Segmentation fault on EPP using Inference-scheduler v0.2.0-rc2

[Luca-Calabria]

Hi could you help to solve this SegFault error?

{"level":"info","ts":"2025-07-21T16:54:54Z","logger":"setup","caller":"runner/runner.go:267","msg":"Enabling pprof handlers"}
{"level":"Level(-2)","ts":"2025-07-21T16:54:54Z","caller":"manager/internal.go:196","msg":"Registering metrics http server extra handler","path":"/debug/pprof/heap"}
{"level":"Level(-2)","ts":"2025-07-21T16:54:54Z","caller":"manager/internal.go:196","msg":"Registering metrics http server extra handler","path":"/debug/pprof/goroutine"}
{"level":"Level(-2)","ts":"2025-07-21T16:54:54Z","caller":"manager/internal.go:196","msg":"Registering metrics http server extra handler","path":"/debug/pprof/allocs"}
{"level":"Level(-2)","ts":"2025-07-21T16:54:54Z","caller":"manager/internal.go:196","msg":"Registering metrics http server extra handler","path":"/debug/pprof/threadcreate"}
{"level":"Level(-2)","ts":"2025-07-21T16:54:54Z","caller":"manager/internal.go:196","msg":"Registering metrics http server extra handler","path":"/debug/pprof/block"}
{"level":"Level(-2)","ts":"2025-07-21T16:54:54Z","caller":"manager/internal.go:196","msg":"Registering metrics http server extra handler","path":"/debug/pprof/mutex"}
panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0x1 addr=0x0 pc=0x1da9531]

goroutine 1 [running]:
sigs.k8s.io/gateway-api-inference-extension/pkg/epp/scheduling.NewSchedulerWithConfig(...)
	/go/pkg/mod/sigs.k8s.io/gateway-api-inference-extension@v0.5.0-rc.3/pkg/epp/scheduling/scheduler.go:40
sigs.k8s.io/gateway-api-inference-extension/cmd/epp/runner.(*Runner).Run(0xc0006e3f30, {0x2efe4c8, 0xc0007e8000})
	/go/pkg/mod/sigs.k8s.io/gateway-api-inference-extension@v0.5.0-rc.3/cmd/epp/runner/runner.go:282 +0x5f1
main.main()
	/workspace/cmd/epp/main.go:46 +0xdb

Running the deployer I can see the epp pod failing during initialization reporting this error. The other pods go up and running successfully

[nirrozenbaum]

hi @Luca-Calabria,
Thanks for opening this issue.
looks like you're missing a plugins config file to initialize the scheduler with the plugins configuration.
few options:

• if you're using a plugins config file - you can post here your configuration so we can try to help you understand what could be the issue.
• if you're not using a config file - you should :). we have in the repo some examples here:
  https://github.com/llm-d/llm-d-inference-scheduler/tree/main/deploy/config

this is definitely a bug, we should do a graceful shutdown and log a more user friendly error. I'll take an action item to fix it.

[Luca-Calabria]

Hi @nirrozenbaum thank you for the suggestion. I'll try to setup a config file and I let you know if this fixes the issue

[nirrozenbaum]

pushed a PR for graceful shutdown with an appropriate message.
kubernetes-sigs/gateway-api-inference-extension#1198
this will be included in the next version.

[Luca-Calabria]

Hey @nirrozenbaum if I would to build llm-d by llm-d-deployer, how I can integrate this config file?

[nirrozenbaum]

@Gregory-Pereira can you please point @Luca-Calabria to the relevant doc?

[Luca-Calabria]

We managed to integrate it into llm-d-deployer presets. Thank you!