AnalyticDB:2023

Improvements

Bug fixes

• Fixed a deadlock between ALTER TABLE ADD PARTITION and VACUUM FULL pg_index that could cause instance unavailability.

• Fixed a core dump in the tdfunc extension caused by improper NULL value handling in the NVL function.

• Fixed a memory leak in the execution engine's Dynamic Index Scan operator during partitioned table scans.

• Fixed incorrect results from a semi-join on a replicated table when using the Orca optimizer.

• Fixed a deadlock on tables with an ANN index during vector database scale-out.

• Fixed a core dump caused by the incremental sort feature.

• Fixed vector search queries hanging after a new column was added to the table.

• Fixed the illegal rescan of motion node: invalid plan error during a Motion Rescan in a Nested Loop Index Join plan.

• Fixed incorrect results from the Orca optimizer for queries involving a semi-join with a DISTINCT clause.

Bug fixes

• Fixed a resource queue holder leak caused by CREATE CURSOR WITH HOLD.

• Fixed a Merge Join plan with a partition selector that led to incorrect results or a core dump.

• Fixed intermittent out-of-memory (OOM) errors during bitmap index scans on partitioned tables.

• Fixed a file registration exception caused by premature release of WorkfileMgr.

• Fixed a deadlock on tables with an ANN index during vector database scale-out.

• Fixed abnormally high memory usage by the checkpointer process.

• Fixed a convert_EXPR_to_join error in the planner.

• Fixed a memory leak in the Global Deadlock Detector (GDD).

• Fixed queries being stuck or delayed when REFRESH MATERIALIZED VIEW was executed on non-materialized views.

• Fixed an unexpected primary coordinator node switchover caused by a prolonged Syncdatadir operation.

• Fixed an incorrect maximum value retrieved for a variable-length sort key when auto-sorting a single-row table.

• Fixed improper semaphore handling during codebook pre-training for a vector database.

• Fixed incorrect results from the Orca optimizer for queries involving a semi-join with a DISTINCT clause.

• Fixed table schema inconsistency when a constraint was deleted from a partitioned table but not from its child partitions.

New features

Vector search

• Supports the pgvector protocol, enabling seamless migration of workloads that use the pgvector extension to AnalyticDB for PostgreSQL vector databases.

• Supports multiple indexes on a single vector column. The system automatically selects the appropriate index based on the distance metric.

• Trains a codebook automatically when you create a vector index on a table that already contains data.

• Adds the following vector UDFs: vector_sum, vector_max, vector_min, vector_avg, vector_norm, vector_angle, and vector_dot.

For more information, see Vector search.

Other new features

Improvements

Bug fixes

• Fixed a core dump caused by the planner during subquery execution.

• Fixed data not being redistributed after a table was renamed during a scheduled instance scale-out.

• Fixed a core dump during an INSERT operation on a real-time materialized view created with GROUP BY and UNION ALL clauses.

• Fixed the cache lookup failed for extension 18713194 error when executing DROP TABLE.

• Fixed the FastANN vector extension failing to install in the pg_catalog schema.

• Fixed the reader could not find writer proc entry error occasionally occurring after parallel query was enabled.

• Fixed the planner failing to prune the default partition of a partitioned table.

• Fixed table schema inconsistency when a constraint was deleted from a partitioned table but not from its child partitions.

• Fixed abnormally high memory usage by the checkpointer process.

• Fixed a convert_EXPR_to_join error in the planner.

• Fixed a core dump on Jsonb data caused by an incorrect plan for a Lateral Join.

• Fixed residual temporary tables remaining on the coordinator node.

• Fixed the MERGE operation on a column-oriented table failing to retrieve the maximum value for a variable-length sort key when only one row existed.

• Fixed a resource queue holder leak caused by CREATE CURSOR WITH HOLD.

• Fixed the can not found block: (-1, -1, -1, -1) error when using the Metascan feature for column-oriented tables.

• Fixed the MIN/MAX subplan has unexpected flowtype: 351 error when generating a plan for a MIN or MAX operation on a system table.

New features

Improvements

Bug fixes

• Fixed Orca optimizer query failures for queries containing implicitly referenced columns.

• Fixed a core dump in a real-time materialized view when an OUTER JOIN with a single operator in the WHERE clause was executed.

• Fixed missing two-stage monitoring statistics for an OUTER JOIN with aggregation in real-time materialized views, and self-join calculation/application information being counted only once.

• Fixed incorrect results caused by the planner converting an EXISTS clause to an INNER JOIN.

• Fixed a resource queue holder leak caused by CREATE CURSOR WITH HOLD.

