GKE ImagePullBackOff: 401 or 403 pulling from the registry
A 401 or 403 on a GKE image pull means the node cannot authenticate. Check image access, the node service account, access scopes and VPC Service Controls.
What you see
Failed to authorize: failed to fetch oauth token: unexpected status: 403 ForbiddenPulling from host HOST_NAME failed with status code: 403 ForbiddenFailed to authorize: failed to fetch oauth token: unexpected status: 401 UnauthorizedUnexpected status code [manifests 1.0]: 401 Unauthorized
What it means
ImagePullBackOff means the kubelet failed to pull the image but keeps retrying with an increasing delay of up to five minutes. A 401 or 403 in the event tells you the failure is authentication or authorization, not the network.
The pull uses the node's service account unless you give an imagePullSecret. So the question is whether that identity may read the repository.
How to confirm the cause
- 1Read the pull error in the pod events
kubectl describe pod POD_NAME
Match the message to the 401 or 403 forms above.
- 2Check the pod and Deployment spec for the image address and any pull secret
kubectl get pod POD_NAME -o jsonpath='{.spec.containers[*].image}{"\n"}{.spec.imagePullSecrets}' - 3Check the node pool's service account and access scope
gcloud container node-pools describe NODE_POOL --cluster CLUSTER --location LOCATION --format='value(config.serviceAccount,config.oauthScopes)'
For a private Artifact Registry repository the node needs access to it through its service account and its access scope.
Causes and fixes
The identity lacks access to the image (403)
- HOW TO TELL
- A 403 and the repository is private or in another project.
- ROOT CAUSE
- The node's service account, or the pull secret's identity, has no read permission on the registry or repository.
- FIX
- Grant the node service account read access to the repository. Verify access to the image first, as Google's guide does.
The node service account is disabled or deleted (401)
- HOW TO TELL
- A 401 and the account used by the node pool is not active.
- ROOT CAUSE
- A deleted or disabled service account cannot get a token, so authentication fails.
- FIX
- Verify the node service account is active, or move the node pool to an active account.
imagePullSecret is wrong
- HOW TO TELL
- A pull secret is in use and pulls still fail.
- ROOT CAUSE
- The secret's credentials are wrong or expired, or the secret is in a different namespace from the pod.
- FIX
- Check the secret's contents and that it is in the pod's namespace and listed in the Deployment's
imagePullSecrets.
Access scope too narrow, or VPC Service Controls
- HOW TO TELL
- Auth looks right and a private Artifact Registry repository is involved.
- ROOT CAUSE
- The node's OAuth access scope may not include read access, or a VPC Service Controls perimeter blocks the registry.
- FIX
- Check the node's access scope for private Artifact Registry and the VPC Service Controls settings.
How to stop it coming back
- Use a dedicated least-privilege service account per node pool, with registry read access.
- Test the pull from a fresh node pool in staging.
- Alert on pods that stay in ImagePullBackOff.
Keep going
- Runbook: GKE image pull fails: not found, DNS, timeout or slow pulls
- Runbook: ImagePullBackOff: the pod can't pull its image
- Cheat sheet: gcloud
- Cheat sheet: kubectl
- Lesson: the kubernetes topic in Learn
Sources
Google Cloud documentation (docs.cloud.google.com), read 18 Nov 2026.