Troubleshooting SkyPilot API Server#

This guide includes tips for troubleshooting common issues with API server deployment.

If this guide does not help resolve your issue, please reach out to us on Slack or GitHub.

Helm deployment troubleshooting#

API server pod is pending#

If the API server pod is pending, you can inspect the pending reason with:

kubectl describe pod -n $NAMESPACE -l app=${RELEASE_NAME}-api

If the pending reason is FailedScheduling and the information indicates there is insufficient cpu/memory, you can either:

  • Add more resources to the Kubernetes cluster, or

  • Use a smaller API server resources request; for example (change the cpu and memory to your desired values):

# Update the resources requests while keeping existing values set in the previous commands
helm upgrade --install $RELEASE_NAME skypilot/skypilot-nightly \
--namespace $NAMESPACE \
--reuse-values \
--set apiService.resources.requests.cpu=4 \
--set apiService.resources.requests.memory=8Gi

Note

API server requires at least 4 CPU cores and 8 GiB memory. Setting lower values may cause degraded performance.