This topic covers engine release notes for AnalyticDB for PostgreSQL in elastic storage mode for 2024 and links to relevant documentation.
November 2024
V6.6.2.8
| Category |
Feature |
Description |
References |
| Optimized feature |
Executor |
Optimizes adaptive scan for bitmap index scans. Optimizes B-tree, Generalized Inverted Index (GIN), and Generalized Search Tree (GiST) indexes to prevent memory usage from growing exponentially with concurrency. |
None |
| Optimizer |
Optimizes nested loop joins by implementing column pruning. |
None |
| Fixed issue |
Stability |
Increases disk capacity to prevent disk exhaustion on the coordinator node. |
None |
| Planner |
Fixes an issue where the query planner incorrectly scans duplicate tables on the coordinator node in specific scenarios. |
None |
October 2024
V6.6.2.7
| Category |
Feature |
Description |
References |
| Fixed issue |
Planner |
Fixes LASJ_NOTIN issues. |
None |
| Resource group |
Fixes issues in resource group redzone scenarios. |
Use resource groups to perform workload management |
| Data link |
Fixes client SDK issues. |
None |
| Vector search |
Fixes incorrect distance records for batch queues in condition pushdown scenarios. |
Perform vector search |
| Optimizer |
Fixes Orca issues. |
None |
September 2024
V6.6.2.6
| Category |
Feature |
Description |
References |
| Optimized feature |
Scaling |
Improves write performance for tables being redistributed during scaling. |
Scale an instance |
| Vector search |
Adds support for the r7a and g7a models in the Vector Edition. Enables dynamic selection of single instruction, multiple data (SIMD) functions for vector distance calculations based on the model. |
Perform vector search |
| Physical backup |
Adds support for disk snapshot backup. |
None |
| Real-time materialized view |
Adds support for eventual consistency in real-time materialized views. |
Real-time materialized views |
| Persistent connection |
Enables non-engine processes to restart. Automatically retries queries or performs primary/secondary switchovers when queries fail due to core dumps. |
None |
| Fixed issue |
Optimizer |
Fixes an optimizer stability issue. |
Optimizers |
| Stability |
Fixes deadlocks when running the DROP OWNED BY or REASSIGN OWNED BY statement. Fixes an issue where the pg_rewind tool deletes pg_logs on secondary compute nodes. |
None |
August 2024
V6.6.2.5
| Category |
Feature |
Description |
References |
| New feature |
adbpg_background extension |
Adds the adbpg_background extension, which provides asynchronous job submission. The extension starts a background process to run statements submitted through a user-defined function (UDF) and records execution status and results in a history table for later reference. |
None |
| pg_buffercache extension |
Adds support for the pg_buffercache extension, which checks whether a table is cached. |
None |
| Optimized feature |
Scaling |
Optimizes write performance for tables being redistributed during scaling. |
Scale an instance |
| Vector search |
Adds a circuit breaker mechanism for queries that time out during vector search. |
Perform vector search |
| Storage engine |
Adds a Grand Unified Configuration (GUC) parameter to enable MetaScan for INSERT INTO SELECT operations. |
Storage engine |
| Fixed issue |
Vector search |
Fixes inconsistent sorting results for zero vectors when using cosine similarity. |
Perform vector search |
|
Fixes a trigger error in the auto_partition extension when running DROP TABLE on a table managed by auto_partition on a secondary coordinator node. Fixes an installation failure in the adbpg_transformer extension. Fixes data motions not working as expected during LEFT JOIN operations with a constant condition on the join key. Fixes an extension update failure that occurs when updating to a minor version with a large gap from the current version. Fixes relfrozenxid errors during VACUUM FREEZE on a secondary coordinator node in a multi-coordinator architecture. Fixes an insufficient memory reserved for statement error when using a resource group. |
None |
July 2024
V6.6.2.3
| Category |
Feature |
Description |
References |
| New feature |
Vector search |
Adds support for sparse vector data types. Adds support for creating Hierarchical Navigable Small World (HNSW) indexes on sparse vectors. Adds the ability to mark vector indexes built with memory mapping technology for deletion. Adds memory preloading for vector tables and vector indexes. Enables reading and writing vector tables during scaling. |
Perform vector search |
| Optimized feature |
Vector search |
