risingwavelabs/risingwave v2.6.0 on GitHub

release v2.6.0

SQL commands:
- Supports using CREATE INDEX to create vector indexes, including flat and HNSW methods, and allows creating vector indexes on function expressions. #22432, #22022, #22808
- CREATE SINK supports auto.schema.change option for Elasticsearch, Redshift, and Snowflake. #22611
- DROP CONNECTION supports CASCADE option to automatically drop dependent objects, except for Iceberg connections. #22557
- DROP FUNCTION supports CASCADE option to automatically remove dependent objects. #22991
- Supports using multiple version columns to resolve upsert conflicts. #22931
SQL functions & operators:
- Supports new data type vector(n). #22019
- Supports jsonb_to_array function to convert a JSONB array into a SQL array of JSONB elements. #22777
- openai_embedding accepts a single constant JSONB config parameter instead of separate api_key and model arguments. #22960
- Supports arg_min and arg_max aggregate functions. #22895
System catalog:
- Adds system function has_database_privilege. #22576

Public preview: Iceberg table engine supports copy-on-write write mode for enhanced compaction. #22713
Supports Redshift and Snowflake v2 sink connectors. You can use intermediate table and leverage S3 to improve sink performance, and auto schema change is supported to automatically add new columns. #22984
Iceberg table engine and Iceberg sink support VACUUM [FULL] schema_name.object_name syntax. #22942
NATS connector no longer creates stream automatically unless new option allow_create_stream is explicitly set to true . #22315
Kafka source supports altering group.id.prefix parameter at runtime. #22492
For Kafka source, when querying with a timestamp beyond the latest message, it starts from the latest offset instead of resetting to the earliest. #22413
MQTT source supports multiple comma-separated topics. #22074
Postgres CDC supports parallelized CDC backfill. #22536
For PostgreSQL CDC, if you’ve already created a publication, fewer permissions are required. #22807
Supports TOASTed data from PostgreSQL when using the CDC connector. #22852

Supports deploying lakekeeper as an Iceberg REST catalog via Docker compose or Kubernetes. #23045
Supports using HashiCorp Vault as secret backend with Token or AppRole authentication methods. #22627

Supports Memory-Only Mode for faster, low-latency queries by fully loading operator states into memory. You can configure it at the operator level to balance performance and memory usage. #22857
Adds session variable streaming_separate_consecutive_join to separate consecutive stream joins to improve performance. #22259
Adds session variable streaming_join_encoding to switch join encoding type. #22416
Adds session variable streaming_separate_sink to separate sink. #22827
Adds protect_drop_table_with_incoming_sink to block table drops if incoming sinks exist. #22347
Adds streaming_now_progress_ratio to decouple now_timestamp from epoch_timestamp. #22487