Railway

On Railway, the server is one service built from your Dockerfile, and the store is an R2 bucket. The service is described in .railway/railway.ts, Railway's infrastructure as code file.

Two things are different from a typical Railway app:

• The service has a volume it never uses. Railway normally starts the new deployment and stops the old one once the new one is healthy, which is the opposite of what Akter needs. It never runs two deployments of a service that has a volume, so attaching one makes every deploy stop the old server first.
• The bucket is R2, not a Railway bucket. Railway's buckets come from Tigris, which no longer accepts the path-style requests Akter sends for new buckets.

You need the src/server.ts, Dockerfile and .dockerignore from Deploying, an R2 bucket, and the railway command.

Describe the service⁠

bun add -d railway

Create .railway/railway.ts:

import { defineRailway, preserve, project, service, volume } from "railway/iac"

export default defineRailway(() => {
  const handoff = volume("handoff")

  const server = service("server", {
    healthcheck: "/ready",
    volumeMounts: { "/handoff": handoff },
    deploy: { drainingSeconds: 15 },
    env: {
      AKTER_STORAGE_BUCKET: "my-app-store",
      AKTER_STORAGE_ENDPOINT: "https://ACCOUNT_ID.r2.cloudflarestorage.com",
      AKTER_STORAGE_REGION: "auto",
      AWS_ACCESS_KEY_ID: preserve(),
      AWS_SECRET_ACCESS_KEY: preserve(),
      AUTH_ISSUER: "https://login.example.com/",
      AUTH_AUDIENCE: "my-app",
      AUTH_JWKS_URL: "https://login.example.com/.well-known/jwks.json",
    },
  })

  return project("my-app", { resources: [server, handoff] })
})

• volumeMounts attaches the volume, for the reason above. A volume also keeps the service to one replica.
• healthcheck makes Railway wait for /ready before it sends traffic to a new deployment.
• drainingSeconds gives the server 15 seconds between SIGTERM and SIGKILL. Railway's default is 0.
• preserve() leaves the bucket's keys out of the file. You set them with the CLI below, and later applies keep them.

Railway sets PORT, and the server listens on it.

Deploy⁠

railway login
railway init --name my-app
railway config apply
railway variable set --service server AWS_ACCESS_KEY_ID=<access key id> AWS_SECRET_ACCESS_KEY=<secret access key>
railway up --service server
railway domain --service server

config apply shows the plan and creates the service and volume. railway up uploads the project, and Railway builds the Dockerfile it finds. railway domain gives the service a public URL. Check the server:

curl https://<your domain>/ready

Releases⁠

Run railway up --service server again, or connect the service to your GitHub repository so each push deploys. Railway stops the old deployment, waits up to drainingSeconds for it to drain, and starts the new one. Clients retry through the gap.

Railway sometimes moves a service to another host on its own. That's a restart of the same release, so it needs nothing from you.