Wal mechanism of MySQL

MySQL Of WAL Mechanism

WAL Its full name is Write-Ahead Logging, Pre write log system . It mainly refers to MySQL When performing a write operation, it is not immediately updated to the disk , But first record it in the log , Then update to disk at the right time . Logs are mainly divided into undo log、redo log、binlog.
When the contents of memory data page and disk data page are inconsistent , We call this memory page “ Dirty page ”. After memory data is written to disk , Memory and disk data page content is consistent , be called “ Clean pages ”.
MySQL Really use WAL The reason is that ： Disk writes are random IO, Comparative consumption performance , So if you write every update operation first log in , Then it becomes a sequential write operation , The actual update operation is performed by the background thread according to log Asynchronous write . So for client End , The delay is reduced . also , Since the sequential write probability is within a disk block , This produces IO The number of times is also greatly reduced . therefore WAL The core of is to transform random writing into sequential writing , Reduced client latency , Improved throughput .

Undo log（ Logical log ）

undo log revoke , Roll back . When a transaction is rolled back or the database crashes , You can use undo log The impact of undoing uncommitted transactions on the database .
undo log Produce and destroy ： Generated before the start of a transaction ; When a transaction is committed , It's not going to be deleted immediately undo log,innodb Will correspond the thing to undo log Put it in the delete list , Later, through the background thread purge thread To recycle .undo log Logical log , Record a process of change . For example, execute a delete sentence ,undo log Will record a insert.
undo log Storage ：undo log Manage and record in the way of segments . stay innodb The data file will contain a rollback segment Rollback segment , It contains 1024 individual undo log paragraph .
undo log The role of

1. Realize the atomicity of transactions
2. Realize multi version concurrency control （MVCC）: Before the transaction is committed ,undo log Saved uncommitted previous version data ,undo log The data in can be used as a snapshot of the old version of data for other concurrent transactions to read , It's equivalent to doing A backup .
   

Redo log

redo log Ensure the atomicity and persistence of transactions , yes InnoDB Engine specific , Records changes to data in transactions . One SQL modify , It will produce a set of redo log Information （ Including a lot of ）, This group is the writing redo log The smallest unit of . When the data is written to the disk redo log Information can be released .
Use redo log The benefits of

1. It guarantees the data persistence .
2. redo log Only changes to the data are recorded , The data is much smaller than a page of data , Greatly reduced IO frequency .
3. redo log Is written sequentially , The modified disk is written randomly .

Redo log working principle
redo log It's a product of the persistence of transactions . Prevent at the point of failure , There are dirty pages not written to the table IBD In file , It's restarting MySQL When , according to redo log To redo , So as to achieve the feature of persistence of transaction's non disk data . If you persist directly to disk , Each commit transaction is persisted to disk , This will increase IO frequency , Reduce efficiency . Persist to redo log And on the disk IO operation , But persistent to redo log Is continuous write , The disk is random , So we use redo log More efficient .

Redo Log Write mechanism
Redo log The contents of the file are written to the file in a sequential loop , When full, it goes back to the first file , Write overlay . There are two pointers inside .
write pos Record the current position , Move back as you write , Write to the end of the last file and go back to 0 The beginning of file No .
checkpoint Is the current location to erase , It's also going back and forth , Before erasing a record, update the record to a data file .
write pos and checkpoint The empty part in the middle can be used to record new operations . If the two coincide , The certificate is full , Cannot perform new update operation , You need to erase some records first .