It's not uncommon to promote a server from slave to master. One of the key things to protect your data integrity is to make sure that the promoted slave is permanently disconnected from its old master. If not, it may get writes from the old master, which can cause all kinds of data corruption. MySQL provides the handy RESET SLAVE command. But as we'll see, its behavior has changed along with the MySQL versions and it's easy to shoot yourself in the foot if you use it incorrectly. So how do you safely disconnect a replication slave?
In short
- For MySQL 5.0 and 5.1, run
STOP SLAVE,CHANGE MASTER TO MASTER_HOST=''and thenRESET SLAVE. - For MySQL 5.5 and 5.6, run
STOP SLAVEand thenRESET SLAVE ALL. - For all versions, ban
master-user,master-hostandmaster-passwordsettings in my.cnf, this may cause huge problems (it's anyway no longer supported from MySQL 5.5).
If you want to know more details, please read on!
MySQL 5.0/5.1
First let's consider MySQL 5.0 and 5.1. RESET SLAVE will remove the master.info and relay-log.info files as well as all the relay log files. This looks great, but does it ensure the replica is disconnected from its master?
Let's try:
mysql> stop slave; Query OK, 0 rows affected (0.00 sec) mysql> reset slave; Query OK, 0 rows affected (0.03 sec) mysql> show slave status\G *************************** 1. row *************************** Slave_IO_State: Master_Host: 127.0.0.1 Master_User: test Master_Port: 3306 Connect_Retry: 60 Master_Log_File: Read_Master_Log_Pos: 4 Relay_Log_File: mysql_sandbox35302-relay-bin.000001 Relay_Log_Pos: 4 Relay_Master_Log_File: Slave_IO_Running: No Slave_SQL_Running: No [...]
This is not expected: instead of removing all settings, some of them are reset to default values. This means that if you run START SLAVE (or if it's done automatically, for instance when restarting the server without the skip-slave-start option), replication may start again. But as the master position has been deleted, replication will restart at the beginning of the first available binary log, which is very likely to corrupt your data by reexecuting some queries.
Here's a trick to make RESET SLAVE work as expected: use CHANGE MASTER TO MASTER_HOST='':
mysql> stop slave; Query OK, 0 rows affected (0.00 sec) mysql> change master to master_host=''; Query OK, 0 rows affected (0.02 sec) mysql> reset slave; Query OK, 0 rows affected (0.04 sec) mysql> show slave status\G Empty set (0.00 sec) mysql> start slave; ERROR 1200 (HY000): The server is not configured as slave; fix in config file or with CHANGE MASTER TO
Much better! If we try to restart replication, it fails. However, I don't like the error message, specifically the 'fix in config file' part. What happens if we specify the master-user, master-password, master-host and master-port in the my.cnf file?
# cat my.cnf [...] master-user=rsandbox master-password=rsandbox master-host=127.0.0.1 master-port=35301 [...]
Let's disconnect the slave:
mysql> stop slave; Query OK, 0 rows affected (0.00 sec) mysql> change master to master_host=''; Query OK, 0 rows affected (0.03 sec) mysql> reset slave; Query OK, 0 rows affected (0.03 sec) mysql> show slave status\G *************************** 1. row *************************** Slave_IO_State: Master_Host: 127.0.0.1 Master_User: rsandbox Master_Port: 35301 Connect_Retry: 60 Master_Log_File: Read_Master_Log_Pos: 4 Relay_Log_File: mysql_sandbox35302-relay-bin.000001 Relay_Log_Pos: 4 Relay_Master_Log_File: Slave_IO_Running: No Slave_SQL_Running: No [...]
Connection settings are automatically restored, which makes disconnecting the replica impossible. And again, if you restart replication, it will read events from the first available binary log file on the master, which is probably not what you want. So never set master-xxx variables in my.cnf!
From MySQL 5.5
Starting with MySQL 5.5, the situation has slightly changed. First the master-xxx variables are no longer supported, which is a great improvement. But theRESET SLAVE statement also behaves differently:
mysql> stop slave; Query OK, 0 rows affected (0,01 sec) mysql > reset slave; Query OK, 0 rows affected (0,11 sec) mysql> show slave status\G *************************** 1. row *************************** Slave_IO_State: Master_Host: 127.0.0.1 Master_User: rsandbox Master_Port: 18675 Connect_Retry: 60 Master_Log_File: Read_Master_Log_Pos: 4 Relay_Log_File: mysql_sandbox18676-relay-bin.000001 Relay_Log_Pos: 4 [...]
As stated in the documentation, the connection parameters are still held in memory. In any case, you will be able to restart replication, but again as no replication coordinate is specified, replication will start at the beginning of the first available binary log file, with all the nasty consequences we can imagine.
Even worse, the CHANGE MASTER TO MASTER_HOST='' trick no longer works:
mysql> stop slave; Query OK, 0 rows affected (0,01 sec) mysql> change master to master_host=''; ERROR 1210 (HY000): Incorrect arguments to MASTER_HOST
Fortunately, the documentation also specifies that we can use RESET SLAVE ALL to remove all replication-related configuration:
mysql> stop slave; Query OK, 0 rows affected (0,00 sec) mysql> reset slave all; Query OK, 0 rows affected (0,04 sec) mysql> show slave status\G Empty set (0,00 sec)
Very good! The command does work as expected without any additional tricks. As soon as you are aware of the difference between RESET SLAVE andRESET SLAVE ALL, disconnecting a replication slave is much easier with MySQL 5.5+.
The post RESET SLAVE vs. RESET SLAVE ALL: Disconnecting a replication slave is easier with MySQL 5.5+ appeared first on MySQL Performance Blog.
No comments:
Post a Comment