tencent cloud

Overview

If multi-row large transaction is updated by a single statement in row format, an event will be generated for each row. As a result, a large number of binlogs are created, and APPLY operations in the replica database become slower during replication, causing replication delays.
After analyzing and optimizing the large transaction replication scenarios, the Tencent Cloud kernel team developed the large transaction replication optimization feature. With this feature, large transactions are automatically identified and binlogs are converted from row format into statement format, thus reducing the quantity of binlogs and increasing the replication performance.

Supported Versions

• Kernel version: MySQL 5.6 20210630 and above.
• Kernel version: MySQL 5.7 20200630 and above.
• Kernel version: MySQL 8.0 20200830 and above.

Use Cases

• This feature accelerates large transaction replay for tables without a primary key in row format. It can be enabled if you are sure that the delay is caused by slow replay due to the lack of primary key.
• This feature aims to solve slow replication when there are large transactions in row format.

Performance Data

The replication time is reduced by 85% for UPDATE operations and about 30% for INSERT operations.

Instructions

The large transaction replication optimization feature judges whether a transaction is large based on the historical execution statistics of the SQL statement. If a transaction is identified as large and optimizable, its isolation level will be automatically upgraded to repeatable read (RR), and the binlogs will be stored in statement format to reduce the time of executing the large transaction in the replica database. Here:

• cdb_optimize_large_trans_binlog is the switch of this feature.
• cdb_sql_statistics is the switch of SQL statement execution statistics collection.
• cdb_optimize_large_trans_binlog_last_affected_rows_threshold and cdb_optimize_large_trans_binlog_aver_affected_rows_threshold are the thresholds for judging a large transaction.
• cdb_sql_statistics_info_threshold is the number of legacy data entries retained in the memory.

To better monitor the transaction execution, the CDB_SQL_STATISTICS table is added in the information_schema database for you to query the statistics of the current transaction.

New parameters

Note：

Currently, you cannot directly modify the values of the above parameters. If needed, submit a ticket for assistance.

Newly added information_schema.CDB_SQL_STATISTICS table