cache-probe.yaml

name: Cache Probe

on:
  push:
    branches:
      - agent/cache-probe-*
  workflow_dispatch:

permissions:
  contents: read
  packages: write

jobs:
  cw-cluster-start:
    name: CW Cluster Start (${{ matrix.mode }})
    runs-on: ubuntu-latest
    timeout-minutes: 120
    strategy:
      fail-fast: false
      matrix:
        mode: [verbose]

    steps:
      - name: Checkout code
        uses: actions/checkout@v4

      - name: Set up Python 3.12
        uses: actions/setup-python@v5
        with:
          python-version: "3.12"

      - name: Install uv
        uses: astral-sh/setup-uv@v7
        with:
          enable-cache: true

      - name: Install dependencies
        run: uv sync --all-packages --extra=cpu --no-default-groups

      - name: Write CoreWeave kubeconfig
        run: |
          mkdir -p ~/.kube
          echo "${{ secrets.CW_KUBECONFIG }}" > ~/.kube/coreweave-iris
          chmod 600 ~/.kube/coreweave-iris

      - name: Log in to GitHub Container Registry
        uses: docker/login-action@v3
        with:
          registry: ghcr.io
          username: ${{ github.actor }}
          password: ${{ secrets.GITHUB_TOKEN }}

      - name: Inspect buildx builder
        run: docker buildx inspect | tee builder-inspect.txt

      - name: Start CoreWeave cluster
        id: start
        shell: bash
        run: |
          set -euo pipefail

          if [[ "${{ matrix.mode }}" == "verbose" ]]; then
            export BUILDKIT_PROGRESS=plain
            iris_args=(-v)
          else
            iris_args=()
          fi

          start_ts=$(date +%s)
          .venv/bin/iris "${iris_args[@]}" --config=lib/iris/examples/coreweave.yaml cluster start 2>&1 | tee cluster-start.log
          end_ts=$(date +%s)

          echo "elapsed_seconds=$((end_ts - start_ts))" >> "$GITHUB_OUTPUT"
        env:
          R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
          R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}

      - name: Summarize cluster-start logs
        if: always()
        shell: bash
        env:
          ELAPSED_SECONDS: ${{ steps.start.outputs.elapsed_seconds }}
        run: |
          set -euo pipefail

          log_file=cluster-start.log
          line_count=0
          cached_count=0
          done_count=0

          if [[ -f "$log_file" ]]; then
            line_count=$(wc -l < "$log_file" | tr -d ' ')
            cached_count=$(grep -c ' CACHED' "$log_file" || true)
            done_count=$(grep -cE '^#.* DONE( |$)' "$log_file" || true)
          fi

          {
            echo "## CW cluster start: ${{ matrix.mode }}"
            echo
            echo "- elapsed_seconds: ${ELAPSED_SECONDS:-unknown}"
            echo "- log_lines: ${line_count}"
            echo "- cached_lines: ${cached_count}"
            echo "- done_lines: ${done_count}"
            echo "- step_outcome: ${{ steps.start.outcome }}"
          } >> "$GITHUB_STEP_SUMMARY"

      - name: Upload cluster-start logs
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: cw-cluster-start-${{ matrix.mode }}-${{ github.run_id }}-${{ github.run_attempt }}
          path: |
            builder-inspect.txt
            cluster-start.log
          if-no-files-found: warn

      - name: Stop CoreWeave cluster
        if: always()
        continue-on-error: true
        run: .venv/bin/iris --config=lib/iris/examples/coreweave.yaml cluster stop
        env:
          R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
          R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}

  seed-iris-task:
    name: Seed iris-task (${{ matrix.mode }})
    runs-on: ubuntu-latest
    timeout-minutes: 45
    strategy:
      fail-fast: false
      matrix:
        mode: [noinline, inline]

    steps:
      - name: Checkout code
        uses: actions/checkout@v4

      - name: Log in to GitHub Container Registry
        uses: docker/login-action@v3
        with:
          registry: ghcr.io
          username: ${{ github.actor }}
          password: ${{ secrets.GITHUB_TOKEN }}

