feat: add graceful shutdown for k8s deployments

[Taure]

Summary

• Add nova:graceful_shutdown/0 public API that handles the full shutdown sequence: delay → suspend listener → drain connections → stop listener
• Called automatically via prep_stop/1 during application shutdown (SIGTERM)
• Configurable via sys.config:
  • shutdown_delay — ms to wait before suspending (lets k8s remove pod from endpoints, default 0)
  • shutdown_drain_timeout — max ms to wait for in-flight requests to complete (default 15000)
• Logs each shutdown phase with structured metadata

Test plan

• Compiles cleanly
• Release builds successfully
• Heartbeat endpoint responds 200
• Graceful shutdown logs all phases (delay → suspend → drain → stop)
• Container stops cleanly on SIGTERM

🤖 Generated with Claude Code

[kivra-hennys]

Looks good.

[burbas]

Should not this behavior be implemented as a callback? Then it will automatically be invoked when calling application:stop(nova) or when the beam machine is shutting down.

[Taure]

I can look into it

[Taure]

But looking at it prep_stop is a callback that is called if application is closed. https://www.erlang.org/doc/apps/kernel/application.html#c:prep_stop/1 so during prep_stop we will do some things before the stop?

[burbas]

But looking at it prep_stop is a callback that is called if application is closed. https://www.erlang.org/doc/apps/kernel/application.html#c:prep_stop/1 so during prep_stop we will do some things before the stop?

You are right. This needs to go into the prep_stop/1 callback since that will be invoked when the application is about to be stopped. stop/1 is called once the application actually is closed, so that won't work in this case.

[burbas]

So my suggestion is that we move the graceful_shutdown into the application behavior. I don't see any value in having it exported in nova-module