aibrix icon indicating copy to clipboard operation
aibrix copied to clipboard

[bug] Probes for quickstart model kill pod

Open jolfr opened this issue 9 months ago • 1 comments

🐛 Describe the bug

Probes for the sample found in /samples/quickstart/model.yaml restart the pod before the model can fully download.

Steps to Reproduce

  1. Deploy AIBrix Dependencies and Core
  2. Deploy /samples/quickstart/model.yaml
  3. Watch for pod restarts with kubectl get pods
  4. Watch for Readiness/Liveliness failures with kubectl describe pod <pod-name>
  5. Watch logs with kubectl logs <pod-name> --follow

Expected behavior

Model starts up as expected. Changing the readiness and liveliness delay to 360 seconds fixes this issue.

Image

Environment

  • AIBrix Version 0.2.0
  • Kubernetes Deployment
  • Google Kubernetes Engine
  • Deepseek-r1-distill-llama-8b

jolfr avatar Feb 28 '25 23:02 jolfr

Just wanted to log this here in case somebody else runs into it. Will have a patch shortly.

jolfr avatar Feb 28 '25 23:02 jolfr