Mysql 45 study notes (23) How does MYSQL ensure that data is not lost

One.binlog writing mechanism

Execution logic:

• Transaction execution, first write the logto binlog cache,
• write the binlog cache when the transaction is committedinto the binlog file

binlog cache

• The system binlog cache allocates a piece of memory, one per thread
• Parameterbinlog_cache_size is used to control the size of the memory occupied by the binlog cache in a single thread
• If exceeds the size specified by this parameter, it will be temporarily saved to disk

When the transaction commits:

• The executor writes the complete transaction in the binlog cache into the binlog and clears the binlog cache

2. Redo log writing mechanism

Three states of redo log:

• Exist in redo log buffer, physically in MySQL process memory
• Writing on disk (write), but not persistent (fsync), physically in the page cache of the file system
• Persist to disk, corresponding to hard disk

There is a problem:

The log is written to the redo log buffer very quickly, and the write to the page cache is similar, but the speed of persistence to the disk is very slow

innodb_flush_log_at_trx_commit parameter

0	Indicates that the redo log is left in the redo log buffer every time a transaction is submitted
1	Indicates that the redo log will be persisted directly to disk every time a transaction is committed
2	Indicates that only the redo log is written to the page cache every time the transaction is committed

• InnoDB has a background thread that will send redo the log in the log buffer, call write to the page cache of the file system with write, then call fsync to persist to disk
• The redo log of the intermediate process of transaction execution is also directly written in the redo log buffer
• May also be persisted to disk by background threads.
• A redo log of a uncommitted transaction may have been persisted to disk.

Three. Two scenarios where the uncommitted redo log is written to disk:

1. When the space occupied by the redo log buffer is about to reach half of innodb_log_buffer_size, the background thread will actively write to the disk
2. When a parallel transaction is committed, the redo log buffer of this transaction is persisted to disk by the way

Two-phase commit:

In terms of timing, redo logprepare first, then write binlog and then redo log commit

Group commit mechanism