Running on Railway?

Railway services crashing or burning credit? We ship it properly.

Railway is a lovely place to prototype and a surprising place to find yourself in production. We pick it up when a Nixpacks build keeps failing at a step nobody added, when a Postgres volume quietly hits the plan ceiling, or when the monthly bill climbs because every PR is spawning a preview environment that forgets to spin down. We tighten the services, cap the spending, and plan whether staying on Railway is still the right call.

  • Since 2015Fixing technology
  • FreeEstimates
  • US teamReal engineers

railway-deploy.log

still failing

What it is

Railway at a glance

Railway is a platform-as-a-service launched in 2020 that aims to feel like Heroku did at its best: push a repo, get a URL, and stop thinking about containers until you have to. It builds with Nixpacks by default, hosts services on a shared infrastructure in US-west, US-east and EU-west, and has managed Postgres, MySQL, MongoDB and Redis as one-click add-ons.

The pricing model is credit-based usage, which is friendlier than Heroku for small services and noisier than a fixed plan once you reach a few environments. Railway has pivoted several times, most recently off Kubernetes onto its own control plane, and the tooling has moved with it. The GitHub integration, PR previews and template marketplace are the features teams reach for first.

Type
Platform-as-a-service for web apps and databases
First released
2020
License
Proprietary SaaS
Maintained by
Railway Corp
Typically used for
Side projects going to production, prototypes, small SaaS backends, preview environments on PRs

The honest version

Railway pros and cons

No technology is all upside. This is where Railway earns its place and where it tends to cause trouble.

Where Railway is strong

  • The developer experience is excellentA repo connected to Railway is building and serving a URL in minutes, and the CLI mirrors the dashboard rather than hiding behind it. Teams productive on day one is the norm.
  • Nixpacks reads most modern projectsNode, Python, Go, Ruby and Rust projects boot without a Dockerfile written by hand, which makes small services genuinely small to deploy.
  • Add-on databases are first-classPostgres, MySQL, MongoDB and Redis are spun up next to the service, networked privately, and connection strings land in environment variables without manual wiring.
  • Preview environments per pull requestA PR gets its own ephemeral copy of the whole stack, which is how modern teams expect to review, and which used to need a lot of custom work elsewhere.
  • Pricing is linear at small sizesA quiet service costs very little because you pay for the resources actually consumed, not a fixed slot that sits idle most of the day.

Where Railway bites

  • Costs are predictable only if somebody watches themCredit-based billing scales with usage, and a forgotten preview environment, a chatty cron, or a service that leaks memory can produce a bill that surprises the finance team.
  • Limited regions and no multi-region failoverYou choose a region, your service runs there, and if that region has a bad day your service has a bad day. Multi-region failover is not part of the product.
  • The platform has pivoted its runtime more than onceTeams who built on Railway in 2022 have learned to expect occasional migrations as the underlying infrastructure changes. Each one has been handled, but each one is work.
  • Observability is basic out of the boxLogs and simple metrics are there, but anything beyond that means wiring up Sentry, Grafana Cloud, Axiom or similar. A production team cannot rely on the built-in surface alone.
  • Private networking is simpler than you may wantServices in a project can talk to each other, but there is no VPC peering to AWS or GCP, no dedicated egress IP without workarounds, and compliance customers will keep asking for both.
  • Database backups need attentionAutomatic backups exist, but the retention and the restore path need to be tested rather than assumed. More than one team has discovered this on the day it mattered.

Sound familiar?

Common Railway problems we fix

The Railway problems that bring people to us, and what you have once each one is fixed.

Nixpacks build failed

Builds fail at a step nobody wrote

A detection step picks the wrong runtime, a lockfile does not match the package manager Nixpacks assumes, or a native dependency needs a system package that is not in the base image.

Builds reliable, cache warm

deploy crashed: exit code 137

Services keep dying out of memory

A Node app with a memory leak, a worker that fans out too wide, or a Postgres that was given a plan size the dataset outgrew months ago.

Memory sized to real load

no space left on device

Postgres volume is full

A table with no retention, an audit log nobody pruned, or WAL files not shipping. The service survives until the moment it runs out of room, then goes read-only.

