KubeSchedulerDown

kube-scheduler used to be monitored in this cluster and now no instance of it is answering.


Severity	critical
Source	kube-scheduler `/metrics` (HTTPS, port 10259)
Key metric	`up`

What it means

The alert only considers clusters where the scheduler was successfully scraped at some point recently, so managed clusters that never expose the scheduler do not trigger it. It fires when every scheduler target has then been down for a sustained period.

Without a scheduler, running pods keep running, but new pods stay Pending forever: rollouts, scale-ups, evicted pods and jobs all stop making progress.

Common causes

The kube-scheduler static pod crashed, often after a manifest edit in /etc/kubernetes/manifests/ with a typo.
Its kubeconfig certificate expired, so it cannot talk to the API server.
A kubeadm upgrade reset --bind-address to 127.0.0.1, making metrics unreachable from Prometheus while the scheduler still works.
The control plane node hosting it is down.

First checks

Is anything actually unscheduled?
```
kubectl get pods -A --field-selector=status.phase=Pending
```
Fresh pods being scheduled means the scheduler works and only the scrape broke.

Check the scheduler pods and restarts:

kubectl -n kube-system get pods -l component=kube-scheduler -o wide
kubectl -n kube-system logs -l component=kube-scheduler --tail=100

See who holds the leader lease:

kubectl -n kube-system get lease kube-scheduler -o yaml

On a control plane node, check the container and the bind address:

sudo crictl ps -a --name kube-scheduler
sudo grep bind-address /etc/kubernetes/manifests/kube-scheduler.yaml

Check certificate expiry: sudo kubeadm certs check-expiration.

Fixing it

Fix the manifest error or renew certificates (kubeadm certs renew scheduler.conf), and the kubelet restarts the static pod automatically. If only metrics are unreachable, set --bind-address=0.0.0.0 (or the node IP) and make sure the scrape uses HTTPS on 10259 with a token that is allowed to read /metrics.

KubeControllerManagerDown: the same static-pod failure modes apply.
KubeAPIDown: the scheduler cannot work without the API server.
KubeVersionMismatch: half-finished upgrades often break components.