Summary
This release enables automatic migration of Databricks Asset Bundles from the Terraform engine to the direct engine when configured and validated, and adds support for declaring environment variables in bundle scripts. Additionally, the experimental SSH connection interactive sessions now resolve Python and pip to the virtual environment interpreter by default, and bundle initialization provides clearer errors for unsupported template URL protocols.
Summary generated by brickster.ai. For the full changelog and any code/binary attachments, follow the GitHub link above.
More from databricks/cli
Explicitly selected profiles now take precedence over authentication environment variables instead of being silently shadowed by them. This release also fixes bundle deployment issues, ensuring permissions from Python mutators deploy correctly and that grants are applied when using the resource selection flag.
This release introduces a new base-environment flag for serverless SSH connections, transitions the AI tools installation command to a plugin-first approach, and limits the labs list command to installable projects. Additionally, it resolves several Databricks Asset Bundle issues, including persistent drift on model serving endpoints, spurious updates on job tasks and clusters, and a migration regression for permissions.
This release improves the workspace export-dir command to sanitize illegal filenames instead of aborting, and updates ssh connect to launch an interactive bash shell starting in the user's workspace home folder. For Asset Bundles, bundle run now outputs a modern job run URL accessible to non-admin users, and several deployment bugs are resolved, including a fix for spurious cluster recreations and a new Terraform-state-based migration implementation.
This release improves the SSH connection experience by fetching server logs on failure, increasing the startup timeout for GPU accelerators, and fixing remote environment setup in Cursor. Asset bundles receive several updates, including support for Unicode letters in variable paths, new Postgres resources for Lakebase branches, and fixes to prevent spurious redeploys caused by backend name normalization.
The direct deployment engine is now generally available and the default for new bundle deployments, which also introduces support for managing Genie spaces as code. New CLI commands include databricks quickstart for an interactive guide and databricks version --check to verify if a newer version is available, alongside several authentication and bundle resource fixes.