Quota and IP exhaustion look alike until you inspect the failed resource
Both failures can leave Pods Pending while the autoscaler tries to add capacity. The decisive evidence is the failed node or address allocation event.
| Signal | Quota or stock | IP exhaustion |
|---|---|---|
| Provider event | Resource quota exceeded or requested machine unavailable | Subnet or Pod secondary range exhausted |
| Check | Regional CPU, accelerator and address quotas; zone capacity | Primary node range, Pod secondary ranges and max Pods per node |
| Fix | Raise quota, change fallback shape or zone | Expand supported ranges, add discontiguous Pod CIDR or redesign density |
| Prevention | Forecast peaks and preflight quota | Size from maximum nodes with growth headroom |
Compute, quota and IP space are three separate capacity budgets.