railway.toml explained, field by field
A railway.toml or railway.json declares one service's build and deploy. See what each field becomes on separate services.
Notes on deployments, databases and operating apps on Unbroker.
A railway.toml or railway.json declares one service's build and deploy. See what each field becomes on separate services.
A render.yaml Blueprint declares services, databases, and env groups. It is an inventory, not an import format.
A Heroku Procfile lists process types and commands. See what web, worker, release phase, and Scheduler become as services.
A self hosted PaaS deploys from Git on your server. You still run the OS, backups, patches, TLS, and recovery when that host is down.
Deploy an Express app with npm start, not a watcher. Listen on 0.0.0.0 and PORT, add /health, and keep secrets in variables. Node.js can qualify for Free.
Deploy a frontend and an API as separate services. The API URL is a public build variable. Fix CORS, cookies, and previews that call production.
Deploy a Django app with gunicorn or uvicorn, not runserver. Set DEBUG, ALLOWED_HOSTS, static files, and migrations. The Dockerfile path requires Pro.
Deploy FastAPI with Uvicorn in a container, not fastapi dev. Bind 0.0.0.0 and PORT, and keep secrets in runtime variables. Dockerfile builds require Pro.
Astro hosting starts with the render mode. Ship dist when the HTML is built ahead. A page that needs each request uses a server adapter.
Deploy a Next.js app with next build and next start, not next dev. NEXT_PUBLIC_ values are fixed at build. The Next.js recipe requires Pro.
Step by step: add a Worker service for a queue consumer that stays up, or a scheduled job on the Jobs tab for a cron command that exits. Both require Pro.
App monitoring and alerts use build logs, runtime logs, metrics, and deploy history. They help you notice a failure sooner. They do not guarantee it.
Store API keys in Variables, not the repo. All, Production, and Preview decide what a preview copies. NEXT_PUBLIC_ and VITE_ are public.
Managed Redis is for cache, sessions, and queues, not a Postgres replacement. Apps nodes and managed clusters both require Pro.
Free backend hosting on Unbroker is $0, no payment method. 1 project, 2 services, Node or static, 3 deploys/day. Sleeps when idle. No Docker or database.
Self host Supabase as a Docker Compose stack. You operate Postgres, Auth, Storage, secrets, backups, and the public URL.
Add a custom domain with HTTPS on Pro. Create the DNS record Cloud shows, wait for verification and the certificate, and keep the default hostname.
Files on the container disk can disappear after deploy. A Pro volume is one path, one service, one replica. An S3-compatible backup is a separate copy.
Debug a failed deployment by the stage that stopped. Read that stage's logs, then change the smallest thing. Redeploying first can hide the evidence.
Deploy on git push for the configured branch, only after Watch paths match a change in that app and GitHub checks have passed.
Database migrations on deploy run once, before the new version is live. A code rollback does not undo the schema, so the previous release has to work.
Unbroker is a PaaS, not a broker. You pick a resource and a region. The provider changes backups and where data is processed, not the deploy.
Docker Compose in production: from the compose file to running services, with health checks, volumes, and a place for data.
The maximum is a ceiling. Replica autoscaling bills the copies that are running. The minimum sets availability and the cost floor.
October 2026 has 30 changelog entries, dated through October 31. This note groups them into five themes and links every entry.
What a stable self-hosted n8n needs: Postgres instead of SQLite, a fixed encryption key, a public webhook URL, and a backup you have restored.
Pull request preview environments need their own variables, a database that is not production, and cleanup when the pull request closes.
A postgres production checklist: the connection limit, long transactions, missing indexes, vacuum, and a restore you have run.
A Lovable app is production ready when secrets, a tested backup, and rollback are in place. What publish and a Git export leave behind.
Usage based pricing on a PaaS follows consumption. Capacity based pricing follows the size you configure. Public rates as of October 28, 2026.
AI agent deploy safely with one scoped token. Claude Code deploy permissions, Codex deploy permissions, and MCP server API token scope.
How to choose a Linux cloud server: size, root, and when Cloud Apps fits. Sizes from $17.95 a month are on the pricing page.
VPS vs PaaS for a Linux cloud server: who operates the OS, how a cloud server running Linux is billed, and when to use a VPS.
No articles in this topic yet.