zkldi_Tachi/.github/workflows/server.yml

name: Server CI/CD

on:
  push:
    branches:
      - "main"
      - "main-dev"
    paths:
      - "typescript/server/**"
      - "typescript/common/**"
      - "typescript/db/**"
      - "typescript/db-migration-engine/**"
      - "docker/**"
      - "db/migrations/**"
      - ".github/workflows/server.yml"
  pull_request:
    branches:
      - "main"
      - "main-dev"
    paths:
      - "typescript/server/**"
      - "typescript/common/**"
      - "typescript/db/**"
      - "typescript/db-migration-engine/**"
      - "docker/**"
      - "db/migrations/**"
      - ".github/workflows/server.yml"
  workflow_dispatch:

permissions:
  contents: read

concurrency:
  group: "server-${{ github.ref }}-${{ github.workflow }}"
  cancel-in-progress: false

jobs:
  test:
    runs-on: ubuntu-latest
    container:
      image: ghcr.io/zkldi/tachi-ci:main
      options: --user root
    env:
      NODE_ENV: "test"
    services:
      # Test Postgres mirrors `tachi-postgres-test` in docker-compose-dev.yml:
      # a tmpfs-backed PG sized for ~8 long-lived worker DBs with autovacuum
      # on. `PGDATA` must be a SUBDIRECTORY of the tmpfs mount, otherwise
      # initdb refuses to clobber the mount root. The runtime-tunable knobs
      # (synchronous_commit, WAL ceiling, autovacuum naptime, etc.) are
      # applied via `ALTER SYSTEM` in a setup step below; restart-required
      # knobs (fsync, full_page_writes, shared_buffers, ...) we leave at
      # the postgres defaults because `services:` does not let us pass
      # postgres CLI args. With tmpfs, fsync is essentially free anyway.
      tachi-postgres-test:
        image: ghcr.io/zkldi/zk-postgres:18@sha256:484fa9e6501ec93c5729ba9f9312d8ee4dfe0c136f1b4881fab5e0aed5aa68fd
        env:
          POSTGRES_USER: tachi
          POSTGRES_PASSWORD: tachi
          POSTGRES_DB: postgres
          PGDATA: /var/lib/postgresql/data/pgdata
        options: >-
          --health-cmd "pg_isready -U tachi"
          --health-interval 5s
          --health-timeout 5s
          --health-retries 10
          --tmpfs /var/lib/postgresql/data:rw,size=3g
      tachi-redis:
        image: redis:7.4-alpine
        options: >-
          --health-cmd "redis-cli ping"
          --health-interval 5s
          --health-timeout 5s
          --health-retries 10
      tachi-s3:
        image: quay.io/minio/minio:RELEASE.2024-10-29T16-01-48Z
        env:
          MINIO_ROOT_USER: minio
          MINIO_ROOT_PASSWORD: password
        entrypoint: /usr/bin/minio
        command: server /data --console-address=:9001
        options: >-
          --health-cmd "curl -f http://localhost:9000/minio/health/live"
          --health-interval 5s
          --health-timeout 5s
          --health-retries 10
    steps:
      - name: Checkout
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          persist-credentials: false

      - name: Link workspace to /tachi
        # tachidb, just recipes, and .env.test all assume the repo lives at /tachi.
        # GHA mounts the workspace elsewhere, so we replace the empty /tachi dir
        # (created by the image's WORKDIR) with a symlink to the real checkout.
        run: rm -rf /tachi && ln -sf "$GITHUB_WORKSPACE" /tachi

      - name: Install dependencies
        run: bun install --frozen-lockfile

      - name: Tune test Postgres for short-lived worker DBs
        # The restart-only knobs (fsync, shared_buffers, ...) cannot be set
        # because the GHA service syntax does not let us pass postgres CLI
        # args. Everything below is reloadable; it lines up with the
        # `-c ...` flags on `tachi-postgres-test` in docker-compose-dev.yml
        # so CI and local behave the same once the suite starts running.
        env:
          PGPASSWORD: tachi
        run: |
          psql -h tachi-postgres-test -U tachi -d postgres <<'SQL'
            ALTER SYSTEM SET synchronous_commit = off;
            ALTER SYSTEM SET max_wal_size = '128MB';
            ALTER SYSTEM SET min_wal_size = '32MB';
            ALTER SYSTEM SET checkpoint_timeout = '30s';
            ALTER SYSTEM SET autovacuum_naptime = '5s';
            ALTER SYSTEM SET autovacuum_vacuum_scale_factor = 0.05;
            ALTER SYSTEM SET autovacuum_analyze_scale_factor = 0.1;
            SELECT pg_reload_conf();
          SQL

      - name: Run tests
        env:
          NODE_ENV: "test"
          PORT: 8080
          POSTGRES_TEST_HOST: tachi-postgres-test
          POSTGRES_TEST_URL: postgresql://tachi:tachi@tachi-postgres-test
          # ubuntu-latest is 4 vCPU; cap workers to leave headroom for the
          # PG service container and avoid oversubscribing tmpfs.
          VITEST_MAX_WORKERS: "4"
        run: bun run --filter tachi-server test

  docker-push:
    runs-on: ubuntu-latest
    needs: [test]
    if: ${{ github.event_name == 'push' || github.event_name == 'workflow_dispatch' }}
    permissions:
      contents: read
      packages: write
    outputs:
      image_tag: ${{ steps.revision.outputs.image_tag }}
    steps:
      - name: Checkout
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          persist-credentials: false

