2025 Releases
2025.5 Engine Release V1.2
-
SQL Engine Enhancements:
-
Query optimizer support for efficient runtime filters. By dynamically generating and applying filter conditions during query execution, significantly reduces data scan volumes, I/O operations, and network overheads. Particularly effective for large table join small table scenarios, delivering substantial performance improvements.
-
Improved SQL engine spill capability; supports adaptive spill to enhance stability for large workloads.
-
Enhanced partitioned table usability: support for truncating parent tables, setting different storage formats per partition (hot/cold tiering), and CDC subscription at the partitioned parent level. See Partitioned Tables.
-
Enhanced pg_hint_plan support; runtime query behavior can be dynamically adjusted via the hint_table configuration. See SQL Hints.
-
-
Incremental Materialized View Enhancements:
-
Improved refresh process with reduced need for table-level locks and better parallel refresh efficiency.
-
New aggregate operators for incremental materialized views, including string_agg(DISTINCT), approximate distinct count (approx_count_distinct), and approximate percentiles (approx_percentile). See Approximate Computing.
-
Support for dynamic modification of materialized view storage format: ALTER MATERIALIZED VIEW mv_name SET ACCESS METHOD columnar/row/hybrid. Dynamic index creation supported. See Materialized Views.
-
-
Multi-Modal Search Enhancements:
-
HNSW index for vector search now supports vector quantization, including int8/fp16 types. Converts original float vectors into more compact types to reduce memory footprint and improve performance. See Vector Search.
-
Feature Store optimization: incremental materialized views now include major window functions: ranking functions (ROW_NUMBER, RANK, DENSE_RANK, NTILE), aggregation over windows (SUM OVER, AVG OVER, COUNT OVER, MIN OVER, MAX OVER), distribution functions (PERCENT_RANK, CUME_DIST), position functions (FIRST_VALUE, LAST_VALUE), lag/lead (LAG, LEAD), population/sample standard deviation (stddev_pop, stddev_samp), population/sample variance (var_pop, var_samp).
-
Added PostGIS extension for geospatial queries in SQL, including spatial relationships, measurement, and geometry operations. See PostGIS Geospatial Query.
-
Optimized relevance ranking for full-text search scenarios. See Full Text Search.
-
Improved JSON support: more JSON PATH query push-down optimizations, index acceleration for CAST filter scenarios, partial indexes for sparse JSON fields for more efficient execution.
-
-
Data Lake Enhancements:
-
Iceberg Table table format supported; integrated with Iceberg REST API Catalog. See Iceberg Foreign Table.
-
Delta Lake table format supported; integrated with Databricks Unity Catalog. See Unity Catalog Foreign Table.
-
Improved experience: Parquet and ORC now support Array types. See External Table Type Mapping.
-
2025.4 Platform Release: Cache
- Accelerated access to cold storage and data lake. See Cache.
2025.4 Platform Release: Nodegroup Auto Suspend
- Auto suspend for Nodegroup instances idle for an extended period, saving compute resources.
2025.3 Platform Release: Catalog
- Enhanced Nodegroup isolation using shared storage, implementing flexible read/write, write/write, and read/read isolation. Suitable for multiple teams sharing a database while maintaining operational independence. See Nodegroup Isolation.
2025.1 Engine Release V1.1
-
Key enhancements in this engine release:
-
Incremental Materialized View: Near-real-time updates via incremental refresh, substantially fresher data with lower resource cost; support for aggregations: COUNT, SUM, MIN, MAX, COUNT DISTINCT, variance, standard deviation; supports HAVING expressions, WITH clauses, and all join types (INNER, LEFT OUTER, RIGHT OUTER, FULL OUTER). See Materialized Views.
-
Improved partitioned table operations: flexible ATTACH/DETACH of partitions, direct data import to parent tables. See Partitioned Tables.
-
Hot/cold tiered storage for balance between performance and cost, reducing storage expenses. See Tiered Storage.
-
Column-level permissions for fine-grained SELECT and UPDATE access control. See Column-level Security.
-
HNSW vector index: improved recall accuracy and performance for billion-scale vector datasets. See Vector Search.
-
SQL Hints mechanism: support for pg_hint_plan extension. See SQL Hints.
-
Enhanced array support: multidimensional arrays. See Arrays.
-
Improved integration and compatibility with ByteBase, AirByte, dlt, dbt.
-
Optimized JSON query efficiency with better column pruning and dictionary encoding.
-
Memory usage optimized for greater efficiency and stability, reducing risk of OOM.
-