GKE Cluster Shows "Starting" Even After its turned on

Kayla — Fri, 13 Jun 2025 11:54:16 GMT

Curious if anyone else has run into this. After changing to GKE based clusters, they all turn on but don't show as turned on - we'll have it show as "Starting" but be able to see the same cluster in the dropdown that's already active. "Changing" to that one will let us run code.

Re: GKE Cluster Shows "Starting" Even After its turned on

mark_ott — Tue, 04 Nov 2025 17:51:26 GMT

Yes, others have reported encountering this exact issue with Databricks clusters on Google Kubernetes Engine (GKE): after transitioning to GKE-based clusters, the UI may show clusters as "Starting" even though the cluster is already up and usable in practice—such as being visible in dropdowns and capable of running code after selecting it. This appears to be a recognized display or state synchronization problem rather than a fundamental failure of the cluster itself.

What Users Are Reporting

Clusters stay in "Starting" or "Pending" states in the Databricks UI, but the compute is accessible and operational from dropdowns or notebook actions.
Users are able to shift execution to the active cluster, which then allows them to run code, indicating the backend state isn't matching the UI state.

Common Causes and Investigations

This may occur due to delays or errors in Databricks' backend synchronizing state information from GKE to the workspace UI.
Issues with network configuration, firewalls, or role permissions can sometimes contribute to similar symptoms, especially after architectural changes like moving to GKE.
There have been reports across Databricks' forums about clusters being stuck in transition states for hours, sometimes resolved only by adjusting firewall rules or reviewing service account permissions.

Workarounds and Notes

"Changing" to the same cluster in the dropdown or refreshing your workspace may allow job execution even if the UI says "Starting".
If the problem persists across different sessions or users, checking cluster logs, Kubernetes events, or escalating to Databricks support is recommended, especially if recent changes to networking, permissions, or Databricks Runtime were made.

Summary Table

Symptom	Usable Workaround	Known Causes	Additional Steps
Cluster UI shows "Starting"	Select active cluster; run code	Backend sync lag	Check network/perm issues
Cluster stuck in "Pending"	Sometimes needs config tweak	Firewall/permissions	Review/adjust cloud settings
Still not working	--	Various/cloud issues	Contact Databricks support