关于mysql redo log的问题。

Question

According to the crash recovery logic of mysql server, we only need to write/sync prepared transaction before writing to binlog. And currently transactions are not grouped during innodb prepare stage.

I changed the related code described as bellow:

don’t sync redo log during innodb prepare stage if binlog enabled.

during flush stage of group commit (In function process_flush_stage_queue)
a) collect group commit queue

b) call ha_flush_logs to sync redo log (make sure all redo log of prepared transaction in current
queue were synced before writing to binlog)
--innobase_flush_logs have to be changed.

c) flush thread cache to binlog file for every thd in queue

Answer 1

重做日志，存储已提交事务的数据，保证在出现机器故障时对已提交事务数据不丢失从而持久化，Mysql操作数据都是会将数据所在的那一页数据加载到缓存池中的，操作也是操作缓存池中的数据，操作完后不会立即刷新到磁盘，而是会将数据写入redo log中，至于缓存池中的数据时要依靠一个随机IO的线程去将缓存池数据持久化，但在缓存池中数据未持久化这个过程中，机器发生故障，就会导致缓存池数据丢失，所以才引入了read log。