The deploy workflow pulled, built and started the app, but never applied
migrations. Shipping a change that adds a table therefore broke the app
until someone remembered to migrate by hand -- the new api_keys table
would have done exactly that to the settings page.
The workflow now builds, applies pending migrations, and only then starts
the app. A failed migration exits non-zero, so the deploy fails instead of
starting an app whose schema is behind its code. Because `up -d` is never
reached in that case, production keeps serving the previous containers.
Migrations run through a new `migrate` compose service built from the
Dockerfile's builder stage, rather than `docker compose exec app npx
prisma migrate deploy` as the README previously suggested. The production
image copies prisma/ and node_modules/.prisma but not the Prisma CLI,
which is a devDependency and is absent from the standalone bundle, so that
command would have had npx download Prisma inside the container at runtime
as the unprivileged nextjs user. The builder stage has the CLI and its
engines already.
The service sits behind a `tools` profile so `docker compose up` never
starts it, and takes DATABASE_URL from compose rather than a .env file --
dotenv does not override variables that are already set, so the compose
value wins.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
twotalesanimation