Atlas: HPA Not Scaling

What this usually means

HPA is missing metrics, cannot compute utilization, or is constrained by min/max replicas and stabilization behavior. It is a control loop, not a promise.

What to inspect first

Start with HPA status conditions and current metrics.

• Look for condition messages about missing metrics.
• Verify the target workload has resource requests set.

kubectl get hpa -n <ns>
kubectl describe hpa <hpa> -n <ns>
# Metrics sanity:
kubectl top pods -n <ns> || true

Likely causes

Autoscaling failures are mostly signal failures.

• Metrics server not installed or not working; custom metrics adapters missing.
• CPU/memory requests absent; utilization targets can’t be computed meaningfully.
• Target is wrong (label selector points to the wrong object; multiple HPAs fight).
• Stabilization windows and scaling policies prevent rapid change.
• Workload is bottlenecked elsewhere (I/O, locks) so scaling replicas does not change throughput.

Resolution guidance

Make scaling observable and bounded.

• Ensure metrics pipeline is healthy and monitored.
• Set requests deliberately; treat them as economic inputs to scheduling.
• Validate min/max and scale policies match the service’s risk posture.
• If scaling doesn’t improve SLOs, fix the real bottleneck and stop chasing replica counts.