Skip to content

nest-monorepo deployment

Every image is built from the monorepo root, because each needs packages/db as well as its app:

Terminal window
docker build -f apps/api/Dockerfile -t shop-api .
docker build -f apps/api/Dockerfile --target migrate -t shop-api-migrate .
docker build -f apps/worker/Dockerfile -t shop-worker .

Each Dockerfile runs turbo prune <app> --docker first: the install layer is built from that app’s manifests and a pruned lockfile, so it is reused until a dependency the app actually uses changes. The build stage runs turbo run build --filter=<app>, which builds @scope/db first, and the runtime stage keeps production dependencies only.

Image Runs Port
shop-api node dist/src/main in apps/api 5000
shop-api-migrate prisma migrate deploy against $DATABASE_URL, then exits none
shop-worker pm2-runtime (or node, without worker-pm2) in apps/worker none
  1. Run shop-api-migrate:<tag> once, with the production DATABASE_URL.
  2. Start shop-api:<tag> and shop-worker:<tag>.

Deploy every app on the same tag. They share one Prisma client, so an API and a worker from different commits can disagree about the schema.

Everything else, the API’s variables and health checks, the worker’s process model and Chromium, is as in nest-api and node-worker.

.forgejo/workflows/ at the root:

Workflow Runs on Does
ci.yml every branch except main and development turbo run lint typecheck test build
test-e2e.yml pull requests into development the API’s e2e suite against service containers
build-and-push.yml main, manual runs builds every image and pushes them under one shared tag