2026

Category

Feature

Description

References

New Feature

Storage node

Log Store migration is supported. You can adjust the Log Store connections of the primary node and virtual warehouses by Content ID. Cluster-level migration of faulty Log Stores, WAL freshness verification, and node isolation are supported.

N/A

File cache

SQL functions are added to manually evict DADI Fragment caches and query Fragment file paths, buckets, cache status, and the number of open files.

N/A

Execution engine

The Laser Runtime Filter size can be configured. Monitoring for Laser Motion network received bytes is added.

N/A

Optimization

Storage node

Page Store Tenant migration capabilities are enhanced. Connection configurations can be persisted after migration and applied to the primary node and virtual warehouses. Offline migration is supported when the source Page Store is permanently unavailable.

N/A

Beam storage engine

For Beam tables without SortKey or statistics collection configured, Stats Fork reads are skipped to reduce unnecessary OSS requests and the risk of triggering QPS throttling.

N/A

Execution engine

Hash Join memory allocation, Runtime Filter, query plan cleanup, and Laser execution logic are optimized to reduce invalid Hash Buckets, redundant Motion operators, invalid InitPlans, and unnecessary execution engine fallbacks.

N/A

Bug Fix

Virtual Warehouse

• The VW Stale Cache issue is fixed. After a Beam Fragment is deleted or reclaimed, DADI stale caches in VW are promptly cleaned through WAL eviction, Pending Evict records, and an Epoch-aware cache invalidation mechanism. This prevents reading old physical paths, accidentally deleting new Fragment files, or causing data corruption after Fragment ID reuse.

• The issue where VW incorrectly reuses the distributed snapshot visibility hint from the primary node is fixed. VW now re-validates DLOG based on its own transaction snapshot to prevent data visibility anomalies.

• The issue where adbpg_inspector incorrectly associates node addresses of other VWs in multi-VW scenarios is fixed. Core file monitoring and cleanup now apply only to the current VW.

N/A

Beam storage engine

• The issue where Beam unique constraint checks do not release B-tree page Pins while waiting for transactions, which may cause concurrent VACUUM operations to be permanently blocked, is fixed.

• The issue where local dictionary offset reads fail under compiler optimizations and data read errors may occur when Beam Delta dictionary columns are not included in filtering is fixed.

N/A

Execution engine

Several execution engine stability and correctness issues are fixed, including abnormal InitPlan parameter delivery and errors, crashes, blocking, or non-terminating queries in Laser scenarios such as Hash Join, Nested Loop, index scan, Bitmap Scan, plan rescan, and SubPlan.

N/A

File cache

The issue where the DADI manual eviction interface does not reject negative Fragment IDs, which may generate invalid file paths or perform abnormal eviction operations, is fixed.

N/A

Category

Feature

Description

References

Optimization

Virtual Warehouse

Optimized the VW feedback mechanism. By default, hot_standby_feedback is enabled for new VWs, and related parameters are adjusted to improve stability in long transaction and concurrent query scenarios.

N/A

File cache

Beam file cache version update: CPU utilization is optimized.

N/A

Bug fix

Virtual Warehouse

The transaction visibility boundary of VW feedback is used to protect DLOG truncation, preventing DLOGs that are still referenced by VW snapshots from being truncated prematurely.

N/A

Storage engine Beam

• Fixed an issue where concurrent execution of Beam Garbage Collect and Fragment ID reuse might incorrectly delete new Fragment OSS files and cause data corruption.

• Fixed an issue where prepared statements executing INSERT on a Beam partitioned table might cause distributed deadlocks when VACUUM is concurrently executed on sub-partitions.

• Fixed an issue where Beam Fragments were not flushed in time in Repeatable Read isolation level and incremental materialized view delayed maintenance scenarios, causing subsequent statements to fail when reading POS Fork and triggering storage node exceptions.

• Optimized Beam storage exception handling to prevent POS Fork file exceptions or temporary DADI/OSS IO errors from being escalated to PANIC, improving node stability and troubleshooting capability.

N/A

Storage node

Fixed an issue where Page Store WAL Receiver might skip connection timeout checks and form zombie connections. Abnormal connections can now be automatically detected, reconnected, and continue receiving WAL.

N/A

Execution engine

Fixed an issue where ORCA might crash when extended statistics are combined with boolean expressions.

N/A

Category

Feature

Description

Related documentation

Added

Storage node

Adds the storage node migration feature.

None

Virtual Warehouse

Supports Read Committed isolation level consistency for VW. A bounded WAL replay waiting mechanism ensures cross-replica data visibility.

None

Optimized

Beam storage engine

• Optimizes Beam Flush performance by enabling ordered B-tree construction and sorted writes to improve write throughput.

