Release Release v0.15.2 · databrickslabs/lakebridge

Release v0.15.2

Simplified pdcr workload extraction (#2654)
The pdcr BI workload extract now groups by just QueryType and WorkloadBucket instead of thirteen dimensions, dropping the PDCRINFO.UserInfo join and the organizational/user/hourly breakdowns that the migration model doesn't need. This eliminates the join spool that could stall production runs on large repositories. Adds a new td_pdcr_bi_concurrency_extract tracking total BI queries and average BI query starts per minute over a 180-day window.

Fixed BigQuery timestamp fidelity (#2594)
BigQuery TIMESTAMP and DATETIME columns are now formatted explicitly (FORMAT_TIMESTAMP/FORMAT_DATETIME) so their row hashes match Spark's, instead of relying on a plain CAST(... AS STRING) that emitted timezone offsets and different fractional-second precision. This clears false row/data/all mismatches on any correctly-migrated table containing those types. Also fixes false schema mismatches on multi-field STRUCTs, and adds warnings for width-lossy types (INT64, FLOAT64, DATETIME) that BigQuery cannot distinguish from narrower targets.
Extracted column transformations into a ColumnTransformer service (#2601)
Refactors the reconcile query builders so column transforms are resolved by a dedicated ColumnTransformer service that holds both the source and target sides, replacing the inline per-type logic previously scattered across the builders. No change to query output — this is a structural change that gives cross-engine numeric rendering a place to live.

Full Changelog: v0.15.1...v0.15.2