      - name: Seed cache image
        shell: bash
        run: |
          set -euo pipefail

          cache_ref="ghcr.io/marin-community/iris-task:cache-probe-${{ github.run_id }}-${{ github.run_attempt }}-${{ matrix.mode }}"
          extra_args=()
          if [[ "${{ matrix.mode }}" == "inline" ]]; then
            extra_args+=(--cache-to type=inline)
          fi

          docker buildx build \
            --platform linux/amd64 \
            --progress plain \
            --build-arg "IRIS_GIT_HASH=$(git rev-parse --short HEAD)" \
            --tag "$cache_ref" \
            "${extra_args[@]}" \
            --push \
            -f lib/iris/Dockerfile.task \
            .

          {
            echo "## Seed iris-task: ${{ matrix.mode }}"
            echo
            echo "- cache_ref: $cache_ref"
          } >> "$GITHUB_STEP_SUMMARY"

  measure-iris-task:
    name: Measure iris-task (${{ matrix.mode }})
    runs-on: ubuntu-latest
    timeout-minutes: 45
    needs: seed-iris-task
    strategy:
      fail-fast: false
      matrix:
        mode: [noinline, inline]

    steps:
      - name: Checkout code
        uses: actions/checkout@v4

      - name: Log in to GitHub Container Registry
        uses: docker/login-action@v3
        with:
          registry: ghcr.io
          username: ${{ github.actor }}
          password: ${{ secrets.GITHUB_TOKEN }}

      - name: Inspect buildx builder
        run: docker buildx inspect | tee builder-inspect.txt

      - name: Measure cache reuse
        id: measure
        shell: bash
        run: |
          set -euo pipefail

          cache_ref="ghcr.io/marin-community/iris-task:cache-probe-${{ github.run_id }}-${{ github.run_attempt }}-${{ matrix.mode }}"
          probe_tag="iris-task:probe-${{ matrix.mode }}"

          start_ts=$(date +%s)
          docker buildx build \
            --platform linux/amd64 \
            --progress plain \
            --build-arg "IRIS_GIT_HASH=$(git rev-parse --short HEAD)" \
            --tag "$probe_tag" \
            --cache-from "type=registry,ref=$cache_ref" \
            --output "type=docker,name=$probe_tag" \
            -f lib/iris/Dockerfile.task \
            . \
            2>&1 | tee measure.log
          end_ts=$(date +%s)

          echo "elapsed_seconds=$((end_ts - start_ts))" >> "$GITHUB_OUTPUT"

      - name: Summarize measurement logs
        if: always()
        shell: bash
        env:
          ELAPSED_SECONDS: ${{ steps.measure.outputs.elapsed_seconds }}
        run: |
          set -euo pipefail

          line_count=0
          cached_count=0
          done_count=0

          if [[ -f measure.log ]]; then
            line_count=$(wc -l < measure.log | tr -d ' ')
            cached_count=$(grep -c ' CACHED' measure.log || true)
            done_count=$(grep -cE '^#.* DONE( |$)' measure.log || true)
          fi

          {
            echo "## Measure iris-task: ${{ matrix.mode }}"
            echo
            echo "- cache_ref: ghcr.io/marin-community/iris-task:cache-probe-${{ github.run_id }}-${{ github.run_attempt }}-${{ matrix.mode }}"
            echo "- elapsed_seconds: ${ELAPSED_SECONDS:-unknown}"
            echo "- log_lines: ${line_count}"
            echo "- cached_lines: ${cached_count}"
            echo "- done_lines: ${done_count}"
            echo "- step_outcome: ${{ steps.measure.outcome }}"
          } >> "$GITHUB_STEP_SUMMARY"

      - name: Upload measurement logs
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: iris-task-measure-${{ matrix.mode }}-${{ github.run_id }}-${{ github.run_attempt }}
          path: |
            builder-inspect.txt
            measure.log
          if-no-files-found: warn