topic Re: How to authenticate databricks provider in terraform using a system-managed identity? in Data Engineering

How to authenticate databricks provider in terraform using a system-managed identity?

felix_counter — Mon, 12 Feb 2024 08:56:29 GMT

Hello,

I want to authenticate the databricks provider using a system-managed identity in Azure. The identity resides in a different subscription than the databricks workspace:

According to the "authentication" section of the databricks provider documentation, I performed the following steps:

Grant the (system-assigned) managed identity the "Contributor" role on Subscription B. I can confirm via Azure portal that the app service behind the managed identity indeed has the "Contributor" role on the subscription in which the databricks workspace resides.
Register the managed identity as a databricks service principal in the databricks workspace using its application id.
Initialize the databricks provider with the following arguments:
- host: host address of the databricks workspace
- azure_workspace_resource_id: resource ID of azure workspace, obtained from an "azurerm_databricks_workspace" data object
- azure_client_id: application id of system-managed identity / registered databricks service principal.
- azure_use_msi: true

I tried to create a resource using this provider.

The terraform plan step looks good, i.e. the resource I want to create shows up in the planning step. However, during the apply step I encounter the following error:

Error: cannot create [redacted]: inner token: token error: ***"error":"invalid_request","error_description":"Identity not found"***

This error appears independent of the created resource (I tried several ones). The problems seems to be in the authentication with the managed identity.

Is it possible to authenticate the databricks provider using a system-managed identity? If yes, what would be the correct configuration for the provider and the environment in this setup? I am a bit confused on how to point the provider at the right identity to use. In order to point the provider to the correct identity / SPN, I set the parameter "azure_client_id" to the managed identity's application id. However, I am not sure whether this is correct.

Re: How to authenticate databricks provider in terraform using a system-managed identity?

felix_counter — Mon, 12 Feb 2024 10:55:19 GMT

Dear @Retired_mod,

thanks a lot for your response describing the step-by-step guide to authenticate Databricks using a managed identity.

However, to my best understanding this is not what I want to achieve. To recap, my goal is to use a system-assigned (i.e., not a user-assigned) managed identity of a web app to authenticate with the terraform databricks provider (i.e., not the CLI). I would be very grateful if you could provide a similar step-by-step guide for this setup.

Re: How to authenticate databricks provider in terraform using a system-managed identity?

felix_counter — Tue, 13 Feb 2024 19:42:48 GMT

I furthermore also tried to authenticate using a user-assigned managed identity. In detail, I performed the following steps using Terraform:

Create a user-assigned managed identity in the same resource group as the databricks workspace
Create a databricks service principal setting 'application_id' to the client id of the managed identity.
Assign the managed identity the "Contributor" role on the subscription in which the databricks workspace is located.
Declare a databricks provider setting 'azure_use_msi' to true, 'host' to the databricks workspace url, 'azure_workspace_resource_id' to the resource id of the databricks workspace, and 'azure_client_id' to the application id of the managed identity.
Create a databricks token using said provider

The same error ("Identity not found") occurs during the terraform apply of step 5 (token creation). I also tried creating other resources, they all fail with above-stated error message. @alexott, do you have a suggestion?

Thanks a lot for your support!

Re: How to authenticate databricks provider in terraform using a system-managed identity?

FarBo — Fri, 22 Mar 2024 10:17:18 GMT

@felix_counter

I think I have your answer.

To create a databricks provider to manage your workspace using an SPN, you need to create the provider like this:

provider "databricks" { alias = "workspace" host = <your workspace URL> azure_client_id = <Application ID of the SPN> azure_client_secret = <Application secret of the SPN> azure_tenant_id = <Your Azure subscription tenant ID> }

I store all these credentials as secrets in my Azure KeyVault and call the keyvault to have access to all its secrets. Then I define data fields to retrieve the secret values from my KeyVault and pass them in the databricks provider definition. You probably know you need to use azurerm provider for this. Below is the full block:

data "azurerm_key_vault" "key_vault" { name = <your keyvault_name> resource_group_name = <your rg_name> } data "azurerm_key_vault_secret" "workspace_url" { name = "<Workspace-URL>" key_vault_id = data.azurerm_key_vault.key_vault.id } data "azurerm_key_vault_secret" "workspace_admin_spn_app_id" { name = "<Workspace-ADMINSPN-APPLICATIONID>" key_vault_id = data.azurerm_key_vault.key_vault.id } data "azurerm_key_vault_secret" "workspace_admin_spn_app_secret" { name = "<Workspace-ADMINSPN-APPLICATIONSECRET>" key_vault_id = data.azurerm_key_vault.key_vault.id } data "azurerm_key_vault_secret" "tenant_id" { name = "<AZURE-TENANTID>" key_vault_id = data.azurerm_key_vault.key_vault.id } provider "databricks" { alias = "workspace" host = data.azurerm_key_vault_secret.workspace_url.value azure_client_id = data.azurerm_key_vault_secret.workspace_admin_spn_app_id.value azure_client_secret = data.azurerm_key_vault_secret.workspace_admin_spn_app_secret.value azure_tenant_id = data.azurerm_key_vault_secret.tenant_id.value }

Re: How to authenticate databricks provider in terraform using a system-managed identity?

LuisArs — Wed, 12 Mar 2025 18:23:49 GMT

Hello,

There is a solution for this issue?, I'm facing similar issue on Azure devops with managed identity too.

│ Error: cannot read spark version: cannot read data spark version: failed during request visitor: inner token: token request: {"error":"invalid_request","error_description":"Identity not found"}

Thanks,

Luis

Re: How to authenticate databricks provider in terraform using a system-managed identity?

eparisca — Thu, 13 Mar 2025 13:58:47 GMT

This answer is for authenticating with a service principal, not a managed identity ("secret-less").

I'm also running into the same error and attempted several permutations of the configuration, including attempting to use a databricks_service_principal_password as a secret. All result in different errors.

data "azurerm_databricks_workspace" "this" { name = var.databricks_workspace_name resource_group_name = var.resource_group_name } provider "databricks" { alias = "spn" host = data.azurerm_databricks_workspace.this.workspace_url azure_workspace_resource_id = data.azurerm_databricks_workspace.this.id azure_client_id = data.azuread_service_principal.access_connector.client_id azure_use_msi = true } resource "databricks_token" "access_connector" { provider = databricks.spn comment = "${data.azuread_service_principal.access_connector.display_name} PAT" }

Neither Azure documentation nor Databrick's Terraform provider documentation approach worked.

Re: How to authenticate databricks provider in terraform using a system-managed identity?

eparisca — Thu, 13 Mar 2025 14:35:36 GMT

Additionally, the error seems to be intermittent. It affects us greatly because we are using terraform in ci/cd. When it fails, we must re-run it manually. Upon re-running it works, sometimes.

When running it locally, it fails on the first attempt, the succeeds on the second.

Re: How to authenticate databricks provider in terraform using a system-managed identity?

goTEEMgo — Sat, 12 Apr 2025 15:01:08 GMT

Add an environment variable to you run environment. Add TF_LOG and set it to true. Scroll through and look for an oauth api call. Look at the resource

I have run into the same problem, and it looks our appreg for AzureDatabricks enterprise application is in a different tenant from the rest of our environment.

example:

[DEBUG] provider.terraform-provider-databricks_v1.72.0: GET /metadata/identity/oauth2/token?api-version=2018-02-01&client_id=<*****************>&resource=<*********************>

That's when searched for resource=<*********************> in EntraID, and found the app reg missing.