OLake (v0.3.17 - v0.3.18)

February 10, 2026 – February 20, 2026

🎯 What's New

Sources

1. Kafka Avro schema normalization -
   Avro decoding could fail when the schema name / namespace contained invalid characters (for example, -). Fixed this by replacing unsupported characters with _ so schemas follow Avro naming conventions and can be decoded.

2. MSSQL connector integration tests -
   Added integration tests for the MSSQL driver covering full refresh, incremental, and CDC sync to Iceberg and Parquet, including datatype coverage plus CRUD and schema evolution cases.

3. Schema discovery parallelism -
   Added --max-discover-threads to control parallel stream discovery at runtime (default is 50). If you have many streams, you can increase this value to speed up discovery.

Catalogs

1. Glue catalog custom endpoint and credentials -
   Added optional glue_* configs (endpoint, creds, catalog ID) so Glue catalog access can use separate credentials/endpoint from S3, while keeping default behavior unchanged when not set.

Destinations

1. Lakekeeper catalog image tag to v0.11.1 -
   Replaced the latest Lakekeeper/catalog image tag with v0.11.1.

🔧 Bug Fixes & Stability

1. Parquet schema extra column fix for normalization -
   With normalization enabled, Parquet output incorrectly included an extra data column. Fixed this by initializing default OLake columns upfront and only add data when normalization is disabled.

2. Primary key order fix for chunk processing (MSSQL, DB2) -
   Fixed non-deterministic composite primary key ordering during backfill by sorting primary key columns in ChunkIterator, ensuring consistent chunk comparisons and scanning.

3. MySQL timezone offset parsing fix -
   Fixed MySQL timezone offsets not being parsed and falling back to UTC by correctly parsing offsets and using them for timezone conversion, so CDC/full refresh use the right timezone ​

4. Performance test CI destination config fix -
   Updated performance test CI to generate iceberg_destination.json (instead of destination.json), matching the path expected by the test utilities config.

5. MSSQL physloc chunk encoding fix -
   Hex-encoded %%physloc%% chunk boundary values to prevent invalid UTF‑8 in gRPC/protobuf strings, with a fallback to re-encode older state chunks for compatibility.