GKE PodUnschedulable: insufficient resources, MatchNodeSelector, taints
A PodUnschedulable pod on GKE has no node that fits. Read the reason: insufficient CPU or memory, a node selector with no matching label, or taints.
What you see
No nodes are available that match all of the predicates: Insufficient cpu (2)- Pod status or events show
MatchNodeSelector PodToleratesNodeTaints,PodFitsHostPorts
What it means
PodUnschedulable means the scheduler cannot place the pod because of insufficient resources or a configuration error. The reason in the event is the predicate that failed.
Important GKE detail from Google's guide: if a pod's resource requests exceed what a single node in any eligible node pool offers, GKE does not schedule it and does not trigger scale-up to add a node. More small nodes will never fit it.
How to confirm the cause
- 1Read the scheduler's reason
kubectl describe pod POD_NAME
The
Eventssection names the failed predicate. - 2See node labels for a node selector problem
kubectl get nodes --show-labels
- 3See what the nodes can offer
kubectl describe nodes
Compare
Allocatablewith the allocated requests. System pods inkube-systemuse resources too.
Causes and fixes
Insufficient resources
- HOW TO TELL
Insufficient cpuor memory in the message.- ROOT CAUSE
- The pod's requests exceed what any eligible node can offer, so GKE does not scale up for it.
- FIX
- Lower the requests in
spec.containers.resources.requests, or add a node pool with larger machines. The default CPU request is 100m.
MatchNodeSelector
- HOW TO TELL
MatchNodeSelectorin the status or events.- ROOT CAUSE
- The labels in the pod's
nodeSelectordo not exist on any node. - FIX
- Make the selector match a real label, or label the node you want.
kubectl label nodes NODE_NAME LABEL_KEY=LABEL_VALUE
Taints the pod does not tolerate
- HOW TO TELL
PodToleratesNodeTaints.- ROOT CAUSE
- Matching nodes carry a taint the pod has no toleration for.
- FIX
- Add a toleration to the pod, or schedule it on nodes without the taint.
Host port conflict
- HOW TO TELL
PodFitsHostPorts.- ROOT CAUSE
- A pod that asks for a
hostPortcan only run where that port is free. - FIX
- Expose the pod through a Service instead of a
hostPort.
How to stop it coming back
- Right-size requests from measured usage.
- Use cluster autoscaler and node auto-provisioning, and make sure a node type large enough for your biggest pod exists.
- Keep node selectors and taints documented.
Keep going
- Runbook: GKE OutOfPods: node has reached its maximum pod count
- Runbook: Pod stuck in Pending with FailedScheduling
- Cheat sheet: kubectl
- Cheat sheet: gcloud
- Lesson: the kubernetes topic in Learn
Sources
Google Cloud documentation (docs.cloud.google.com), read 18 Nov 2026.