qinglong/.github/workflows/ql3-cloudnativepg-dr-live.yml

name: QingLong 3.0 CloudNativePG DR live evidence

on:
  workflow_dispatch:

permissions:
  contents: read

concurrency:
  group: ql3-cloudnativepg-dr-live-${{ github.ref }}
  cancel-in-progress: false

jobs:
  cloudnativepg-dr-live:
    name: Barman WAL, latest and PITR recovery
    runs-on: ubuntu-24.04
    timeout-minutes: 120
    steps:
      - uses: actions/checkout@v6
      - uses: pnpm/action-setup@v6
        with:
          version: '8.3.1'
      - uses: actions/setup-node@v6
        with:
          node-version: '24.18.0'
          cache: pnpm
          cache-dependency-path: pnpm-lock.yaml

      - name: Require enough ephemeral disk for four isolated K3s nodes
        shell: bash
        run: |
          set -euo pipefail
          available_kib="$(df -Pk / | awk 'NR == 2 { print $4 }')"
          minimum_kib="$((35 * 1024 * 1024))"
          if (( available_kib < minimum_kib )); then
            echo "CloudNativePG DR live gate requires at least 35 GiB free; found ${available_kib} KiB" >&2
            exit 1
          fi
          docker volume ls --quiet --filter dangling=true | sort > \
            "${RUNNER_TEMP}/ql3-dangling-volumes.before"
          docker system df

      - name: Install workspace dependencies without lifecycle scripts
        run: pnpm install --frozen-lockfile --ignore-scripts

      - name: Install verified kubectl v1.32.8
        shell: bash
        run: |
          set -euo pipefail
          curl --fail --location --silent --show-error \
            --output "${RUNNER_TEMP}/kubectl" \
            https://dl.k8s.io/release/v1.32.8/bin/linux/amd64/kubectl
          curl --fail --location --silent --show-error \
            --output "${RUNNER_TEMP}/kubectl.sha256" \
            https://dl.k8s.io/release/v1.32.8/bin/linux/amd64/kubectl.sha256
          test "$(cat "${RUNNER_TEMP}/kubectl.sha256")" = \
            "$(sha256sum "${RUNNER_TEMP}/kubectl" | cut -d ' ' -f 1)"
          chmod 0755 "${RUNNER_TEMP}/kubectl"

      - name: Fetch checksum-locked operator manifests
        shell: bash
        run: |
          set -euo pipefail
          curl --fail --location --silent --show-error \
            --output "${RUNNER_TEMP}/cert-manager.yaml" \
            https://github.com/cert-manager/cert-manager/releases/download/v1.20.3/cert-manager.yaml
          echo "7ee74ba06845213e96d8ceaff3d20dd51e682765c1418eddda4e8780ba082261  ${RUNNER_TEMP}/cert-manager.yaml" | sha256sum --check
          curl --fail --location --silent --show-error \
            --output "${RUNNER_TEMP}/barman-cloud.yaml" \
            https://github.com/cloudnative-pg/plugin-barman-cloud/releases/download/v0.13.0/manifest.yaml
          echo "d2e71e7b06822448f1a421f05781846cfdb9cc621e7ef32eef5e20c5133213b0  ${RUNNER_TEMP}/barman-cloud.yaml" | sha256sum --check
          curl --fail --location --silent --show-error \
            --output "${RUNNER_TEMP}/cloudnative-pg.yaml" \
            https://github.com/cloudnative-pg/cloudnative-pg/releases/download/v1.30.0/cnpg-1.30.0.yaml
          echo "f8bede43fe4ee0d478c2355b204a36876b2ae4faac60f2a9452280b293da3b88  ${RUNNER_TEMP}/cloudnative-pg.yaml" | sha256sum --check

      - name: Recheck the static backup and supply-chain contracts
        run: |
          pnpm audit:cloudnativepg-backup:ql3
          pnpm audit:barman-cloud-supply-chain:ql3
          pnpm audit:cert-manager-selection:ql3
          node --test \
            test/back/ql3CloudNativePgBarmanLiveContract.test.cjs \
            test/back/ql3CloudNativePgDrEvidenceAudit.test.cjs \
            test/back/ql3CloudNativePgBarmanWorkflow.test.cjs

      - name: Prove TLS object storage, continuous WAL, latest restore and PITR
        env:
          QL3_CLOUDNATIVEPG_BARMAN_LIVE: '1'
          QL3_SOURCE_REVISION: ${{ github.sha }}
          QL3_KUBECTL_BIN: ${{ runner.temp }}/kubectl
          QL3_CERT_MANAGER_MANIFEST_FILE: ${{ runner.temp }}/cert-manager.yaml
          QL3_BARMAN_MANIFEST_FILE: ${{ runner.temp }}/barman-cloud.yaml
          QL3_CNPG_OPERATOR_MANIFEST_FILE: ${{ runner.temp }}/cloudnative-pg.yaml
          QL3_DR_REPORT: ${{ runner.temp }}/ql3-cloudnativepg-dr/report.json
        shell: bash
        run: |
          set -euo pipefail
          umask 077
          mkdir -m 0700 "$(dirname "${QL3_DR_REPORT}")"
          pnpm test:cloudnativepg-barman-live:ql3 -- \
            "--report=${QL3_DR_REPORT}"

      - name: Re-audit the exact private report and isolated cleanup
        env:
          QL3_DR_REPORT: ${{ runner.temp }}/ql3-cloudnativepg-dr/report.json
        shell: bash
        run: |
          set -euo pipefail
          test "$(stat -c '%a' "${QL3_DR_REPORT}")" = '600'
          pnpm audit:cloudnativepg-dr-evidence:ql3 -- \
            "--report=${QL3_DR_REPORT}"
          sha256sum "${QL3_DR_REPORT}"
          test -z "$(docker ps -aq --filter name=ql3-barman-dr-)"
          test -z "$(docker network ls -q --filter name=ql3-barman-dr-)"
          test -z "$(docker ps -aq --filter label=io.qinglong.ql3.live=cloudnativepg-barman-disaster-recovery)"
          test -z "$(docker network ls -q --filter label=io.qinglong.ql3.live=cloudnativepg-barman-disaster-recovery)"
          docker volume ls --quiet --filter dangling=true | sort > \
            "${RUNNER_TEMP}/ql3-dangling-volumes.after"
          diff --unified \
            "${RUNNER_TEMP}/ql3-dangling-volumes.before" \
            "${RUNNER_TEMP}/ql3-dangling-volumes.after"
          docker system df

      - name: Upload the audited content-free DR evidence
        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        with:
          name: ql3-cloudnativepg-dr-${{ github.run_id }}-${{ github.run_attempt }}
          path: ${{ runner.temp }}/ql3-cloudnativepg-dr/report.json
          if-no-files-found: error
          retention-days: 14
          compression-level: 9
          overwrite: false
          include-hidden-files: false