December 2023
V6.3.10.26
Category | Feature | Description | References |
Optimized feature | Optimizer | When multiple tables are joined, the amount of time consumed and memory usage are not significantly increased if the value of the join_collapse_limit parameter is increased. The join_collapse_limit parameter specifies the maximum number of tables that can be merged by the join operation. | None |
Optimizer | The amount of time consumed by executing the EXPLAIN statement is reduced when a partitioned table that has a large number of partitions is joined with a MaxCompute external table. | None |
Fixed issue | The following issue is fixed: When the ALTER TABLE ADD PARTITION and VACUUM FULL statements are executed on pg_index, deadlocks occur. As a result, the instance becomes unavailable. |
The following issue is fixed: Core dumps occur when the NVL function improperly handles NULL values in the tdfunc extension. |
The following issue is fixed: Memory leaks occur at the execution layer when the execution engine handles the Dynamic Index Scan operator for partitioned table scanning. |
The following issue is fixed: Semi-join results are incorrect in specific scenarios when the Orca optimizer generates an execution plan for a semi-join operation on replicated tables. |
The following issue is fixed: Deadlocks occur on the table that is used to create approximate nearest neighbor (ANN) indexes when you scale out a vector database. |
The following issue is fixed: The incremental sort feature causes core dumps. |
The following issue is fixed: Query statements executed by the vector search feature are suspended when new columns are added. |
The following issue is fixed: The illegal rescan of motion node: invalid plan error occurs when the Motion Rescan operator is executed in the Nestloop Index Join plan. |
The following issue is fixed: The results are incorrect in specific scenarios when the Orac optimizer generates an execution plan that involves a semi-join operation and the DISTINCT clause. |
November 2023
V6.3.10.24
Category | Description |
Fixed issue | The following issue is fixed: The resource queue holders are leaked when you execute the CREATE CURSOR WITH HOLD statement. |
The following issue is fixed: Incorrect query results or core dumps occur when the optimizer attempts to generate an execution plan that involves a merge join by using a partition selector. |
The following issue is fixed: An out of memory (OOM) issue occasionally occurs when bitmap index scans are performed on partitioned tables. |
The following issue is fixed: Early release of WorkfileMgr causes a file registration exception. |
The following issue is fixed: Deadlocks occur on the table that has ANN indexes when you scale out a vector database. |
The following issue is fixed: The memory usage of the checkpointer process is abnormally high. |
The following issue is fixed: An error occurs when the planner optimizer attempts to generate an execution plan that involves convert_EXPR_to_join. |
The following issue is fixed: Memory leaks occur when the memory usage of the GDD process is unexpectedly high. |
The following issue is fixed: Queries are stuck or delayed when the REFRESH MATERIALIZED VIEW statement is used to refresh non-materialized views. |
The following issue is fixed: An abnormal switchover to the primary server occurs due to a long Syncdatadir time. |
The following issue is fixed: The maximum value of a variable-length field is incorrect when automatic sorting is performed on a table that has only one row of data. |
The following issue is fixed: Abnormal semaphores are captured during codebook pre-training of vector databases. |
The following issue is fixed: Query results may be incorrect in specific scenarios when the Orac optimizer attempts to generate an execution plan that involves a semi-join operation and the DISTINCT clause. |
The following issue is fixed: If you delete a constraint from a partitioned table but not from partitions, table schemas are inconsistent. |
V6.6.0.0
Category | Feature | Description | References |
New feature | Vector search | The pgvector extension is supported. You can seamlessly migrate business that uses the pgvector extension as the SDK to AnalyticDB for PostgreSQL vector databases. Multiple indexes can be created for a single vector column, and indexes can be automatically selected based on the distance measurement. Product Quantization (PQ) codebooks can be trained before indexes are created. If a vector table has data, a codebook is trained before a vector index is created for the vector table. Vector-related user-defined functions (UDFs) are supported, such as vector_sum, vector_max, vector_min, vector_avg, vector_norm, vector_angle, and vector_dot.