Reduces the time to validate vector data size requirements for product quantization (PQ) training when enabling the PQ feature on vector indexes. Blocks PQ codebook training jobs during scaling to prevent conflicts with index rebuilding jobs. Optimizes inner product calculation functions using the latest SIMD computing library. Optimizes computational hotspots in vector indexes built with PostgreSQL shared buffer, improving creation, insertion, and query performance. Optimizes computational hotspots in vector indexes built with memory mapping technology, improving creation, insertion, and query performance. Updates default engine parameters for vector index creation and queries based on the latest best practices. Optimizes execution plans in brute-force search mode for hybrid query scenarios. |
Perform vector search |
| Fixed issue |
Vector search |
Fixes a decrease in recall rate when using PQ-enabled vector indexes for inner product distance search. Fixes inaccurate vector data size statistics during PQ codebook training on partitioned tables. Fixes memory leaks during PQ codebook training when data size becomes too large. Fixes vector index creation failures caused by PQ codebook training conflicts. Fixes a Block not found error when using PostgreSQL shared buffer vector indexes after many DELETE and UPDATE operations followed by VACUUM. Fixes PQ codebook training not exiting gracefully. Fixes PQ codebooks becoming invalid after ALTER TABLE RENAME on vector tables. Fixes memory mapping files not being deleted after removing vector indexes. Fixes a codebook version 1 not found error when using PQ codebook-enabled indexes on 5 million rows of vector data. |
Perform vector search |
| Fixed issue |
auto_partition extension |
Fixes a user execution permission issue in the auto_partition extension. Fixes tables not being automatically partitioned when deleted tables remain in the auto-partition configuration table. |
None |
March 2024
V6.6.2.1
| Category |
Feature |
Description |
References |
| New feature |
pg_jieba extension |
Adds the pg_jieba extension, which provides Chinese word segmentation for efficient Chinese full-text search. |
pg_jieba |
| btree_gist extension |
Adds the btree_gist extension, which uses a GiST index structure to simulate B-tree behavior. |
btree_gist |
| Optimized feature |
Resource group management |
Enables modifying CPU, memory, and concurrency parameters for a resource group to improve performance stability and compatibility. |
Use resource groups to perform workload management |
| Fixed issue |
Security |
Fixes the CVE-2022-1552 vulnerability. |
None |
| Stability |
Fixes deadlocks when running ALTER TABLE or ANALYZE on partitioned tables. Fixes out of memory (OOM) errors during bitmap index scans on partitioned tables. Fixes data truncation when a gpfdist external table with a custom format fails to identify the \0x00 string. |
None |
January 2024
V6.6.2.0
| Category |
Feature |
Description |
References |
| New feature |
Elastic scaling |
Adds elastic scaling for compute nodes. Tables remain readable, writable, and updatable without business interruption while nodes are being added. |
Change compute node configurations |
| Real-time materialized view |
Adds window function support for real-time materialized views. |
Real-time materialized views |
| Optimized feature |
SSL encryption |
Adds support for SSL 1.3. |
Configure SSL encryption |
| Ganos spatio-temporal engine |
Updates Ganos to V6.3. |
None |
| Fixed issue |
Stability |
Fixes a Host key verification failed error during data backup and restoration. Fixes index scan execution plans not being selected when creating an index on a partitioned table using the partition key. Fixes incorrect execution plans when the JOIN ON clause contains cross-level subqueries. Fixes performance degradation or errors when invoking functions on replicated tables from coordinator nodes. Fixes queries being suspended when LIMIT pushdown is used in non-vector search scenarios. Fixes system tables retaining residual data after running DROP FOREIGN SERVER on foreign tables. Fixes core dumps when using the NVL function of the tdfunc extension to replace null values. Fixes performance degradation caused by the Legacy query optimizer generating extra redistribution motions under LEFT JOIN and WHERE conditions. Fixes memory leaks when running the full-text search probe process. Fixes child partitioned tables inheriting only the primary key—not the unique index—when the parent partitioned table has both. Fixes an illegal rescan of motion node: invalid plan error on nested-loop index join plans. |
None |