2025

Category

Feature

Description

References

New

Logical backup

Adapts logical backup to the DBS storage pool API.

None

Long-term memory

Optimizes how long-term memory saves operation history. Adds an API for users to view history.

None

Optimized

High availability

Optimizes the cluster restart time.

None

Bug fixes

High availability

Fixes an issue where an ECS instance breakdown may cause statements, such as queries and COPY, to hang.

None

Iceberg

Upgrades the log4j-core component version.

None

Vector search

Fixes an out-of-memory (OOM) issue caused by an excessively long query in pgsearch.

None

Storage engine

Fixes stability issues with Beam.

None

Real-time materialized view

Fixes an issue of slow updates when writing to an empty base table of an incremental materialized view (IMV).

None

GraphRAG

Fixes stability issues with GraphRAG.

None

Scaling

Fixes stability issues with scaling.

None

CDC

Fixes stability issues with CDC.

None

Category

Feature

Description

References

Bug fixes

Startup

Fixes a startup failure issue after a breakdown in specific scenarios.

None

Scaling

Fixes an issue where real-time materialized views fail during scale-out.

None

Beam

Fixes several stability issues.

None

pgsearch

Fixes an OOM issue caused by an excessively long query.

None

CDC

Fixes several stability issues related to data synchronization.

None

Optimized

Restart

Optimizes the cluster restart time to resolve slow restarts with a high number of connections.

None

Category

Feature

Description

References

New

Long-term memory

Adds a parameter to customize the number of memories for search. Supports automatic memory categorization and custom category labels.

Advanced features of long-term memory for AnalyticDB for PostgreSQL

OSS FDW

OSS FDW supports COPY.

None

Extension

Allows rds_superusers to create the fixnumeric extension.

fixnumeric

Optimized

Vector engine

Hardens dependencies for pgsearch.

None

GraphRAG

• Supports adjusting the connection pool size.

• Adds loading of 'age' to shared_preload_libraries.

None

Vector engine

Optimizes vector engine parameters.

None

Security

Security hardening.

None

Bug fixes

CDC

Fixes a memory leak issue during insert, update, and delete (IUD) operations when a Beam table is not a subscribed table.

None

Version upgrade

Fixes an issue where the read-only state cannot be released due to a script execution failure during a minor version upgrade.

None

AOCS

Fixes an issue with ALTER COLUMN TYPE for AOCS tables.

None

Real-time materialized view

Fixes stability issues with IMV.

None

Optimizer

Fixes stability issues with the optimizer.

None

Storage engine

Fixes stability issues with Beam.

None

Category

Feature

Description

References

Bug fixes

AOCS

Fixes stability issues with ALTER COLUMN TYPE for AOCS tables.

None

Beam

Fixes an issue where delta indexes are repeatedly written in specific scenarios.

None

Optimizer

Fixes several stability issues during dictionary processing.

None

Partitioned table

Fixes an issue of high memory usage when importing dictionaries for partitioned tables.

None

CDC

Fixes several stability issues related to data synchronization.

None

Optimized

Security

Security hardening.

None

Category

Feature

Description

References

New

Long-term memory

Adds support for custom filter conditions during long-term memory retrieval.

Advanced features of long-term memory for AnalyticDB for PostgreSQL

Multi-zone

Supports forced failover between the primary and secondary zones.

None

Optimized

Real-time materialized view

• Optimizes MV & IMV REPLACE to support operations such as rebuilding sort keys.

• Materialized views support ALTER ACCESS METHOD.

None

Image size reduction

Reduces the image size for AnalyticDB for PostgreSQL V7.0.

None

Storage engine

• Improves real-time write performance for Beam.

• Optimizes high memory usage when importing dictionaries for partitioned tables.

None

Bug fixes

Vector engine

• Fixes stability issues with the vector engine.

• Fixes an issue where mirror redo cannot find a dictionary because the pgsearch dictionary contains segid.

None

Storage engine

Fixes a performance issue when checking column-level reloptions for Beam partitioned tables.

None

Real-time materialized view

Fixes stability issues.

None

CDC

Fixes stability issues with CDC.

None

Category

Feature

Description

References

Bug fixes

Optimizer

• Fixes a column pruning issue in "unionall" caused by optimizer_transform_or_join_to_join_under_unionall in CTEs.

• Fixes an issue where the ORCA optimizer duplicates data and broadcasts the right table when it processes a RIGHT SEMI ANTI JOIN ... NOT IN operation and all associated columns are non-null.

