Evicted
The kubelet evicted the pod because its node crossed an eviction threshold, or because the pod exceeded its own ephemeral-storage limit. The pod object stays behind in phase Failed with reason Evicted; its controller creates a replacement.
Diagnose in order
Section titled “Diagnose in order”- Which resource? It’s named in the message:
memory,ephemeral-storage,nodefs,imagefs, orpids. - Is it one node or many?
kubectl get pods -A --field-selector status.phase=Failed -o wide. - Who actually used the resource? The message often names the container and compares usage with its request. A pod using 1450Mi against a 256Mi request is why the node was overcommitted, even if it isn’t the pod that got evicted.
Usual causes
Section titled “Usual causes”| Resource | Cause |
|---|---|
memory |
Pods using far more than they request; the node is overcommitted |
ephemeral-storage |
Container logs, emptyDir volumes, or writes to the container filesystem |
nodefs / imagefs |
Node disk full: images, logs, or local volumes |
pids |
A process leak, often a shell loop or a runaway worker pool |
Set requests close to real usage, add ephemeral-storage limits to pods that write to disk, rotate logs, and size nodes for their workloads. Clean up the leftover records with kubectl delete pods -A --field-selector status.phase=Failed once you’ve read them.
Read next: Node pressure and eviction.