Pod stuck in Pending with FailedScheduling
A Pending pod has not been placed on a node. Read the FailedScheduling event to see if CPU, memory, taints or a hostPort is the blocker.
What you see
- STATUS stays
Pendingand the pod has no node - Event:
Warning FailedScheduling default-scheduler 0/42 nodes available: insufficient cpu
What it means
The scheduler could not find any node where the pod fits. Until a place is found the pod stays unscheduled, and the scheduler records an event each time it fails.
A pod that is Pending is not broken. It is waiting. The event tells you what it is waiting for.
How to confirm the cause
- 1Read the scheduler's event
kubectl describe pod POD_NAME | grep -A 9999999999 Events
Example from the official docs
Warning FailedScheduling 23s default-scheduler 0/42 nodes available: insufficient cpu
This is the output shape from the Kubernetes docs. Your numbers will differ.
- 2See what each node has left
kubectl describe nodes
Compare
Allocatablewith theAllocated resourcesblock. Allocatable is lower than capacity because system daemons use some of it. - 3List recent events in the namespace
kubectl get events
Events are namespaced. Add
--namespace NAMEif the pod is not indefault.
Causes and fixes
Not enough free CPU or memory
- HOW TO TELL
- The event says
insufficient cpuor points at memory, and the allocated totals on the nodes are close to allocatable. - FIX
- Add nodes, remove pods you do not need, or lower the pod's resource requests if they are higher than the app really needs.
The pod is bigger than every node
- HOW TO TELL
- A single pod requests more than any one node can offer, such as
cpu: 1.1on nodes withcpu: 1. - FIX
- Lower the request, or add a larger node type. More small nodes will never fit it.
Node taints the pod does not tolerate
- HOW TO TELL
- Most nodes carry a taint and the pod has no matching toleration.
- FIX
- Add a toleration to the pod if that is intended, or schedule it to nodes without the taint.
`hostPort` conflict
- HOW TO TELL
- The pod uses a
hostPort. A pod bound to a hostPort can only run on as many nodes as the cluster has. - FIX
- Expose the pod through a Service instead, unless you really need
hostPort.
How to stop it coming back
- Set realistic resource requests from measured usage, not guesses.
- Alert on pods that stay Pending for more than a few minutes.
- Let the cluster autoscaler add nodes if your platform supports it.
Keep going
- Runbook: ImagePullBackOff: the pod can't pull its image
- Runbook: OOMKilled and exit code 137
- Cheat sheet: kubectl
- Lesson: the kubernetes topic in Learn
Sources
- Kubernetes docs: Resource management, FailedScheduling
- Kubernetes docs: Debug Running Pods, Pending Pods
- Kubernetes docs: Debug Pods
Kubernetes documentation (kubernetes.io), read 18 Nov 2026. The docs version selector listed v1.33 to v1.37.