• Fixes an error when the parameter for text <> all(text[]) is "{NULL}".

• Fixes numerous dictionary-related issues across various scenarios.

None

Upgrades and downgrades

Fixes an issue where the read-only state is not fully released after a minor version upgrade.

None

Storage engine

• Fixes an issue of improper handling of predicate indexes during presorting and index writing in the Beam engine.

• Fixes an issue where Zonemap cannot be applied to timestamptz and timestamp filter conditions.

• Fixes issues such as sort key memory leaks.

None

Optimized

Image

Reduces the image size for AnalyticDB for PostgreSQL, decreasing disk space usage.

None

GraphRAG

Implements numerous optimizations for the GraphRAG extension and enables its installation by default.

None

auto-pipeload

Installs the auto-pipeload extension by default.

None

Category

Feature

Description

References

Bug fixes

CDC

Fixes stability issues.

None

ANALYZE

Fixes an issue where the calculation method for the number of sampled rows for AO table statistics might miss new data.

None

Optimized

Real-time materialized view

• Supports forcing creation with Beam storage.

• Supports modifying the storage method.

None

Executor

Optimizes support for Beam tables.

None

Long-term memory

Supports adding custom filter conditions during long-term memory retrieval.

None

Category

Feature

Description

References

New

Automate the import of unstructured data using Auto-PipeLoad

AnalyticDB for PostgreSQL introduces the Auto-PipeLoad extension to seamlessly connect Object Storage Service (OSS) with the database and automate the extract, transform, and load (ETL) process for unstructured data. This extension is designed for modern AI applications. It natively supports text extraction, intelligent chunking, and vectorization for unstructured data, such as .pdf, .docx, and .txt files. The extension efficiently transforms raw data into structured or vectorized formats for analysis, which greatly simplifies the data pipeline for AI applications.

Automate unstructured data ingestion with Auto-PipeLoad

Category

Feature

Description

References

Bug fixes

Stability

Fixes a "could not find Shared Local Snapshot!" error that might occur during cursor queries.

None

Storage

Fixes an issue of duplicate primary keys caused by updates during the Beam sorting process.

None

Executor

Fixes a coredump issue in dictionary encoding that occurred when the dictionary maintenance length exceeded the USSR maximum value during dictionary initialization.

None

CDC

Resolves issues caused by excessively frequent updates to the `adbpg_replication_lsn_map` table in high-concurrency scenarios. These issues include severe table bloat and degraded replication performance.

None

Optimizer

Fixes an issue of incorrect results caused by nested loops not preserving order in some scenarios.

None

Category

Feature

Description

References

New

High availability

Supports different shared_buffers configurations for primary and secondary instances.

None

GraphRAG

Supports importing decision trees.

None

Vector search

GIN supports stable early stopping.

None

Bug fixes

Storage engine

Fixes stability issues with Beam.

None

Extension management

Fixes packaging for rdkit.

None

Transaction management

Fixes stability issues, such as failed deadlock detection, caused by the conversion from int8 to int4 in gp_dist_wait_status.

None

Category

Feature

Description

References

New

Build long-term memory applications

Large Language Models (LLMs) can remember conversations to a certain extent. However, limitations in their attention mechanism can cause them to miss information, which degrades the quality of generated content. Expanding the context window of an LLM may still not capture all key information, and it also increases costs and reduces model efficiency. Long-term memory technology extracts and persistently stores key information from conversations, such as user characteristics and preferences. This allows AI applications to reference this stored memory in future interactions to improve conversation quality and user experience.

Build long-term memory applications based on AnalyticDB for PostgreSQL

Category

Feature

Description

References

New

Connection pool

Supports PgBouncer.

Server-side connection pool (PgBouncer)

Appearance

Parquet foreign tables support MAP and LIST types.

None

Memory management

vmem supports startup memory configuration.

None

Long-term memory

Supports the long-term memory extension.

Build long-term memory applications based on AnalyticDB for PostgreSQL

GraphRAG

Supports the GraphRAG extension.

Use the GraphRAG service

Optimized

Tiered storage

Supports obtaining correct statistics for OSS ORC foreign tables and allows auto analyze.

None

Storage engine

Implements lock-free optimization for Beam Optimize and optimization for REINDEX.

None

Vector engine

Optimizes parameters.

