What risingwavelabs/risingwave shipped

RisingWave v3.1.1 is out. This week: a new MariaDB CDC source, fixed scan range pushdown results, and no more empty BigQuery AppendRows requests. BigQuery upserts are now ordered by change sequence number, and SQL Server and Postgres CDC handle dotted table names.

RisingWave v3.1.1 is released. This week also added a first-class MariaDB CDC source and fixed wrong results and a possible panic in scan range pushdown. The BigQuery sink no longer sends empty AppendRows requests, and BigQuery upsert changes are ordered with a change sequence number, both backported to earlier release branches. SQL Server and Postgres CDC now support table names containing a dot, DROP CONNECTOR works on refreshable tables, and stored definitions that fail to parse return an error instead of being accepted.

Weekly roundup: generated columns now enforce NOT NULL, IS NOT DISTINCT FROM parses tighter than AND/OR, escaped quoted identifiers work, and several query correctness fixes landed for outer joins, array subqueries and watermarks.

This week's changes to the repository focus on query correctness and connector behavior. Generated columns now enforce NOT NULL, and IS [NOT] DISTINCT FROM is parsed tighter than AND and OR. Escaped double-quoted identifiers are tokenized correctly, full outer hash joins with non-equi conditions no longer add NULL rows, correlated ARRAY subqueries no longer emit transient NULLs in materialized views, and watermarks are no longer derived through timestamp plus or minus months. On the CDC side, offsets persisted on non-checkpoint barriers are committed, Kinesis reader retry intervals parse from strings, and Debezium topic names keep table names unsanitized.

This week: Iceberg orphan file cleanup, coordinated PK-index compaction, Iceberg V3 change log truncation, GCS vended credentials with Google auth, native Pulsar JSON schemas, and database filters in diagnostic reports. Several fixes for correlated subqueries, BigQuery upsert ord

This week the project added Iceberg orphan file cleanup, meta coordination of PK-index Iceberg compaction, Iceberg V3 change log truncation, and GCS vended credentials with native Google authentication for Iceberg REST catalogs. Connector support widened with native Pulsar JSON schemas, and diagnostic reports can now be filtered by database. A range of fixes landed as well: correlated references bind to their owning scope, BigQuery upsert changes are ordered by _CHANGE_SEQUENCE_NUMBER, state table reads hide rows below the pending watermark, and base64 Pub/Sub credentials work again. CDC sources also received fixes for privilege checks, SQL Server name casing and binlog monitoring.

This week: asof joins clean both sides by watermark, Iceberg V3 compaction with barriers, new mssql_query table function, automatic Oracle heartbeat tables, and fixes for streaming sum, TTL expiry and CDC backfill recovery.

This week's RisingWave changes cover streaming correctness, connectors and storage. As of joins now clean both tables by the watermarks of both sides, Iceberg V3 compaction is applied with barriers, and the primary key index gains bounded compaction planning. A mssql_query table function was added, Oracle heartbeat tables can initialize automatically, and Pulsar consumer termination is retried. Fixes include streaming sum outputting NULL when no non-NULL input remains, TTL rows expiring by the forwarded watermark, arrangement backfill forwarding buffered rows on a barrier, and CDC backfill recovering from snapshot read failures.

This week: Iceberg V3 as an independent job, native Pulsar Avro schemas, a bounded-memory sort buffer for the locality provider, explicit table cache warm-up, and fixes across CDC recovery, sink output on primary key changes, and streaming time travel.

This week in the project: Iceberg V3 can be maintained as an independent job in the meta service. Pulsar sources now support native Avro schemas, so they can infer columns and decode messages written with different schema versions. A bounded-memory sort buffer for the locality provider was added to mitigate join amplification, and table caches can now be warmed up explicitly. Iceberg ADLS file IO configuration is shared, and variant access paths are prebuilt. On the fix side, sinks keep the old primary key delete when an update changes the sink primary key, upstream rows of finished vnodes are forwarded after a backfill reschedule, SQL Server CDC validates table name casing, and time travel is rejected in streaming jobs.

Weekly digest: improved CDC connection shutdown for PostgreSQL and SQL Server, upgraded storage cache library, fixed object store metadata error handling, and refactored checkpoint job coordination.

This week's RisingWave updates improve operational stability and performance. CDC sources now properly force-close PostgreSQL and SQL Server connections during shutdown, preventing thread hangs and resource leaks. Storage performance improved via Foyer 0.22.6 upgrade. Object store listing now correctly propagates metadata errors. Additional checkpoint job coordination improvements in the metadata layer.

Risingwave 3.1.0 is out. New: broadcast temporal joins, keyed changelog execution, Oracle CDC plumbing, Iceberg compaction health metrics. Plus stability fixes across streams and refresh handling.

Risingwave 3.1.0 is available. This release adds broadcast temporal join support for stream processing, keyed changelog planning and execution, and Oracle CDC connector plumbing. New Iceberg compaction health metrics are now exposed, and multiple stability improvements cover table refresh handling, MATCH_RECOGNIZE scans, and source privilege enforcement.

V3.0.4 shipped with compaction stability, watermark improvements in over-window queries, Google Pub/Sub ADC auth, and connector fixes for SQL Server primary keys.

V3.0.4 released with compaction group stabilization, watermark forwarding through over-window operators for reduced latency, Google Pub/Sub Application Default Credentials support, and SQL Server composite primary key ordering fixes.