Hi,

I have a question about scaling down workers.

I am running a worker service on Railway. Each replica runs a long process that pulls jobs from Redis and processes them. Some of these jobs take a long time to finish.

I am using the GraphQL API to autoscale the replicas based on load. Scaling up works fine. But when scaling down, Railway seems to randomly stop some replicas. Sometimes it stops a replica that is still processing a job, and that job gets interrupted.

I wanted to ask:

• Is there any graceful shutdown or draining when replicas are scaled down?

• Is there a way to let a replica finish its current job before it is stopped?

• Can we control which replica gets terminated, or add some delay before shutdown?

• What is the recommended way to run long-running background workers with autoscaling on Railway?

Right now it looks like the replica is stopped immediately, which is risky for job processing. I just want to check if there is a safer way to handle this.

Thanks for your help!