Summary
dbt-databricks 1.12.3 adds `catalog_database` support to route Unity catalog models independently of dbt project catalog names. The release fixes connection initialization races, streaming table tag spurious changes, view refresh materialization, shallow clone rebuilds, Iceberg Python model formatting, non-ASCII column handling, and dropping governed columns.
Summary generated by brickster.ai. For the full changelog and any code/binary attachments, follow the GitHub link above.
More from databricks/dbt-databricks
Version 1.12.2 adds catalogs.yml v2 support, a skip_optimize model config for granular control over post-materialization optimization, and support for the SQL connector's Rust kernel backend. The release fixes numerous issues with incremental materializations, including constraints being dropped and re-added incorrectly, column-level tags not persisting, and view metadata (comments, properties) being lost during updates.
1.12.1
Version 1.12.1 enables correlation between dbt runs and Databricks workflows via exposed job/run IDs and adds SPOG account-level URL routing support. The release fixes unexpected schema diffs on streaming tables, snapshot column tagging, and Iceberg incremental clustering loss, and now requires explicit `contract.enforced: true` for column-level constraints (a breaking change).
v1.12.0 adds metric views, row filters, Python UDFs, and new SCHEDULE EVERY and TRIGGER ON UPDATE refresh modes for materialized views and streaming tables, plus key-only tag support. Breaking change: databricks_tags now merge additively across hierarchy levels rather than child configs completely replacing parent.
v1.11.8 adds invocation_id to default query comments and enforces a 255-character identifier length limit for Databricks relations. The release also eliminates multiple spurious warnings that previously appeared on every run and fixes metadata handling for Hudi/Iceberg tables, materialized views, and catalog merges.
Added support for notebook-scoped packages in command submits and notebook job runs. Fixed duplicate alias generation in --empty mode, added insert-by-name support to microbatch and replace_where merge strategies, and improved capability detection for named compute.
Concurrent microbatch execution is now enabled and optimize() has been added to snapshot and table materializations for improved performance. Fixes address catalog name quoting in schema commands, streaming table cluster clauses, column-level tag application, and constraint configuration validation.