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.

Ingress-nginx controller conflicts

If you encounter similar errors as below when deploying the API server:

Error: Unable to continue with install: IngressClass "nginx" in namespace "" exists and cannot be imported into the current release: invalid ownership metadata; annotation validation error: key "meta.helm.sh/release-name" must equal "test": current value is "skypilot"

It is likely you have an existing ingress-nginx controller deployed in the Kubernetes cluster. Or have an existing SkyPilot API server deployed which by default installs an ingress-nginx controller. If this is the case, you can follow the instructions in Reusing ingress controller for API server to reuse the existing ingress-nginx controller to solve this issue.