1
              return
            fi
            if (( value > MAX_SECONDS )); then
              echo "::error::${label} step took ${value}s (> ${MAX_SECONDS}s threshold). Check for dangling open handles (Http2Session, axios keep-alive, fs watcher) at the end of the action body."
              fail=1
            fi
          }

          assert_under_threshold "setup" "$setup_duration"
          assert_under_threshold "post"  "$post_duration"

          if (( fail )); then
            exit 1
          fi

          echo ""
          echo "Both steps are within threshold."