Verify installation¶

Pods¶

kubectl get pods -A | grep -E 'shaide|harbor|istio|serving'

All pods should be Running or Completed. Model serving pods are the slowest to become ready - large weights take several minutes to load on first start.

Endpoint¶

kubectl -n shaide get svc shaide-server

An EXTERNAL-IP should be assigned.

API¶

curl https://<endpoint>/v1/models -H "Authorization: Bearer <key>"

Expect a JSON list of served models.

Generation¶

curl https://<endpoint>/v1/chat/completions \
  -H "Authorization: Bearer <key>" \
  -H "Content-Type: application/json" \
  -d '{"model":"<model-id>","messages":[{"role":"user","content":"ping"}]}'

GPU allocation¶

kubectl describe node <gpu-node> | grep -A5 "Allocated resources"

nvidia.com/gpu requests should be non-zero once models are scheduled.

If something fails¶

Symptom	Check
Pods `Pending`	Node capacity, GPU availability, PVC binding
Pods `ImagePullBackOff`	Registry trust - Node registry trust
`503` from the API	Model still loading; watch pod status
No `EXTERNAL-IP`	No LoadBalancer provider - Networking

See Troubleshooting.