Skip to content

Scaling

Replicas

Capacity for concurrent requests is a function of replicas per model. Increase the replica count in the serving stack configuration and re-apply.

Each replica needs its own GPU allocation - VRAM requirements are per replica, not per model. See Compute.

Nodes

When replicas cannot be scheduled, add GPU nodes:

kubectl get pods -n shaide-serving --field-selector status.phase=Pending
kubectl describe pod <pending-pod> | grep -A10 Events

Insufficient nvidia.com/gpu means the cluster is out of GPU capacity.

Signals

Signal Meaning
429 responses Replicas saturated
Rising time-to-first-token Queueing
Pods Pending Out of GPU capacity

Grafana dashboards ship with the platform - see Observability.

Cost

Scale unused node pools to zero when idle. GPU nodes dominate cost, and models hold VRAM for as long as they are scheduled.