None

Extension management

Prohibits non-superusers from creating extensions using commands.

None

Bug fixes

Vector search

Fixes an issue where vector indexes do not support partial indexes.

None

Optimizer

Fixes stability issues with dictionary encoding during dictionary initialization.

None

Storage engine

Fixes stability issues with Beam.

None

CDC

Fixes stability issues with CDC.

None

Transaction management

Ensures that V7.x is compatible with the usage of DISCARD ALL to correctly delete temporary tables.

None

Category

Feature

Description

References

New

Access Iceberg data

AnalyticDB for PostgreSQL supports accessing Iceberg data through OSS foreign tables. You can directly access Iceberg data or perform operations such as association analysis between local tables and Iceberg tables. The feature also supports data sharing with Iceberg-compatible systems, such as Hive, Spark, StarRocks, and cloud-based EMR.

Use an OSS foreign table to access Iceberg data

Category

Feature

Description

References

New

Graph analysis engine

AnalyticDB for PostgreSQL integrates a graph analysis engine as an extension. It supports the Cypher query language to efficiently query and operate on graph data. The graph analysis engine excels at processing highly interconnected datasets and is widely used in scenarios such as social networking, fraud detection, recommendation engines, knowledge graphs, and network or IT O&M.

Graph analysis engine

Category

Feature

Description

References

New

fixnumeric extension

AnalyticDB for PostgreSQL introduces the FIXNUMERIC type. This type combines fixed precision with underlying integer storage. In compute-intensive scenarios, its performance is significantly better than the native NUMERIC or DECIMAL types. The performance of operations such as aggregations and basic arithmetic is improved several times over, while lossless precision is maintained within a specified range.

fixnumeric

Category

Feature

Description

References

New

pgml extension

• The pgml extension supports fine-tuning for text embedding models.

• The pgml extension supports image embedding.

In-Database AI/ML

OSS foreign table

• Supports parallel reading of data in the ORC format.

• Supports Auto Analyze (automatic statistics collection) for foreign tables. This feature collects more accurate statistics from OSS ORC foreign tables to optimize query performance.

• Use an OSS foreign table to import data from OSS at high speed

• Automatically collect statistics

Flink CDC

Flink change data capture (CDC) supports subscribing to real-time materialized views and the strong consistency feature for DTS synchronization.

Use Flink CDC to subscribe to full and incremental data in real time

Tiered storage

Tiered storage supports adding and dropping columns.

None

Optimized

OSS External Tables

Optimizes the link between the OSS foreign data wrapper (FDW) and ORC files to enhance write support for bytea data.

Use an OSS foreign table to import data from OSS at high speed

Storage engine

The default storage for Beam is set to auto.

Self-developed high-performance storage engine Beam

ANALYZE

Optimizes the memory usage of ANALYZE when it merges table statistics.

Manually collect statistics

Bug fixes

Upgrade

Fixes several stability issues related to minor version upgrades.

None

ANALYZE

Fixes an issue where an error might occur when you run ANALYZE on the primary table of a partitioned foreign table or a table that uses tiered storage.

Manually collect statistics

Execution engine

• Fixes an issue in queries that contain LATERAL JOIN where the Gather Motion operation did not correctly pass the sort key, causing results to be sorted incorrectly.

• Fixes an issue where the LEAST and GREATEST functions for the DATE type might return incorrect calculation results when both ORCA and LASER are enabled.

None

Storage engine

• Fixes an issue where duplicate data appeared in Beam tables with primary keys.

• Fixes an issue where ALTER TABLE SET (querycache_enabled = ON/OFF) did not take effect on Beam tables.

Self-developed high-performance storage engine Beam

Optimizer

• Fixes an issue where Chinese alias strings might be incorrectly displayed as Unknown when the ORCA optimizer is enabled.

• Fixes an issue where the ORCA optimizer might cause a program to crash (coredump) when processing tables with combined statistics due to specific formats of Common Table Expression (CTE) or ROLLUP operations. This improves stability and compatibility in complex query scenarios.

• Fixes several stability issues with the ORCA optimizer.

• Fixes an issue where the could not find pathkey item to sort error was reported in some scenarios because the sort basis (PathKey) could not be found.

None

pg_search

Fixes an issue where the pgsearch extension might return incorrect search results when processing TIMESTAMPTZ fields.

