Fixed: In GKE versions earlier than 1.34.6-gke.1154000 and 1.35.2-gke.1691000, mounting Cloud Storage buckets by using...

In GKE versions earlier than 1.34.6-gke.1154000 and 1.35.2-gke.1691000, mounting Cloud Storage buckets by using the Cloud Storage FUSE CSI driver can experience significant delays. This issue typically manifests as a CreateContainer error that states the following message: failed to reserve container name. This error is self-healing and resolves automatically after the underlying mount operation completes and the container runtime releases the reservation.

The delay is caused by an inefficient bucket access check performed by the CSI driver sidecar by using the ListObjects API method, which can take several hours to complete on buckets that contain millions of empty folders.

The error occurs because the kubelet enforces a strict two-minute timeout for the container creation request. If the FUSE mount process exceeds this time limit while the sidecar is performing the initial bucket access check, then the kubelet cancels the operation and retries. However, the container runtime remains blocked on the first attempt and retains the reservation for the container name.

The new GKE releases fix this issue by replacing the ListObjects check with the GetStorageLayout API method, which performs the same validation but returns almost instantly in most cases.

To resolve this issue, upgrade your cluster to one of the following versions:

• 1.34.6-gke.1154000 or later • 1.35.2-gke.1691000 or later

For GKE version 1.33 clusters running version 1.33.5-gke.2435000 or later, you can mitigate this issue by setting the skipCSIBucketAccessCheck: "true" volume attribute to bypass the check.

There is no supported fix for this issue in cluster versions 1.33.5-gke.2435000 and earlier.