Databricks workspace in our own VNET

[ArjenSmedes]

We have setup a Databricks workspace in our own Azure VNET, including a private endpoint. Connecting to the WS works fine (through the private ip address). However, when creating my first cluster, I run into this problem:

"ADD_NODES_FAILED

...

Failed to get instance bootstrap steps from the Databricks Control Plane. Please check that instances have connectivity to the Databricks Control Plane.

...

and then further on: 

Non-authoritative answer: adb-3485221775054017.17.azuredatabricks.net canonical name = adb-3485221775054017.17.privatelink.azuredatabricks.net. adb-3485221775054017.17.privatelink.azuredatabricks.net canonical name = westeurope-c2.azuredatabricks.net. Name: westeurope-c2.azuredatabricks.net Address: 40.74.30.80."

Now, what we haven't yet done is setup an entry in our private DNS zone (we need to setup privatelink.azuredatabricks.net on our site, which will be done tomorrow morning hopefully). 

Question: Is it true that the data plane is running IN our VNET and thus, when DNS has been fixed, it will pick up the right ip address and provisioning of the cluster will work?

(just so I understood the concepts properly).

Kind Regards,

Arjen

[Kaniz · Accepted solution]

Hi @ArjenSmedes, Yes, it is true. The Databricks data plane runs within your VNet. When you set up your private DNS zone, it will pick up the correct IP address, and the provisioning of the cluster should work.

The error message you're seeing, "Failed to get instance bootstrap steps from the Databricks Control Plane", indicates that your Databricks instances cannot connect to the Databricks Control Plane. This is a common issue when the cases are deployed inside a VNet and do not have the necessary connectivity to the Databricks Control Plane. Setting up a private DNS Zone is recommended when deploying Databricks in your own VNet. This DNS setup is needed to resolve the Databricks Control Plane's URL to the private IP address of the remote endpoint.

Here is a summary of the information from the given sources:- When you deploy a Databricks workspace in your own VNet, the workspace data plane (which includes the Databricks clusters) runs within your VNet. This is a VNet injection.
 
- When you use the private link, the control plane of your workspace is accessed over a private network connection. When you create a private endpoint for your workspace, you should create a DNS record in your private DNS zone that maps the URL of the Databricks Control Plane to the private IP address of the private endpoint.

Therefore, once your DNS has been set up correctly, your instances should be able to connect to the Databricks Control Plane, and the provisioning of your cluster should proceed without issues.