BM25 high-performance full-text search

Category

Feature

Description

References

New

Real-time data subscription with Flink CDC

AnalyticDB for PostgreSQL provides a self-developed CDC connector. This connector uses the logical replication feature of PostgreSQL to subscribe to full and incremental data and can be seamlessly integrated with Flink.

Use Flink CDC to subscribe to full and incremental data in real time

Optimized

Storage engine

For AnalyticDB for PostgreSQL V7.0 instances with version 7.1.1.4 or later, the default storage engine is Beam. Upgrading the kernel version of existing instances does not change their default storage format.

Self-developed high-performance storage engine Beam

Category

Feature

Description

References

New

Storage engine

• The row lock feature for Beam supports table-level options.

• The Beam storage engine supports sort key filtering for types such as VARCHAR.

Self-developed high-performance storage engine Beam

Optimized

Real-time materialized view

Supports creating real-time materialized views on Beam tables.

Real-time materialized view

Bug fixes

Vector engine

• Fixes a memory overflow issue caused by loading multi-version codebooks.

• Fixed a Parallel Worker conflict that occurred when multiple vector indexes were built in parallel.

Guide to the vector search engine

Storage engine

Fixes an error in the Beam storage engine that occurs during ALTER TYPE operations involving USING helper tables. This issue occurred when converting to the same data type.

Self-developed high-performance storage engine Beam

Category

Feature

Description

References

New

High-performance full-text search

Custom tokenizer dictionaries

You can add proper nouns, industry terms, or even new buzzwords to a dictionary based on your industry or specific application scenario. This allows for text segmentation that better meets your needs, which improves tokenization accuracy and search efficiency.

Configure a custom tokenizer dictionary

Stop word dictionaries

Configuring stop words can improve search efficiency, enhance search accuracy, and reduce storage usage. You can customize a dedicated stop word dictionary based on your industry or specific application scenario.

Configure a stopword dictionary

The pgsearch.field() and pgsearch.tokenizer() functions are added for creating BM25 indexes.

• pgsearch.field(): Defines how a field is indexed.

• pgsearch.tokenizer(): Defines the tokenizer type, the dictionary to use, and other tokenizer settings.

BM25 high-performance full-text search

The hmm and search parameters are added to the built-in jieba tokenizer for high-performance full-text search.

• hmm: When set to true, the tokenizer uses a Hidden Markov Model (HMM) to process out-of-vocabulary words. HMM-assisted tokenization uses contextual information to determine word boundaries, which improves tokenization accuracy.

• search: When set to true, the text is tokenized at the finest granularity. This further splits long words to increase phrase matching.

BM25 high-performance full-text search

Category

Feature

Description

References

Optimized

Storage engine

• Optimizes row-level locking at the table level in the Beam storage engine.

• Optimizes writes when an Optimize operation is performed on a Beam index.

Self-developed high-performance storage engine Beam

Real-time materialized view

Optimizes the handling of mutable functions and supports time-related functions such as now() and current_date().

Real-time materialized view

Bug fixes

Storage engine

• Fixes an error that occurred when modifying the schema of a Beam table that used a USING helper table.

• ZoneMaps in Beam now support data types such as VARCHAR.

Self-developed high-performance storage engine Beam

Search

Fixes an issue where the pgsearch extension returned incorrect search results for Timestampz.

BM25 high-performance full-text search

Planner

Fixes an issue with the Planner where if an execution plan contained an InitPlan node with a SubqueryScan node above it, the target list would be empty. This caused queries to return the correct number of rows, but the value of each row was NULL.

None

Stability

Fixes an issue where Chinese table headers in query results were replaced with UNKNOWN after some instances were upgraded.

None

Category

Feature

Description

References

Optimized

Storage engine

Optimizes row-level locking at the table level in the Beam storage engine.

Self-developed high-performance storage engine Beam

Real-time materialized view

Optimizes the handling of mutable functions and supports time-related functions such as now() and current_date().

Real-time materialized view

Bug fixes

Storage engine

ZoneMaps in Beam now support data types such as VARCHAR.

Self-developed high-performance storage engine Beam

Optimizer

Fixes an issue where ORCA might output all system columns of a table when processing multiple SQL clauses, such as COUNT DISTINCT. This caused an error for Beam and AOCS tables because system columns such as xmin were output.

None

Planner

Fixes an issue with the Planner where if an execution plan contained an InitPlan node with a SubqueryScan node above it, the target list would be empty. This caused queries to return the correct number of rows, but the value of each row was NULL.

None