Stop slave is stuck -- the event of the transaction is not copied completely

In the article stop slave Get stuck , A preliminary introduction to stop slave And some reasons .

And articles stop slave Stuck simulation – Big business scenario in , Introduced the large transaction scenario ,stop slave The simulation .

This article introduces another case ： The events of a transaction are not completely copied to the slave database .

In the master-slave architecture , After updating the data on the main database , Will write binlog, And copy to the slave Library .

In transactional storage engines , Such as innoDB, Data updates are in transactions .

A transaction usually has many event form , for example begin, commit,query, write_rows wait .

Commit a transaction on the master database , Event write binlog, Then copy to the slave Library , Start playback from the Library binlog. Playback from the library binlog, Not all of a transaction event After copying them , Just started playback , But one side I/O Thread receive binlog journal , On one side SQL Thread application binlog journal .

such , There may be a problem , If a transaction event More , Only a part event Copy to from library , At this time, the main library fails or The network is disconnected , The remaining... Cannot be received from the library event, Lead to SQL The thread has been waiting for the rest event.

At this time , Execute... On the slave library stop slave,SQL Thread will wait , Until the waiting time is exceeded （ Default 60s）, Then roll back the transaction .

Basic troubleshooting ideas ：

1. Check the major tasks being implemented event（ Big business monitoring in advance , You can collect what is being executed event）.
2. View the error log from the Library , Determine the log location of the master database that is being copied from the slave database . Use this log point , In the old main warehouse binlog Find the corresponding event in .
3. The analysis reason ：
   • If what is being executed is begin event , And the event corresponding to the replication site from the library is not commit, Then we can get , Lead to stop slave The reason for the timeout is that the event of the transaction is not received completely .
   • Ongoing SQL yes Normal update SQL, Lead to stop slave The reason for the timeout may be that the transaction is relatively large , The actual implementation is time-consuming , It may also be that the event is not received completely （ Because it is impossible to determine whether the executing transaction and the replicating transaction are the same ）.