OLake Go (v0.6.0 - v0.6.5)

March 21, 2026 – April 20, 2026

🎯 What's New

Sources

1. Filters for CDC and incremental syncs -
   Added filtering support for CDC and incremental syncs so users can selectively sync records based on configured conditions.

2. Postgres strict SSL verification support -
   Added verify-ca and verify-full SSL support for Postgres using PEM certificate content input. Also removed support for passing sslrootcert, sslcert, and sslkey as file paths; these fields now expect the actual PEM content instead.

3. Skip ProduceSchema overhead during sync -
   Optimized sync performance by skipping heavy schema inference (like MongoDB collection scans or Kafka message decoding) during sync. Drivers now rely on the pre-discovered schema from streams.json via a new SyncContext, and ProduceSchema is only executed for user-selected streams rather than all available source tables/collections.

4. Partitioning support for non-normalized streams -
   Partition transforms now work correctly when normalization=false for both legacy and Arrow Iceberg writers. Previously, non-normalized streams silently wrote all data to a null partition due to missing column values during pre-shaping and case mismatches in column name lookups.

🔧 Bug Fixes & Stability

1. Datatype utility unit tests -
   Added unit test coverage for utils/typeutils/datatype.go across type detection, comparison, timestamp precision, and SQL type mapping, and fixed invalid reflect.Value handling to map to types.Null.

2. Oracle incremental cursor timezone fix -
   Fixed incremental cursor handling for Oracle TIMESTAMP and DATE columns by stripping the session timezone offset before saving cursor values, so TZ-naive columns don’t shift and re-read already synced records.

3. MongoDB date out of range fix -
   Fixed a fatal crash (json: error calling MarshalJSON) during MongoDB syncs with normalization: false caused by dates exceeding Go's [0, 9999] year limit. Added bounds checking to clamp years < 1 to 1970 and years > 9999 to 9999, ensuring dates safely map to valid JSON timestamps.

4. Clear destination support for GCP -
   Fixed clear destination for GCP buckets by handling the unsupported bulk delete operation and added retry backoff.

5. Increase Java writer startup timeout -
   Extended the Java writer process startup timeout from 30 seconds to 600 seconds (10 minutes).

6. MongoDB nested dates out of range fix -
   Fixed fatal crashes (json: error calling MarshalJSON) when unnormalized MongoDB documents contained nested DateTime values with years outside Go’s supported [0, 9999] range by configuring the MongoDB client to decode BSON DateTime into safe, clamped time.Time values at decode-time.

7. PostgreSQL backfill chunking for multi-level partitions -
   Fixed a bug where multi-level partitioned PostgreSQL tables were silently skipped during backfill. The chunk discovery query now properly finds all leaf partitions (using pg_partition_tree for PG 12+ and a recursive pg_inherits CTE for older versions) instead of only looking one level deep, and stale statistics without an ANALYZE run now return a clear error rather than silently skipping the table.

8. Column selection integration tests -
   Added integration tests for column selection to verify that only specified columns are synced. Tests cover both inclusion and exclusion logic, as well as validating that columns removed from the source during a backfill stop being ingested while the remaining columns continue processing correctly.

9. Go Security workflow vulnerability and Go version update -
   Updated the Go Security workflow to make govulncheck respect defined exemptions (failing only on unexempted vulnerabilities) and upgraded the Go version to v1.25.9 (which includes recent security fixes).

10. Kafka integration test -
    Added an integration test for the Kafka driver to validate end-to-end sync behavior, and fixed parsing and comparison bugs that could cause incorrect data transformation and validation mismatches.

11. MSSQL boolean filter literal fix -
    MSSQL does not support TRUE/FALSE as SQL literals, so filter SQL generation now converts boolean values to 1/0 instead. This fix is applied consistently across both the structured filter_config and legacy string filter paths.

12. Kafka sync: skip unparseable last partition message -
    When the last message in a partition was neither JSON nor Avro, the sync had no way to parse it and would stall indefinitely, waiting until the context was cancelled. Fixed by skipping such messages — the offset is still marked in Kafka to advance past it, but the message is not written to the destination, allowing the sync to continue cleanly.