I'm also getting inconsistent responses across my replicas even after getting a successful deployment. Sometimes the requests will work, other times they don't. Seems like the deployment may not have replicated correctly.

Apologies but this looks like an issue with the application level code. Due to volume, we can only answer platform level issues.

I've made this thread public so that the community might be able to help with you query.

It was not an application level code issue; the same working image from three days ago when redeployed also started failing after working intermittently for a while, so I very much reject that assessment and consider it an unacceptable answer. The application ultimately ended up not being able to connect to the database for about 10 minutes before coming back online without being touched. It's back now, the latest code is working without changes, deployments are now succeeding.

I'm not going to resolve this thread because I really think you should look into this.

Are you building with nixpacks or a dockerfile?

Nixpacks.

Can you try swapping over to use a Dockerfile? Would be very happy to help you work through setting that up if needed. Generally more control in a dockerfile and its a method of debugging

No, I'm not spending time swapping over to a Dockerfile when it's worked flawlessly up to this point and is currently working, having spontaneously resolved itself. I reject the idea that it's the code, so I won't be spending time debugging my own code when I know it works.

I appreciate your attempt to help, but I'm feeling quite spurned by support on what is firmly a platform issue. Instead, I believe I'll spend my time finding a new home for the clients workloads.

I can understand that. I would love to help but in the end can not force you to do something. I think there are better methods compared to nixpacks but respect your decision.
I can suggest coolify as an alternative but still firmly believe railway would work completely fine.

I'm unwilling to do it because I do not believe that, in this instance, it will be worth it. One of the reasons I value (valued?) Railway so highly is the 1-click deployment of my existing apps without any extra configuration required. For that purpose, nixpacks is not only very good but it's been bullet-proof for the last 18 months over hundreds of deployments. On the strength of the offering I've also recommended Railway to several people.

In this instance, I am 100% confident there was a (hopefully temporary, one-time) platform problem. My problem is not with the outage, outages are a fact of life. In fact, as I type this, Vercel has now gone down and crashed my frontend. Typical. My problem is with the official Railway attitude and response (or lack thereof). I realise I'm not their largest customer (or even a big or medium sized customer), but the absolute lack of willingness to engage or even investigate it on equal terms means I feel I can no longer trust them as a provider.

Anyway, I'm ranting now. Thanks for the coolify recommendation, I will look into it.