databricks/terraform-provider-databricks v1.121.0

on GitHub

Important Changes

• [Backward Incompatible] Resource databricks_mws_ncc_private_endpoint_rule has been refactored in this version to fix multiple bugs in its implementation and behavior. Server-populated attributes are now strictly read-only (writes that previously produced perpetual plan drift now error at plan time), plan churn on previously-Optional output fields is eliminated, and the new Plugin Framework implementation (behind DATABRICKS_TF_ENABLED_PF_RESOURCES) polls the rule's connection_state during Create so a failed or unusable provisioning state surfaces at apply time instead of on the next plan. See Breaking Changes and New Features and Improvements below for specifics.

Breaking Changes

• Tighten read-only attributes on databricks_mws_ncc_private_endpoint_rule (#5819). The attributes rule_id, account_id, endpoint_name, vpc_endpoint_id, connection_state, creation_time, updated_time, deactivated, deactivated_at, error_message, and gcp_endpoint.psc_endpoint_uri are now computed-only and can no longer be set in HCL. They are populated by the server in every API response (and psc_endpoint_uri is unconditionally overwritten from the cloud-platform truth on every read), and previous releases accepted writes to these attributes silently while the API ignored or overwrote the values, producing perpetual drift on every plan. Configurations that explicitly assigned any of these attributes must remove the assignment; the value is still available in state after apply.

New Features and Improvements

• A Plugin Framework implementation of databricks_mws_ncc_private_endpoint_rule is now available behind DATABRICKS_TF_ENABLED_PF_RESOURCES=databricks_mws_ncc_private_endpoint_rule (#5819). Once opted in, terraform apply waits for the rule to leave CREATING before returning: PENDING and ESTABLISHED succeed, while a CREATE_FAILED, REJECTED, DISCONNECTED, or EXPIRED connection state surfaces as an apply-time error instead of on the next plan. This differs from the SDKv2 implementation, which returns immediately without polling. The Plugin Framework implementation will replace the SDKv2 implementation in a follow-up version.
• Deprecate the SDKv2 fallback implementations of databricks_library, databricks_quality_monitor, and databricks_share resources, and the databricks_share, databricks_shares, and databricks_volumes data sources. These resources have been served by the Plugin Framework by default since their migration; the SDKv2 implementations remain only as opt-in fallbacks via the USE_SDK_V2_RESOURCES / USE_SDK_V2_DATA_SOURCES environment variables. Setting either environment variable for any of these names now emits a runtime warning (visible with TF_LOG=WARN or higher), and the SDKv2 implementations will be removed in the next major release of the provider.
• Add resource and data sources for databricks_ai_search_endpoint.
• Add resource and data sources for databricks_ai_search_index.
• Add clear_cloud_attributes_on_remove to databricks_cluster (#5812). When set to true, removing a cloud attributes block (aws_attributes, azure_attributes, gcp_attributes) from the configuration clears it instead of the removal being silently suppressed. The flag defaults to false, preserving the existing diff-suppression behavior that prevents perpetual drift from platform-returned cloud attribute defaults. Keeping a block, even partially specified, is unaffected; only removing the whole block clears.

Bug Fixes

• Fix import for jobs with >100 tasks (#5417).

Documentation

• Added an example to databricks_budget for creating budgets to control Genie usage costs (#5817).

Exporter

• Generate code in import.sh more safely (#5848).

Internal Changes

• Deprecate instance_profiles attribute, and replace it with roles in databricks_group data source (#5086)