      - name: Revision tag (YYYYMMDD-shortsha + full SHA aliases)
        id: revision
        run: |
          set -eu
          short=$(git rev-parse --short=7 HEAD)
          d=$(TZ=UTC0 git log -1 --format=%cd --date=format:%Y%m%d HEAD)
          echo "image_tag=${d}-${short}" >> "$GITHUB_OUTPUT"

      - name: Log in to GHCR
        uses: docker/login-action@4907a6ddec9925e35a0a9e82d7399ccc52663121 # v4.1.0
        with:
          registry: ghcr.io
          username: ${{ github.actor }}
          password: ${{ secrets.GITHUB_TOKEN }}

      - name: Set up Docker Buildx
        uses: docker/setup-buildx-action@4d04d5d9486b7bd6fa91e7baf45bbb4f8b9deedd # v4.0.0

      - name: Read server package version
        id: server_version
        run: |
          echo "version=$(jq -r .version typescript/server/package.json)" >> "$GITHUB_OUTPUT"

      - name: Set build time
        id: build_time
        run: echo "value=$(date -u +%Y-%m-%dT%H:%M:%SZ)" >> "$GITHUB_OUTPUT"

      - name: Compute floating image tags
        id: floating_tags
        run: |
          if [ "$GITHUB_REF" = "refs/heads/main" ]; then
            echo "value=ghcr.io/zkldi/tachi:main\nghcr.io/zkldi/tachi:latest" >> "$GITHUB_OUTPUT"
          else
            echo "value=ghcr.io/zkldi/tachi:main-dev" >> "$GITHUB_OUTPUT"
          fi

      - name: Build and push unified server image
        uses: docker/build-push-action@bcafcacb16a39f128d818304e6c9c0c18556b85f # v7.1.0
        with:
          context: .
          file: docker/Dockerfile.server
          target: prod
          push: ${{ github.ref == 'refs/heads/main' || github.ref == 'refs/heads/main-dev' }}
          provenance: true
          sbom: true
          tags: |
            ghcr.io/zkldi/tachi:${{ steps.revision.outputs.image_tag }}
            ghcr.io/zkldi/tachi:${{ github.sha }}
            ${{ steps.floating_tags.outputs.value }}
          build-args: |
            VERSION=${{ steps.server_version.outputs.version }}
            VERSION_DETAIL=${{ steps.revision.outputs.image_tag }}
            GIT_REVISION=${{ github.sha }}
            BUILD_TIME=${{ steps.build_time.outputs.value }}
          cache-from: type=gha,scope=server-${{ github.ref_name }}
          cache-to: type=gha,mode=max,scope=server-${{ github.ref_name }}

  dispatch-deploy-prod:
    # Deploys boku and kamai from main.
    # See server.yml for rationale — deploy logs live in the private server-ceres repo.
    runs-on: ubuntu-latest
    needs: [test, docker-push]
    if: ${{ github.ref == 'refs/heads/main' && (github.event_name == 'push' || github.event_name == 'workflow_dispatch') }}
    strategy:
      fail-fast: false
      matrix:
        destination: [boku, kamai]
    permissions:
      contents: read
    steps:
      - name: Dispatch deploy
        env:
          # Fine-grained PAT, repo-scoped to zkldi/server-ceres, Contents: write.
          GH_TOKEN: ${{ secrets.DEPLOY_DISPATCH_TOKEN }}
          DEST: ${{ matrix.destination }}
          SHA: ${{ github.sha }}
          IMAGE_TAG: ${{ needs.docker-push.outputs.image_tag }}
        run: |
          gh api repos/zkldi/server-ceres/dispatches \
            -f event_type=deploy \
            -F "client_payload[app]=server" \
            -F "client_payload[dest]=${DEST}" \
            -F "client_payload[sha]=${SHA}" \
            -F "client_payload[image_tag]=${IMAGE_TAG}"
          echo "::notice::Dispatched server@${DEST} ${IMAGE_TAG} (${SHA}) — logs: https://github.com/zkldi/server-ceres/actions/workflows/tachi-deploy.yml"

  dispatch-deploy-dev:
    # Deploys dev from main-dev.
    # See server.yml for rationale — deploy logs live in the private server-ceres repo.
    runs-on: ubuntu-latest
    needs: [test, docker-push]
    if: ${{ github.ref == 'refs/heads/main-dev' && (github.event_name == 'push' || github.event_name == 'workflow_dispatch') }}
    permissions:
      contents: read
    steps:
      - name: Dispatch deploy
        env:
          # Fine-grained PAT, repo-scoped to zkldi/server-ceres, Contents: write.
          GH_TOKEN: ${{ secrets.DEPLOY_DISPATCH_TOKEN }}
          SHA: ${{ github.sha }}
          IMAGE_TAG: ${{ needs.docker-push.outputs.image_tag }}
        run: |
          gh api repos/zkldi/server-ceres/dispatches \
            -f event_type=deploy \
            -F "client_payload[app]=server" \
            -F "client_payload[dest]=dev" \
            -F "client_payload[sha]=${SHA}" \
            -F "client_payload[image_tag]=${IMAGE_TAG}"
          echo "::notice::Dispatched server@dev ${IMAGE_TAG} (${SHA}) — logs: https://github.com/zkldi/server-ceres/actions/workflows/tachi-deploy.yml"