Vertex AI Workbench Instance provisions seemingly forever?

[jvdb]

Creating a Vertex AI Workbench instance works without a problem. However, I wanted to follow a tutorial to connect to an instance from a local IDE, and for that I need to create a Vertex AI Workbench instance that doesn't have an external IP.

When I uncheck the "external ip"-setting when creating a new instance, the provisioning does not terminate. I've tried several times now and even after 10 hours, it's still provisioning. When I create the exact same instance (same hw config, etc.) but don't uncheck that one setting, the provisioning completes in a few minutes.

What can be the cause of this?

[Poala_Tenorio]

It sounds like there might be an issue with the configuration or setup when you uncheck the "external IP" setting during the creation of the Vertex AI Workbench instance. When you choose not to allocate an external IP address to the instance, it might affect the networking configuration or cause delays in provisioning.

When you deselect the option for an external IP address, it may require additional configuration for the internal networking setup. Double-check that the networking settings are correctly configured, including any firewall rules or routes that might be impacting the provisioning process.

Check the logs and monitoring metrics provided by the cloud platform to see if there are any errors or warnings during the provisioning process. This can provide insights into what might be causing the delay.

It's possible that the specific configuration you're selecting without an external IP address is causing resource contention or constraints within the environment. This could lead to longer provisioning times. Try selecting different configurations or sizes to see if the issue persists.

[nimrah-waqar]

Sounds like there might be an issue with the provisioning process when creating a Vertex AI Workbench instance without an external IP. One potential solution could be to create the instance with an external IP first, allowing it to provision successfully, and then modifying the instance to remove the external IP afterward. This approach may circumvent the issue you're encountering during the initial provisioning process. Once the instance is provisioned with the external IP, you can then follow any necessary steps to remove it while ensuring the instance remains accessible as required for your tutorial.

[jvdb · Accepted solution]

Support has helped me resolve this. Even though there is no mention of this in the documentation that I could find, Private Google Access needs to be turned on for the zone you're instantiating a Workbench Instance with External IP disabled.

This can be done in the console at VPC Network -> VPC Networks -> Subnets and then editing the settings for the zone.