Read mysql45 the next day

mysql The medium log system has the following functions: server Layer. binlog journal （ Archive log ）, also InnoDB In the engine layer redolog journal （ Redo log ）;binlog Is stored in sql Modified logical log .

show binary logs Commands can be viewed binlog Log files , I am here dbeaver Use this command to view the display 860 and 861

show binglog events in 'mysql-bin.000861’ You can view the contents of specific log files , But I read it for a while and didn't understand what was in it .

redolog The physical log is stored , That is, which rows change the value . I still don't know what the log content is , Just remember that one is a physical log and the other is a logical log .

show variables like ‘innodb_log%’

You can see redo Log related configuration ：

Here, the log size log is configured as 524288000,524288000/1024/1024 Is almost 500MB, One group is 2 File , So the size that this group of files can record is 1000MB about ;

In the figure write pos What is recorded is where the log is currently written , Move back as you write , When moved to ib_logfile_3 At the end of , It will start from ib_logfile_0 restart ;
check point It is the place that needs to be erased at present . Every time an update statement is executed ,InnoDB It doesn't have to be done immediately , Instead, store the records in redolog In the Journal , Then update the memory ,
At this time, it is finished , In fact, it has not been updated to the disk . If every time an update statement comes , Write to the disk immediately , Repetitive operations can also be costly , So first record it in redolog In the Journal ,
Then execute when the database is idle , Such execution efficiency will be relatively high , It's kind of similar IO Buffer in .

When there are data records that need to be changed ,InnoDB The engine will first write the change value in redolog In the Journal , And then update it to memory , When the database is idle, update it to the disk , This process is called WAL,
The full name is Write-Ahead Loggin The key point is to write a log first And then update it to disk .redolog The log makes the database restart abnormally , No records submitted before will be lost , This can become Crash-Safe.

redo Logs are fixed size , It's recycled , So it will keep overwriting the previous logs ,binlog You can add log files by appending , Therefore, the previous log files will not be overwritten .
General process of updating records ：

1. actuator ： take id=2 The record of num Liega 2
2. Engine layer ： The engine interface is based on id=2 Find that record , If the record is in memory, it will be returned directly , If not, read from the disk to the memory and return （ In this place, I am curious that the record is not in the memory, so I look for it from the disk , So what happens
Invalidate previous records in memory ？ There are too many things to be stored in the memory, which will delete the things that first existed in the memory ？）
3. actuator ： After getting this line of value , to num+2 Then return to the engine layer
4. Engine layer ： take num+2 The value of is written to memory , At the same time, write this value to redolog In the log , here redolog be in prepare state , Then inform the actuator that it has finished , You can commit a transaction .
5. The actuator generates this update operation binglog journal , Then write to binglog In the disk file of
6. The executor calls the commit transaction interface of the engine to commit transactions
7. When the engine interface receives the request to commit a transaction, it will send the following message redolog Change the status of the log to the submission status , Update complete .

You can see that redolog There are two log operations , The first write is prepare state , The second, commit state . This is the two-stage submission .

Two phase commit can prevent abnormal restart of the database , Give Way redolog and binglog Inconsistent operation records appear , That is, two-phase submission is to make redolog and binglog Try to keep the log records consistent .

innodb_flush_log_at_trx_commit This parameter is set to 1 Every time said redolog When logging, it will be persisted to the disk , This prevents the loss of operation records ;
sync_binlog Parameter is 1 Every time said binlog Log operations are persisted to disk files .