AliSQL-8.0.44-2

AliSQL 8.0.44-2

Prebuilt AliSQL binaries and Docker images for this release. / 本次发布提供 AliSQL 预编译二进制包和 Docker 镜像。

Binary packages / 二进制包

Platform	Package
Linux x86_64 (glibc 2.17)	alisql-8.0.44-2-linux-glibc2.17-x86_64.tar.xz
Linux aarch64 (glibc 2.17)	alisql-8.0.44-2-linux-glibc2.17-aarch64.tar.xz

Docker image / Docker 镜像

The multi-architecture image supports linux/amd64 and linux/arm64. It is published after the binary packages are available.

• Image: songhuaxiong/alisql:8.0.44-2

• Build: GitHub Actions

  docker pull songhuaxiong/alisql:8.0.44-2

English changelog

Standalone English changelog | 中文变更说明

AliSQL 8.0.44 is based on MySQL 8.0.44. This release updates the DuckDB integration and adds native vector indexes, Binlog Cache Free Flush, Persist Binlog Into Redo V2, and Native Flashback. Most new execution paths are disabled by default. The Flashback query gate is enabled, but snapshot generation and undo retention are not.

Overview

Area	Feature	Default
Analytics	DuckDB storage engine enhancements	duckdb_mode=NONE
Vector search	Native VECTOR type and HNSW index	vidx_disabled=ON
Large transactions	Binlog Cache Free Flush	OFF
Binlog durability	Persist Binlog Into Redo V2	persist_binlog_to_redo=OFF
Historical queries	Native Flashback AS OF TIMESTAMP	Snapshot task OFF

1. DuckDB Storage Engine Enhancements

AliSQL integrates DuckDB as a native analytical storage engine behind the MySQL protocol and SQL interface. This update upgrades DuckDB to v1.4.4 and extends compatibility, DDL, replication, and resource governance.

New Features

• Optional SQL normalization for cross-database queries, temporal expressions, more MySQL functions, and prepared-statement reprepare.
• Generated-column table conversion, Latin1, BLOB/VARCHAR comparison, and improved BINARY, VARBINARY, and BIT handling.
• Separate per-query thread limits for user and replication workloads.
• Optional high-precision calculation paths.
• INSERT ... SELECT acceleration for DuckDB-to-DuckDB Copy DDL.
• CPU governance for DuckDB replication work.

Reliability and Replication Fixes

• Fixed crashes in JSON paths, CONCAT, UPDATE/DELETE subqueries, partition Copy DDL, and read-only prepared-statement cursors.
• Fixed orphan-table cleanup and shutdown during startup conversion.
• Multi-Trx Batch requires replica_parallel_workers=0; change duckdb_multi_trx_in_batch only while replication is stopped.
• Fixed Exec_Master_Log_Pos, invalid gno=0, partial-batch rollback, and GTID release handling.
• Improved log_replica_updates=OFF, empty-transaction, temporary-table, cross-database, and idempotent-replay behavior.

New Variables

Variable	Scope	Default	Range / Description
duckdb_sql_normalization	Global, Session	OFF	Rewrite supported MySQL SQL for DuckDB
duckdb_max_threads_per_query	Global, Session	1000000	1 to 4611686018427387904; user query thread limit
duckdb_max_threads_per_query_rpl	Global	1000000	Same range; replication query thread limit
duckdb_psmt_cursor_send_extra_eof	Global, Session	ON	Extra EOF for empty cursor results
duckdb_prefer_high_precision	Global, Session	OFF	Prefer high-precision calculations
duckdb_convert_tables_with_generated_columns	Global	ON	Allow generated-column conversion
duckdb_copy_data_between_tables_use_ins_sel	Global, Session	OFF	Use INSERT ... SELECT during Copy DDL
ignore_index_hint_error	Global	OFF	Warn instead of failing for missing hinted indexes

2. Native Vector Index (VIDX)

VIDX adds VECTOR(N), scalar vector functions, and HNSW approximate nearest-neighbor search. It supports Euclidean and cosine distance with up to 16,383 dimensions.

SET GLOBAL vidx_disabled = OFF;
SET SESSION transaction_isolation = 'READ-COMMITTED';

CREATE TABLE embeddings (
    id BIGINT PRIMARY KEY,
    embedding VECTOR(3)
) ENGINE=InnoDB;

INSERT INTO embeddings VALUES
    (1, VEC_FROMTEXT('[0.1,0.2,0.3]')),
    (2, VEC_FROMTEXT('[0.2,0.1,0.4]'));

CREATE VECTOR INDEX embedding_hnsw
    ON embeddings(embedding) M=6 DISTANCE=COSINE;

SELECT id,
       VEC_DISTANCE_COSINE(
           embedding, VEC_FROMTEXT('[0.1,0.2,0.3]')
       ) AS distance
FROM embeddings
ORDER BY distance
LIMIT 10;

Variable	Scope	Default	Range
vidx_disabled	Global	ON	ON, OFF
vidx_default_distance	Global, Session	EUCLIDEAN	EUCLIDEAN, COSINE
vidx_hnsw_default_m	Global, Session	6	3 to 200
vidx_hnsw_ef_search	Global, Session	20	1 to 10000
vidx_hnsw_cache_size	Global	16 MiB	1 MiB to ULLONG_MAX bytes

Vector indexes require InnoDB and READ COMMITTED. They do not support ALGORITHM=INPLACE or invisible indexes. Nullable vectors are allowed but are omitted from HNSW; scalar distance returns NULL for them. Randomized and heuristic HNSW construction means replicas are not guaranteed to have byte-identical graph topology.