| Perform vector search |
pldbgapi extension | The pldbgapi extension and the function debugging feature of PL/pgSQL are supported. | pldbgapi |
Dynamic data masking | Different masking functions can be used to mask sensitive columns and display raw data in a masked form. This protects sensitive data and improves data security. | Dynamic data masking |
Real-time materialized view | A sample dataset is supported for real-time data warehouses to test real-time data warehouses that are built based on real-time materialized views. | None |
Optimized feature | Optimizer | The memory usage of coordinator nodes and compute nodes is significantly reduced in specific scenarios. | None |
Real-time materialized view | The performance in complex query and outer join scenarios is improved. | Real-time materialized views |
Vector search | The AVX512F instruction set can be used for vector distance calculation. | None |
The sorting distance can be obtained from a vector index to prevent repeated calculation. |
Original vectors can be stored in indexes to prevent multiple lookups during index creation and query. |
Normalized vectors can be stored in indexes to optimize the performance of calculating the cosine similarity. |
Resource group | The underlying support for using Cgroups in proxy mode is added. | None |
Fixed issue | The following issue is fixed: Core dumps occur when the planner optimizer generates an execution plan for subqueries. The following issue is fixed: Data is not redistributed after you rename a table during an instance scale-out. The following issue is fixed: If you create a real-time materialized view by using a query that contains the GROUP BY and UNIONALL clauses, core dumps occur when you execute an INSERT statement. The following issue is fixed: The cache lookup failed for extension 18713194 error occurs when you execute a DROP TABLE statement. The following issue is fixed: The FastANN extension cannot be installed in the pg_catalog schema. The following issue is fixed: After you enable the parallel query feature for an instance, the reader could not find writer proc entry error occasionally occurs. The following issue is fixed: The planner optimizer cannot prune the default partitions of partitioned tables. The following issue is fixed: If you delete a constraint from a partitioned table but not from partitions, table schemas are inconsistent. The following issue is fixed: The memory usage of the checkpointer process is abnormally high. The following issue is fixed: An error occurs when the planner optimizer attempts to generate an execution plan that involves convert_EXPR_to_join. The following issue is fixed: When the optimizer attempts to generate an execution plan that involves a lateral join, the generated plan is incorrect and a core dump occurs on JSONB fields. The following issue is fixed: Residual temporary tables exist on coordinator nodes. The following issue is fixed: The maximum value of a variable-length field is incorrect when you execute the MERGE statement on a column-oriented table that has only one row of data. The following issue is fixed: The resource queue holders are leaked when you execute the CREATE CURSOR WITH HOLD statement. The following issue is fixed: The can not found block: (-1, -1, -1, -1) error occurs when MetaScan is used for column-oriented tables. The following issue is fixed: The MIN/MAX subplan has unexpected flowtype: 351 error occurs when the MIN/MAX subplan is generated for system tables.
|
September 2023
V6.3.11.3
Category | Feature | Description | References |
New feature | Real-time materialized view | Real-time materialized views can be monitored. | Real-time materialized views |
Cross-database query | A foreign data wrapper (FDW) is supported to allow you to perform associated queries across AnalyticDB for PostgreSQL databases. This ensures data timeliness and reduces data redundancy. | Cross-database query |
Cross-instance query | Real-time interactive queries between AnalyticDB for PostgreSQL instances are supported and can be used for federated analytics. | Cross-instance query |
Optimized feature | Optimizer | The IN list optimization feature is enabled by default. | None |
Compatibility with Oracle | The UNKNOWN data type can be converted. | None |
Fixed issue | The following issue is fixed: A query fails if the query uses the Orca optimizer and contains columns that are not explicitly specified. The following issue is fixed: In real-time materialized views, core dumps occur if single operators exist in the WHERE clause of the OUTER JOIN operation. The following issue is fixed: In real-time materialized views, the two-stage monitoring statistics are missing when you combine outer join with an aggregate operation, and the self join calc/apply information is counted only once. The following issue is fixed: The planner optimizer incorrectly converts the EXISTS clause into an INNER JOIN operation. The following issue is fixed: The resource queue holders are leaked when you execute the CREATE CURSOR WITH HOLD statement. The following issue is fixed: The could not retrieve tle for sort-from-groupcols error occurs when the optimizer attempts to generate an execution plan that involves an aggregate operation on a table that has a primary key.
|
August 2023
V6.3.11.2
Category | Feature | Description | References |
New feature | Vector search | Functions related to cosine similarity are added. Vector indexes can be created for text data and vector data in parallel. The hybrid search feature is enhanced to support semi-structured data indexes. Vector search and full-text search can be used to implement two-way retrieval.
