Why is Kubernetes autoscaling not triggering?

Kubernetes autoscaling can fail to trigger for a few common reasons, and the exact cause depends on which autoscaler you mean:

• HPA: Horizontal Pod Autoscaler
• VPA: Vertical Pod Autoscaler
• Cluster Autoscaler: adds/removes nodes

Here’s a practical checklist.

1) Metrics aren’t available

Autoscaling depends on metrics.

HPA common issues

• metrics-server not installed or broken
• Resource requests are missing, so HPA has nothing to compare against
• CPU/memory metrics aren’t being reported

Check:

kubectl top pods
kubectl top nodes
kubectl get apiservices | grep metrics
kubectl describe hpa <name>

If kubectl top fails, HPA usually won’t work properly.

2) Your deployment has no resource requests

HPA typically scales based on usage vs requested CPU/memory.

Example problem:

• Pod uses 200m CPU
• But CPU request is not set
• HPA can’t calculate target utilization correctly

Check:

kubectl get deploy <name> -o yaml | grep -A5 resources

Add requests/limits like:

resources:
  requests:
    cpu: 100m
    memory: 128Mi
  limits:
    cpu: 500m
    memory: 512Mi

3) HPA target is too high or workload isn’t exceeding it

Maybe autoscaling is working, but load never crosses the threshold.

Check current HPA status:

kubectl describe hpa <name>

Look for:

• current CPU/memory utilization
• target value
• desired replicas

4) Scale-up is being delayed by stabilization or limits

Kubernetes adds protections to prevent flapping.

Reasons:

• stabilizationWindowSeconds
• behavior policies limiting rapid scale-up
• max replicas reached

Check:

kubectl get hpa <name> -o yaml

5) The app isn’t actually under load

Sometimes requests aren’t hitting the pods you expect.

Possible causes:

• load balancer not sending traffic
• caching hides load
• traffic is going to another service
• app is bottlenecked elsewhere

Generate load and observe:

kubectl top pods

6) HPA object is misconfigured

Common mistakes:

• wrong target name
• wrong metric type
• selecting the wrong label set
• scaling a workload that isn’t the owner of the pods

Check:

kubectl describe hpa <name>
kubectl get hpa <name> -o yaml

7) Cluster Autoscaler is blocked

If the issue is node scaling:

• pods are not pending
• node group min/max size reached
• pod requests fit on existing nodes
• cloud provider permissions are wrong
• taints/tolerations or node selectors prevent scheduling

Check pending pods:

kubectl get pods -A | grep Pending
kubectl describe pod <pod>

8) VPA is not designed to “trigger” like HPA

If you’re using VPA:

• it changes pod resource recommendations
• it may restart pods to apply new requests
• it doesn’t scale replica count

So if you expect more pods, VPA won’t do that.

---

Fast debugging path

Run these:

kubectl get hpa
kubectl describe hpa <name>
kubectl top pods
kubectl top nodes
kubectl get deploy <name> -o yaml

Look for:

• missing metrics
• no resource requests
• target not being exceeded
• max replicas reached
• events/errors in describe

If you want, paste the output of:

kubectl describe hpa <name>
kubectl get deploy <name> -o yaml
kubectl top pods

and I can help pinpoint the exact reason.