3. Large TX Optimization: Binlog Cache Free Flush

Free Flush finalizes a spilled transaction binlog-cache file and renames it as the next binlog instead of copying the large cache again during commit. This reduces commit I/O amplification.

Free Flush supports large InnoDB transactions when the registered 2PC participants are the binlog and InnoDB. The current DuckDB integration registers another 2PC participant, so builds that include DuckDB use normal binlog group commit. Large-transaction optimization for DuckDB will be added in the next release.

Variable	Scope	Default	Range
binlog_cache_free_flush	Global, dynamic	OFF	ON, OFF
binlog_cache_free_flush_limit_size	Global, dynamic	256 MiB	10 MiB to ULLONG_MAX bytes

The cache must exceed the threshold, be spilled and finalized, contain no statement-cache event or incident, remain unencrypted, and not modify mysql.gtid_executed. InnoDB must be the only registered non-binlog 2PC engine. If any check fails, the transaction uses normal binlog group commit.

4. Persist Binlog Into Redo V2

Eligible binlog events are persisted in InnoDB redo during commit. Background flush and sync threads update the binlog file; crash recovery reconstructs a missing binlog tail from redo.

The implementation guards require sync_binlog=1, binlog_order_commits=OFF, and replica_preserve_commit_order=OFF. Keep innodb_flush_log_at_trx_commit=1 for the intended full-durability baseline; it is a recommendation rather than an enablement guard.

Variable	Scope	Default	Range / Unit
persist_binlog_to_redo	Global, dynamic	OFF	ON, OFF
persist_binlog_to_redo_size_limit	Global, dynamic	1 MiB	0 to 10 MiB
sync_binlog_interval	Global, dynamic	50000 us	1 to 100000000 us
binlog_buffer_size	Global, read-only	20 MiB	20 MiB to 1 GiB
wait_binlog_flush	Global, dynamic	ON	ON, OFF
binlog_group_delay	Global, dynamic	100 ns	0 to 1000000000 ns
binlog_group_delay_running_threads	Global, dynamic	100	0 to 100000

Only one active non-binlog 2PC engine is supported. Enabling is rejected while DuckDB mode is active. Transactions using nontransactional tables, the statement binlog cache, or a cache above the size limit use normal group commit.

5. Native Flashback

Native Flashback reconstructs an InnoDB historical read view from retained undo and exposes it through AS OF TIMESTAMP.

The background thread is created if innodb_rds_flashback_task_enabled=ON or innodb_undo_retention>0 at server startup, but actual snapshot generation requires both the task to be enabled and undo retention to be nonzero:

[mysqld]
innodb_rds_flashback_task_enabled=ON
innodb_undo_retention=3600
innodb_rds_flashback_interval=1

SELECT *
FROM orders AS OF TIMESTAMP DATE_SUB(NOW(), INTERVAL 5 MINUTE);

The read view is selected from retained snapshot history and may not match the requested wall-clock time exactly. innodb_rds_flashback_allow_gap bounds the accepted difference. For recovery, write historical rows to a separate table, validate them, and stop application traffic before replacing the original table.

Variable	Default	Range / Unit
innodb_rds_flashback_task_enabled	OFF	Boolean
innodb_rds_flashback_task_stop_all	OFF	Boolean
innodb_rds_flashback_interval	1	1 to 86400 seconds
innodb_rds_flashback_enabled	ON	Boolean
innodb_rds_flashback_allow_gap	30	0 to UINT32_MAX minutes
innodb_rds_flashback_print_warning	ON	Boolean
innodb_undo_retention	0	0 to UINT32_MAX seconds
innodb_undo_space_supremum_size	10240	0 to UINT32_MAX MiB
innodb_undo_space_reserved_size	0	0 to UINT32_MAX MiB

Administrative procedures are dbms_admin.analyze_flashback_snapshots(), dbms_admin.show_flashback_snapshots(timestamp, order, limit), and dbms_admin.del_flashback_snapshots(timestamp, order). Analyze returns the minimum and maximum snapshot timestamps plus total undo-space usage reported by InnoDB in MiB. Show and delete require SUPER.

Flashback supports SELECT references to InnoDB base tables. Temporary tables, views, locking reads, and historical reads across incompatible DDL are not supported. The effective history window is constrained by both retention time and undo-space pressure.

Alibaba Cloud RDS MySQL

RDS MySQL provides managed versions of these features, together with product topology, data synchronization, backup, monitoring, and support. Supported versions, parameter names, defaults, and ranges can differ from this source tree; use the product documentation for an RDS instance.

Capability	Official documentation
DuckDB analytical instances	English / 中文
Vector storage	English / 中文
Native Flashback	English / 中文
Binlog Cache Free Flush	English / 中文
Binlog in Redo	English / 中文