KubeHpaReplicasMismatch

A HorizontalPodAutoscaler has decided on a replica count, but the workload has been stuck at a different number for a while.


Severity	warning
Source	kube-state-metrics v2.x (autoscaling/v2)
Key metrics	`kube_horizontalpodautoscaler_status_desired_replicas`, `kube_horizontalpodautoscaler_status_current_replicas`

What it means

The HPA computes a desired replica count from its metrics. Normally the current count catches up within a minute or two. This alert fires when desired and current differ, the HPA is somewhere between its min and max, and the current count has not changed at all for a sustained period. In other words: scaling was requested and nothing happened.

The usual impact is under-provisioning during a load spike: latency rises while the autoscaler “thinks” it has already reacted.

Common causes

New pods are Pending: not enough node capacity, or the cluster autoscaler cannot add nodes.
A namespace ResourceQuota blocks new pods.
Pods start but never become Ready (failing probes, crash loops, image pull errors).
Another controller or a GitOps tool keeps resetting spec.replicas on the Deployment.
Scale-down is held back by a behavior stabilization window or policy (expected, but long windows can trigger this).

First checks

Compare desired and current across HPAs:

kube_horizontalpodautoscaler_status_desired_replicas - kube_horizontalpodautoscaler_status_current_replicas != 0

Read the HPA’s conditions and events:
```
kubectl -n <ns> describe hpa <hpa>
```

Look for pods that are not running and why:

kubectl -n <ns> get pods -l <selector> --field-selector=status.phase=Pending
kubectl -n <ns> describe pod <pending-pod> | sed -n '/Events/,$p'

Check the ReplicaSet for quota or admission errors:

kubectl -n <ns> describe rs <replicaset> | grep -iE 'FailedCreate|quota|forbidden'

Check whether something else is writing replicas (look at managedFields managers):

kubectl -n <ns> get deploy <name> --show-managed-fields -o yaml | grep -E 'manager:|replicas'

Fixing it

Unblock the thing new pods are waiting on: add node capacity or fix the cluster autoscaler, raise the quota, or fix the readiness problem. If a GitOps tool owns spec.replicas, remove that field from the manifest (or ignore it in the sync diff) so the HPA is the single owner.

KubeHpaMaxedOut: the HPA has hit its ceiling instead.
KubeHpaUnableToScale: the HPA cannot even update the scale subresource.
KubeQuotaFullyUsed: a full quota is a common blocker.
KubeDeploymentReplicasMismatch: the Deployment side of the same gap.