comet-ml--opik/.github/workflows/cleanup_e2e_docker.yaml

name: Cleanup E2E Docker Images

on:
  pull_request:
    types: [closed]
  schedule:
    - cron: "0 3 * * 6"  # Every Saturday at 03:00 UTC
  workflow_dispatch:
    inputs:
      max_age_days:
        description: "Delete CI tags older than this many days (0 = delete all)"
        required: false
        default: 7
        type: number

permissions:
  contents: read
  packages: write

env:
  IMAGES: >-
    opik-backend
    opik-python-backend
    opik-guardrails-backend

jobs:
  cleanup:
    name: Cleanup CI images
    runs-on: ubuntu-latest
    timeout-minutes: 15
    steps:
      - name: Delete CI images
        env:
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          EVENT_NAME: ${{ github.event_name }}
          PR_NUMBER: ${{ github.event.pull_request.number }}
          MAX_AGE_DAYS: ${{ github.event.inputs.max_age_days || '7' }}
        run: |
          DELETED=0
          SKIPPED=0
          FAILED=0

          # Build the jq filter and age cutoff based on trigger type
          if [ "$EVENT_NAME" = "pull_request" ]; then
            echo "Mode: PR close — deleting all ci-e2e-pr${PR_NUMBER}-* images"
            JQ_FILTER=".[] | select(.metadata.container.tags | any(startswith(\"ci-e2e-pr${PR_NUMBER}-\"))) | {id: .id, tags: .metadata.container.tags, created_at: .created_at}"
            CUTOFF=""
          else
            echo "Mode: stale sweep — deleting ci-e2e-* images older than ${MAX_AGE_DAYS} days"
            JQ_FILTER='.[] | select(.metadata.container.tags | any(startswith("ci-e2e-"))) | {id: .id, tags: .metadata.container.tags, created_at: .created_at}'
            CUTOFF=$(date -u -d "${MAX_AGE_DAYS} days ago" +%s 2>/dev/null \
              || date -u -v-"${MAX_AGE_DAYS}"d +%s)
          fi

          for IMAGE in ${{ env.IMAGES }}; do
            echo "=== Scanning ${IMAGE} ==="

            if ! VERSIONS=$(gh api \
              "/orgs/comet-ml/packages/container/opik%2F${IMAGE}/versions" \
              --paginate \
              --jq "$JQ_FILTER" 2>&1); then
              echo "  Failed to list versions: ${VERSIONS}"
              continue
            fi

            if [ -z "$VERSIONS" ]; then
              echo "  No matching versions found"
              continue
            fi

            COUNT=$(echo "$VERSIONS" | wc -l | tr -d ' ')
            echo "  Found ${COUNT} matching versions"

            while read -r VERSION; do
              [ -z "$VERSION" ] && continue
              VERSION_ID=$(echo "$VERSION" | jq -r '.id')
              CI_TAG=$(echo "$VERSION" | jq -r '.tags[] | select(startswith("ci-e2e-"))' | head -1)

              # For stale sweep, check age; for PR close, delete unconditionally
              if [ -n "$CUTOFF" ]; then
                CREATED_AT=$(echo "$VERSION" | jq -r '.created_at')
                CREATED_TS=$(date -u -d "$CREATED_AT" +%s 2>/dev/null \
                  || date -u -j -f "%Y-%m-%dT%H:%M:%S" "${CREATED_AT%%.*}" +%s 2>/dev/null \
                  || echo "")
                if [ -z "$CREATED_TS" ]; then
                  echo "  Skipping ${IMAGE}:${CI_TAG} (could not parse created_at: ${CREATED_AT})"
                  SKIPPED=$((SKIPPED + 1))
                  continue
                fi
                if [ "$CREATED_TS" -ge "$CUTOFF" ]; then
                  SKIPPED=$((SKIPPED + 1))
                  continue
                fi
                echo "  Deleting ${IMAGE}:${CI_TAG} (older than ${MAX_AGE_DAYS} days)"
              else
                echo "  Deleting ${IMAGE}:${CI_TAG}"
              fi

              if gh api --method DELETE \
                "/orgs/comet-ml/packages/container/opik%2F${IMAGE}/versions/${VERSION_ID}"; then
                DELETED=$((DELETED + 1))
              else
                FAILED=$((FAILED + 1))
              fi
            done <<< "$VERSIONS"
          done

          echo ""
          echo "=== Summary ==="
          echo "Deleted: ${DELETED}"
          echo "Skipped: ${SKIPPED}"
          echo "Failed: ${FAILED}"