| Vector database overview |
Optimizer | | Dynamic Join Filter |
Real-time materialized view | Append-optimized column-oriented (AOCO) tables can serve as base tables of real-time materialized views. | Real-time materialized views |
Tiered storage of hot and cold data | Data that is infrequently accessed can be stored in Object Storage Service (OSS) to reduce storage costs. | Tiered storage of hot and cold data |
Optimized feature | Vector search | | None |
Optimizer | The planner optimizes the nested loop join cost and index-related SQL plans to achieve better execution performance. The memory of partitioned tables is used and managed in a more optimal manner to prevent high memory usage in large partition scenarios. The Orca optimizer supports dynamic partition pruning. The Orca optimizer improves the performance of generating execution plans for partitioned tables. The Orca optimizer improves the accuracy of estimating the cardinality of time-type columns.
| None |
Real-time materialized view | The performance of complex queries and outer joins is improved. | None |
V6.3.10.20
Category | Feature | Description | References |
New feature | Vector search | The inner product distance and cosine distance are supported for vector databases. | Vector database overview |
Fixed issue | The following issue is fixed: The pg_cron extension cannot use a non-superuser account to create scheduled tasks for a specific user. The compatibility issue of Oracle string concatenation is fixed. The following issue is fixed: The could not retrieve tle for sort-from-groupcols error occurs when the optimizer attempts to generate an execution plan that involves an aggregate operation on a table that has a primary key. The following issue is fixed: Incorrect query results or core dumps occur when the optimizer attempts to generate an execution plan that involves a merge join by using a partition selector. The following issue is fixed: Core dumps or panic situations occur when the optimizer attempts to generate an execution plan that involves a lateral join with a LIMIT clause.
|
July 2023
V6.3.10.16
Category | Feature | Description | References |
New feature | Real-time materialized view | Real-time materialized views can rewrite queries that involve the GROUP BY clause with a rollup or cube aggregation. | Real-time materialized views |
Missing GROUP BY columns are automatically supplemented for the columns in the query result when a real-time materialized view is created. |
Optimized feature | Real-time materialized view | The excessive length of nested materialized view names that is caused by the excessive length of view names or subquery aliases is optimized. | None |
Optimizer | The extended amount of time for the Orca optimizer to plan tables that contain a large number of subpartitions is reduced. | None |
Orafce extension | Orafce 3.25 is backward compatible. | Supported extensions |
Query Cache | The query cache provides more scenarios to cache query results. | Supported extensions |
Fixed issue | The following issue is fixed: Scale-in failures occur when a CTAS plan is run in the Orca optimizer. The issue that causes upgrade or downgrade failures is fixed. The following issue is fixed: Core dumps occur due to files that are sorted in automatic merge operations during a scaling process. The following issue is fixed: Scale-in failures occur when vector search is enabled. The following issue is fixed: The cache lookup failed for extension ... error occurs when the DROP TABLE statement is executed. The issue that the Orca optimizer does not support aggregate operations on multiple distinct objects is fixed. The following issue is fixed: The reader could not find writer proc entry error occurs when SQL statements are executed in parallel. The following issue is fixed: The can not found block: (-1, -1, -1, -1) error occurs when MetaScan is used for column-oriented tables. The issue that partitions inherit different storage options from the partitioned table is fixed. The issue that the MMAP and shared_buffer engines use different mechanisms to handle vector data when interfacing with PostgreSQL is fixed. The issue that MMAP index files in vector databases are incompletely recycled is fixed. The following issue is fixed: The vector algorithm library of the MMAP engine cannot increase the recall rate from a low value to 100% after parameters are modified. The following issue is fixed: Core dumps occur when the UPDATE ... RETURNING ... statement is executed on append-only (AO) materialized views and the RETURNING clause contains the ctid column. The following issue is fixed: When nested real-time materialized views are automatically created in the multi-coordinator architecture, dependencies on other coordinator nodes are missing.
