installcheck.yml

#
# Installcheck Tests Workflow
#
# This workflow runs installcheck tests on node n1 of a three-node Spock cluster.
# Environment variables are loaded from tests/docker/pgedge.env
#
# NOTES:
# The 'security_label' test has no chance to pass without extra coding. So, just
# skip it for now. This decision may be revised later if something is changed
# in the PostgreSQL core.
# The sad fact is that n2 and n3 can't pass these tests without errors, because
# the LR is limited and doesn't support many features. That turns their logs
# into a mess, but we still save them to detect any potential issues.
#

name: Installcheck Tests (Three-Node Cluster)
run-name: Running installcheck tests on three-node Spock cluster

on:
  workflow_dispatch:
  pull_request:
    types: [opened, synchronize, reopened]

permissions:
  contents: read

jobs:
  installcheck:
    strategy:
      fail-fast: false
      matrix:
        os: [ubuntu-latest]
        pgver: [15, 16, 17, 18]

    runs-on: ${{ matrix.os }}

    steps:
      - name: Checkout spock
        uses: actions/checkout@v4
        with:
          ref: ${{ github.ref }}

      - name: Add permissions
        run: |
          sudo chmod -R a+w ${GITHUB_WORKSPACE}

      - name: Set up Docker
        uses: docker/setup-buildx-action@e468171a9de216ec08956ac3ada2f0791b6bd435 # v3

      - name: Set up docker compose
        uses: docker/setup-compose-action@364cc21a5de5b1ee4a7f5f9d3fa374ce0ccde746 # v1
        with:
          version: latest

      - name: Start docker cluster
        id: start_cluster
        run: |
          cd ${GITHUB_WORKSPACE}/tests/docker/
          # To minimize regression tests difference, override pgedge.env with
          # standard PostgreSQL regression test defaults
          sed -i 's/^DBUSER=.*/DBUSER=regression/' pgedge.env
          sed -i 's/^DBNAME=.*/DBNAME=regression/' pgedge.env
          # Build and start the cluster
          PGVER=${{ matrix.pgver }} DBUSER=regression DBNAME=regression \
            docker compose up --build --wait -d
        timeout-minutes: 20
        continue-on-error: true

      - name: Diagnose cluster startup failure
        if: steps.start_cluster.outcome == 'failure'
        run: |
          cd ${GITHUB_WORKSPACE}/tests/docker/
          echo "=== Docker container status ==="
          docker compose ps -a
          for node in n1 n2 n3; do
            echo ""
            echo "=== Container logs: $node ==="
            docker compose logs pgedge-$node 2>&1 | tail -80 || true
            echo ""
            echo "=== PostgreSQL logfile: $node ==="
            docker compose cp pgedge-$node:/home/pgedge/logfile.log /dev/stdout 2>/dev/null | tail -80 || echo "(not available)"
          done

      - name: Run installcheck on node n1
        id: installcheck
        run: |
          cd ${GITHUB_WORKSPACE}/tests/docker/

          # Run installcheck on n1 and capture output
          docker compose exec -T pgedge-n1 bash -c '~/run-installcheck.sh' \
            | tee installcheck_output.txt || true

          # Analyze the output for test failures
          echo "Analyzing test results..."

          # Check for 'exited with exit code' which indicates a crash or abnormal termination
          if grep -q "exited with exit code" installcheck_output.txt; then
            echo "ERROR: Found 'exited with exit code' in output - indicates abnormal termination"
            echo "test_result=failed" >> $GITHUB_OUTPUT
            exit 0
          fi

          # Check for 'not ok' test results, excluding 'security_label'
          # Extract all 'not ok' lines and filter out security_label
          FAILED_TESTS=$(grep "not ok" installcheck_output.txt | grep -v "security_label" || true)

          if [ -n "$FAILED_TESTS" ]; then
            echo "ERROR: Found failing tests (excluding security_label):"
            echo "$FAILED_TESTS"
            echo "test_result=failed" >> $GITHUB_OUTPUT
          else
            echo "All tests passed (security_label failures are expected and ignored)"
            echo "test_result=passed" >> $GITHUB_OUTPUT
          fi
        timeout-minutes: 30

      - name: Collect installcheck artifacts
        if: ${{ always() }}
        run: |
          cd ${GITHUB_WORKSPACE}/tests/docker/
          mkdir -p installcheck-logs

          # Copy the output file we captured
          cp installcheck_output.txt installcheck-logs/

          # Copy regression output from n1 (where tests run)
          # pg_regress outputs to /home/pgedge/installcheck-output/
          docker compose cp pgedge-n1:/home/pgedge/installcheck-output/regression.diffs installcheck-logs/ || true
          docker compose cp pgedge-n1:/home/pgedge/installcheck-output/regression.out installcheck-logs/ || true
          docker compose cp pgedge-n1:/home/pgedge/installcheck-output/results installcheck-logs/results/ || true

          # Collect PostgreSQL logs from all nodes
          for node in n1 n2 n3; do
            echo "Collecting logs from $node..."
            # PostgreSQL log file (from pg_ctl -l in entrypoint.sh)
            docker compose cp pgedge-$node:/home/pgedge/logfile.log installcheck-logs/$node-logfile.log || true
            # Container logs
            docker compose logs pgedge-$node > installcheck-logs/$node-container.log 2>&1 || true
          done

      - name: Upload installcheck artifacts
        if: ${{ always() }}
        uses: actions/upload-artifact@v4
        with:
          name: installcheck-logs-pg${{ matrix.pgver }}
          path: tests/docker/installcheck-logs/
          if-no-files-found: error
          retention-days: 7

      - name: Cleanup docker cluster
        if: ${{ always() }}
        run: |
          cd ${GITHUB_WORKSPACE}/tests/docker/
          docker compose down

      - name: Check test result
        run: |
          if [ "${{ steps.installcheck.outputs.test_result }}" = "failed" ]; then
            echo "Installcheck tests failed (see artifacts for details)"
            exit 1
          fi
          echo "Installcheck tests passed successfully"