Separates a transient shortfall from nodes still rejoining after an upgrade, genuine cluster-wide CPU contention needing more capacity, and a pod whose CPU request is simply too large to fit on any node in the pool at all. It does not cover FailedScheduling caused by node affinity, taints/tolerations, or Insufficient memory/pods — the same describe-pod approach applies, but the resource name in the message and the fix differ.
Unverifiedno reproductions yetWhy this confidence?8 steps. Every step is written out below in full — the interactive version simply follows the branches for you.
This is a capacity-fit problem: the sum of already-scheduled CPU requests plus this pod's request exceeds what any single node currently has free. It's common right after a cluster or node-pool upgrade because upgrades typically cordon and replace nodes in batches, temporarily shrinking available capacity.
What happens next
Read the FailedScheduling event text precisely — the same mechanism applies to memory, taints, or affinity, but the fix is different for each.
kubectl describe pod <pod-name> -n <namespace>Expected result
Warning FailedScheduling 12s (x8 over 4m) default-scheduler 0/6 nodes are available: 6 Insufficient cpu.What happens next
Look for NotReady nodes, nodes marked SchedulingDisabled (cordoned), or nodes that were only just created and haven't finished joining.
kubectl get nodes -o wideExpected result
NAME STATUS ROLES AGE VERSION
node-3 Ready,SchedulingDisabled <none> 400d v1.33.0
node-4 Ready <none> 2m v1.33.0What happens next
If the message names a different resource (Insufficient memory, Insufficient pods) or mentions node affinity/taints instead, the diagnosis approach here still applies (check what the scheduler is actually comparing) but the specific fix does not — resize or reconfigure the resource actually named in the message.
Some nodes are cordoned or NotReady, or were only just created — this is the node rollout that usually accompanies a cluster or node-pool version upgrade. Capacity should recover once the rollout finishes; if using a managed node pool or cluster autoscaler, confirm it isn't itself stuck (quota limits, a node template that fails to boot) before assuming it will resolve on its own.
All nodes are Ready and none are cordoned — the question now is whether the pod could fit anywhere at all, or whether it's simply contention for space that does exist.
kubectl describe node <node-name> | grep -A5 Allocatable; kubectl get pod <pod-name> -n <namespace> -o jsonpath='{.spec.containers[*].resources.requests.cpu}'Expected result
Allocatable:
cpu: 3800m
memory: 14Gi
2000mWhat happens next
The pod's request does fit under some node's allocatable CPU, so this is contention rather than an impossible fit: the cluster is at capacity for schedulable CPU right now. Check that the cluster autoscaler (or manual node pool scaling) is enabled, has headroom against any cloud-provider quota, and is actually configured to add the node type this workload needs.
If the pod's CPU request exceeds the allocatable CPU of every node type available, no amount of autoscaling fixes it — the scheduler will never find room. Reduce the request to something a real node can satisfy, or move the workload to a larger node pool/instance type.
Changes system or service state. Review before running.
Patches the deployment's pod template with the new CPU request, which triggers a rolling restart of the deployment's pods.
kubectl set resources deployment/<deployment-name> -c=<container-name> --requests=cpu=500m -n <namespace>What would strengthen it: 6 more independent reproductions. Reproductions from 3 more distinct environments.