|
May 2023
V6.3.10.14
Category | Feature | Description | References |
New feature | Real-time materialized view | The FROM clause is supported for real-time materialized views. | Real-time materialized views |
Optimized feature | GanosBase engine | The GanosBase engine is upgraded to version 5.5. | Ganos overview |
Fixed issue | The CVE-2023-2454 vulnerability is fixed. We recommend that you update the minor version of your AnalyticDB for PostgreSQL V6.0 instance in elastic storage mode to V6.3.10.14 or later. The issue that UDFs used by the PL/Java extension may cause residual PostgreSQL processes is fixed. The issue that causes errors when the system adds a secondary coordinator node during data recovery is fixed. The issue that causes scheduling failures of auto-vacuum in fast worker mode is fixed. The issue that causes waiting queries due to locks occupied by auto-merge is fixed. The issue that causes incorrect scaling progress display is fixed. The issue that causes query resource skew due to the interrupted full data recovery process is fixed. The issue that causes the transaction is already committed error during the recovery of distributed transactions on compute nodes is fixed. The issue that causes core dumps when you enable parallel query of AOCO tables after instance scaling is fixed. The issue that the copy of XLOG files by pg_rewind is subject to the statement_timeout parameter is fixed. The issue that causes an extended amount of time for the Orca optimizer to plan tables that contain a large number of subpartitions is fixed.
|
March 2023
V6.3.10.12
Category | Feature | Description | References |
New feature | Real-time materialized view | Multiple columns can be specified in the PARTITION BY clause for real-time materialized views. | Real-time materialized views |
Optimized feature | Orafce extension | The Orafce extension is upgraded to version 3.25.0. | Use the Orafce extension |
Fixed issue | The issue that causes memory leaks during the execution of INSERT ON CONFLICT UPDATE or COPY ON CONFLICT UPDATE is fixed. The issue that causes failures to find distributed clogs in the column-oriented auxiliary table is fixed. The issue that causes scaling failures due to partitioned external tables is fixed. The issue that causes scaling failures due to zero-data materialized views is fixed. The issue that causes incorrect scaling progress display is fixed. The issue that causes query errors on replicated tables in specific scenarios is fixed. The issue that causes core dumps during the ANALYZE operation of small-volume external tables is fixed. The issue that causes ERROR: could not find hash function for hash operator 58393736 during the generation of execution plans is fixed. The issue that causes Orca optimizer-based query errors on the expansion_progress view is fixed. The issue that causes deadlocks and data inconsistency is fixed. The errors occurred when concurrent deletions were performed on real-time materialized views in aggregation-excluded scenarios.
|
January 2023
V6.3.10.9
Category | Feature | Description | References |
New feature | Partitioned table | The following features are supported for partitioned tables in AnalyticDB for PostgreSQL: Subpartitions can be renamed. The names of partitions can be specified when partitions are created. The DETACH and ATTACH statements are supported for partitioned tables
| Define table partitioning |
Change to the number of compute nodes | The number of compute nodes can be reduced for AnalyticDB for PostgreSQL V6.0 instances in elastic storage mode. | Change compute node configurations |
Sample dataset | A sample dataset is provided to help you use or test AnalyticDB for PostgreSQL instances. | Manage the sample dataset |
Real-time materialized view | The following features are supported for real-time materialized views: UNION ALL syntax FROM syntax
| Real-time materialized views |
Optimized feature | Real-time materialized view | The performance of OUTER JOIN operations is optimized for real-time materialized views. | Real-time materialized views |
Column-oriented partitioned table | The execution of ALTER TABLE ADD COLUMN statements is accelerated for column-oriented partitioned tables. | None |
GanosBase spatio-temporal engine | The GanosBase engine is upgraded to version 5.1. | GanosBase overview |
Diagnostics and optimization | Statistics can be used to speed up diagnostics on data bloats, data skews, and indexes. | Data bloat, data skew, and index statistics |
SQL optimizer | The default value of the join_collapse_limit parameter is set to 10. This helps reduce the amount of memory that is consumed when multiple tables are joined and the amount of time that is required to join multiple tables. | None |
Fixed issue | The data type of the table size column is changed from INTEGER to BIGINT. This way, complete numbers are displayed in the tables that are generated from intelligent diagnostics. The issue that causes the tuple already updated by self error is fixed. The error occurred when the distribution key columns of a partitioned table were updated. The issue that causes OOM errors on compute nodes in high concurrency scenarios is fixed. OOM errors occurred because the coordinator node does not reclaim writer gangs. The issue that causes index scan errors on column-oriented tables is fixed. The issue that causes auxiliary table errors on AOCO tables due to incorrect snapshots is fixed. The issue that causes index bloats and repeated reading failures on AO tables or AOCO tables is fixed. The issue that causes materialized node errors on nested-loop joins is fixed.
|