• Supports Batch Split Compaction for Beam tables to reduce I/O overhead during compaction of large tables.

None

Security

Enhances server file security: restricts non-superusers from using file handling functions (enabled by default), and adds rds_superuser and default role privilege controls for all roles.

None

Bug fix

Beam storage engine

• Fixes a data corruption issue caused by Fragment Prune not cleaning up OSS residual files and GC relfilenode mismatch.

• Fixes a crash caused by rd_beam being null when pg_database_size is executed concurrently with ALTER TABLE SET ACCESS METHOD heap.

• Fixes the FM_RECYCLED Fragment auxiliary tuple freeze bug to prevent abnormal relfrozenxid advancement and data corruption after CLOG truncation.

• Fixes a crash when Beam Tablesample selects the last batch of data.

• Fixes the fragment_id epoch mismatch in the Laser dictionary pruning path.

None

Execution engine

• Fixes a SIGSEGV during InitPlan dispatch caused by a SerializedParams serialization bug.

• Fixes a crash caused by unprotected NULL return from cdbpath_create_motion_path in Lateral scenarios.

None

Virtual Warehouse

Fixes stability issues with VW horizontal scaling.

None

Category

Feature

Description

Related documentation

Added

Storage node

• Supports querying Page Store Timeline monitoring metrics through SQL functions.

• Supports updating Page Store Tenant configurations through SQL functions.

• Supports triggering Page Store forced compaction through SQL functions.

None

Optimized

Storage node

• The Compact operation is no longer performed during the checkpoint phase, reducing I/O overhead.

• Adds default write I/O throttling to prevent write overload.

None

Backup and restoration

Verifies the Page Store status before restoring a cluster to prevent Attach operations when storage nodes are abnormal.

None

Bug fix

GUC configuration

• Fixes an issue where 41 duplicate GUC registrations cause gpconfig errors and duplicate rows in pg_settings.

• Fixes an issue where appending parameters to postgresql.conf without a trailing newline causes configuration parsing failures.

None

Backup and restoration

Fixes a JSON parsing error that occurs when Garbage Collect is triggered asynchronously.

None

File cache

Updates the Beam file cache version to fix stability issues.

None

Storage node

Fixes a crash caused by the Prefetch Ring Buffer not being cleared after a Page Store connection is disconnected.

None

Category

Feature

Description

Related documentation

Bug fixes

Beam

• Fixed a metadata update error caused by Beam garbage collection (GC).

• Fixed a crash that occurred when running the beaminspect function on a Beam root partition table due to a null rd_beam value.

-

File cache

Updated the Beam file cache version to fix stability issues.

-

Category

Feature

Description

Related documentation

Bug fixes

virtual warehouse

Fixed a data visibility issue caused by the initialization of latestCompletedGxid during virtual warehouse (VW) restarts.

-

Beam

• Fixed an issue where the backend process failed to close all dadi files on exit, preventing potential file handle leaks.

• Fixed issues related to garbage collection.

• Fixed an issue where the Beam delta relcache was invalidated before the primary relation relcache, ensuring the correct cache invalidation order.

-

Category

Feature

Description

Related documentation

Optimizations

storage node

Optimized the Write-Ahead Logging (WAL) backup logic to significantly reduce Object Storage Service (OSS) storage consumption.

-

backup and recovery

Optimized the recovery point status query to significantly reduce the number of database connections used during backups.

-

local file cache

Enabled by default.

-

Bug fixes

Start/Stop

Fixed an issue where the SSL configuration file was lost when a cluster was paused.

Configure SSL encryption

Creation

Fixed an issue where the postgresql.conf file had an incorrect format after a cluster was created.

-

Category

Feature

Description

Related documentation

New feature

vector index

Added support for Hierarchical Navigable Small World (HNSW) vector indexes in a compute-storage separation architecture.

Vector analysis

Bug fixes

file cache

Updated the Beam file cache version to improve stability.

-

Start/Stop

Fixed an issue where configuration files were lost when a cluster was paused.

-

Beam

• Fixed an issue where VACUUM could not reclaim Beam auxiliary tables.

• Fixed an issue where Beam ZoneMap filtering did not correctly adapt to expressions.

-

Category

Feature

Description

Related documentation

Optimizations

Local disk scaling

The Beam file cache size is now set to 60% of the disk capacity upon disk expansion.

-

file cache

Updated the Beam file cache version to extend the retention period for its internal logs.

-

Bug fixes

Beam

Fixed an issue causing residual Beam temporary tables.

-

high availability

Fixed a startup failure caused by incorrect permissions for the data directory.

-

storage node

Fixed an issue where residual WAL partial files caused backup failures and led to full disks.

-