Troubleshooting

The following is a list of problems that usually occur during the work process.

Pod Not Running

This syntax is used to detect errors.

kubectl logs <pod-name> -n vllm

Service Not Exposed

Make sure the service is of type LoadBalancer and that your cluster supports the LoadBalancer service.

kubectl describe service llama-31-70b-instruct -n vllm

Resource Limits

Ensure the nodes have sufficient resources (CPU, memory, GPU) to meet the resource requests and limits.