It would be great to have a rolling release feature for services. Vercel's implementation is really good: https://vercel.com/docs/rolling-releases

The core flow:

• Deploy a new version alongside the existing one (prod and canary)

• Gradually shift traffic to the new version (percentage-based or stepwise)

• Increase rollout and then finally promote canary to prod - based on metrics or manually

• Ability to halt or roll back instantly if errors spike (and manually)

• Might also be nice to have some web hooks that you could setup with an external observability/product-analytics service to halt or advance rollout

Currently railway only offers to check a health endpoint and abort deployment if that fails. But being able to get real usage with incremental traffic would allow bugs to be caught and prevented from going wide.

There have been lots of high profile outages/crashes in recent memory that probably could have been mitigated by some incremental rollout strategy. And staging is often not enough because it basically never fully mimics production.