• Fixed the could not retrieve tle for sort-from-groupcols error when generating a plan for an aggregate operation on a table with a primary key.

New features

Improvements

New features

Bug fixes

• Fixed the pg_cron extension failing to allow non-superuser accounts to create scheduled tasks for a specific user.

• Fixed a compatibility issue with Oracle string concatenation.

• Fixed the could not retrieve tle for sort-from-groupcols error when the optimizer generated a plan for an aggregate operation on a table with a primary key.

• Fixed a Merge Join plan with a partition selector that led to incorrect results or a core dump.

• Fixed a core dump or panic caused by the optimizer when handling a Lateral Join with a LIMIT clause.

New features

Improvements

Bug fixes

• Fixed scale-in failures when a CTAS plan was run in the Orca optimizer.

• Fixed vertical scaling failures.

• Fixed a core dump caused by sorted files during automatic background merging in a scale-out process.

• Fixed instance scale-in failures when vector search was enabled.

• Fixed the cache lookup failed for extension ... error when the kernel executed DROP TABLE.

• Fixed the Orca optimizer not supporting aggregate functions on multiple DISTINCT objects.

• Fixed the reader could not find writer proc entry error caused by the parallel query feature.

• Fixed the can not found block: (-1, -1, -1, -1) error when using the Metascan feature for column-oriented tables.

• Fixed inconsistent storage options (row-oriented or column-oriented) between a partitioned table and its child partitions.

• Fixed an inconsistency in how the mmap engine and shared_buffer engine handle Product Quantization (PQ) for vector data.

• Fixed mmap index files in the vector database not being fully recycled, leading to file leaks.

• Fixed the vector algorithm library of the mmap engine failing to achieve a 100% recall rate by adjusting parameters when the initial recall rate was very low.

• Fixed a core dump when executing UPDATE ... RETURNING ... on an append-optimized (AO) materialized view if the RETURNING clause contained Ctid.

• Fixed missing dependencies on other coordinator nodes when automatically creating nested real-time materialized views in a multi-coordinator architecture.

New features

Improvements

Bug fixes

• Fixed the kernel vulnerability CVE-2023-2454.

• Fixed UDFs used by the PL/Java extension causing residual pg processes.

• Fixed an error that occurred when detecting and adding a secondary coordinator node during data recovery.

• Fixed autovacuum jobs not being scheduled promptly in Fastworker mode.

• Fixed Automerge holding a lock for an extended period, causing other queries to wait.

• Fixed inaccurate display of scaling progress.

• Fixed query resource skew caused by an interruption during full data recovery.

• Fixed the transaction is already committed error during recovery of a distributed transaction on a compute node.

• Fixed a core dump when accessing AOCO tables with parallel query enabled after an instance scale-out.

• Fixed pg_rewind replication of WAL (Xlog) being affected by the statement_timeout parameter.

• Fixed excessively long Orca optimizer planning time for tables with many subpartitions.

New features

Improvements

Bug fixes

• Fixed a potential memory leak when using INSERT ON CONFLICT UPDATE or COPY ON CONFLICT UPDATE.

• Fixed a missing distributed commit log (clog) file for a column-oriented auxiliary table.

• Fixed scaling failures when the database contained partitioned external tables.

• Fixed scaling failures when the database contained materialized views with no data.

• Fixed inaccurate display of scaling progress.

• Fixed query errors on replicated tables in certain scenarios.

• Fixed a core dump when running ANALYZE on an external table with a very small amount of data.

• Fixed the ERROR: could not find hash function for hash operator 58393736 error during SQL plan generation.

• Fixed an error when querying the expansion_progress view after enabling the Orca optimizer.

• Fixed deadlocks and data inconsistencies caused by concurrent delete tasks on a real-time materialized view in a non-aggregation scenario.

New features

Improvements

Bug fixes

• Fixed a potential integer overflow in the intelligent diagnosis feature by changing the table size column data type from integer to bigint.

• Fixed the tuple already updated by self error when updating a distribution key column in a partitioned table.

• Fixed coordinator nodes failing to reclaim writer gangs, which could lead to high memory usage and out-of-memory (OOM) errors on compute nodes in high-concurrency scenarios.

• Fixed incorrect INDEX SCAN results on column-oriented tables.

• Fixed an error caused by using an incorrect snapshot when adding an auxiliary table to an AOCO table.

• Fixed index bloat and non-repeatable reads on append-optimized (AO) and AOCO tables.

• Fixed a Nested Loop Join failing to correctly select a materialization node.