Storage reclaimed, retention set

estimated monthly: $412 (was $138)

The bill has jumped and nobody knows why

Preview environments from closed PRs still running, a cron that spawned duplicates during a retry, or a service left in a crashloop that bills for restarts.

Spend back in line with the plan

healthcheck /healthz: 502

Deploys never go healthy

A healthcheck path that the application does not actually serve, a bind to 127.0.0.1 instead of 0.0.0.0, or a startup time longer than the healthcheck deadline.

Healthchecks honest, deploys stable

connection pool exhausted

Postgres runs out of connections

A framework defaulting to large pools per worker, a lambda-style surge of short-lived connections, or PgBouncer absent on a service that needed it from the start.

Pooling in place, limits documented

cron timeout waiting for lock

Scheduled jobs are running twice

A cron that slowly climbed past its interval, so a new run started before the old one finished, and the lack of a lock meant both ran against the same rows.

Jobs serialized and observable

custom domain: certificate pending

Custom domain will not finish provisioning

A CNAME still pointing at Heroku, a CAA record from a previous host, or a DNS provider adding a stray proxy that breaks ACME validation.

Domain live on TLS

What we do

Railway help, from stabilization to steady state

Call us for one unhealthy deploy, or hand us the whole workspace.

Rescue
Services in crashloop, databases out of room, or a bill climbing by the hour. We stabilize the workspace and then unpick what caused it.
Cost control
A sweep through the project for forgotten preview environments, oversized add-ons, and services that never spin down, with the before and after numbers on paper.
Deploy reliability
Nixpacks or Dockerfile builds made deterministic, healthchecks honest, and rollbacks that work before you ever need them.
Database upkeep
Postgres sizing, backups verified by actual restore drills, and PgBouncer put in front of workloads that exhausted the connection pool.
Migrations on and off Railway
Bringing a service onto Railway from Heroku or Vercel, or taking a service that outgrew the model over to Fly, Render or a VM, without dropping traffic.
Managed Tech
Daily log watch, monthly spend review, patched builds and a human at the other end of the chat when a region has a wobble.

How it works

From first message to fixed

  1. Look Free estimate

    Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.

  2. Fix

    We fix the cause, not the symptom. If something is down, getting it back up comes first.

  3. Harden

    Backups, monitoring, updates and access, so the same problem can't quietly come back.

  4. Hand back or stay on

    You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.

Often found together

Running Railway alongside something else?

We work on the whole stack, so the problem can be wherever it actually is.

Questions

Railway questions

Something else? Call+1 737 727 7327

Why is our Railway deploy stuck on build?

In practice it is usually Nixpacks picking the wrong provider for a polyglot repo, a native dependency missing a system package in the base image, or a lockfile that disagrees with the package manager Nixpacks detected. The fix is either to pin the build in railway.json, swap to a hand-written Dockerfile for that service, or correct the lockfile so detection does the right thing.

Why has our Railway bill jumped this month?

The three common causes are preview environments that stayed running after their PRs closed, add-on databases that were upsized during an incident and never downsized, and services stuck in crashloop that keep consuming build and runtime minutes. The dashboard shows the breakdown; the trick is acting on it before the next cycle.

Should we stay on Railway or move to Fly, Render or AWS?

If the workload is a small number of services that fit comfortably in one region, Railway is often the right call. Once you need multi-region failover, VPC peering, or heavy data residency commitments, the gap widens and Fly or AWS begins to pay for itself. We tell you honestly which side of that line you are on.

Can you take over a Railway workspace from a former developer?

Yes. We work with you to transfer ownership, rotate the project tokens, reconnect the GitHub integration under your account, and audit which services and add-ons are actually in use so nothing goes dark during the handover.

Why do our Railway Postgres connections keep running out?

The service is almost certainly doing per-request connections or running many workers that each open their own pool, and Railway's Postgres has a finite connection limit on each plan. The honest fix is PgBouncer in front of the database and a review of pool sizing in the framework, not a bigger plan size.

How much does Railway help cost?

We give estimates at no charge. Send over a description of the workspace and what is going wrong, and we return a quote before anyone touches the account.

Free estimate

Railway giving you trouble?

Tell us what you're running and what's going wrong. A person reads it and comes back with an estimate.

  • No charge to find out what it will take
  • If it's urgent, say so

Replies come from a person. By sending, you authorize us to email you back. No marketing.