Automatic Failover With MariaDB Monitor
Automatic Failover With MariaDB Monitor
The is not only capable
of monitoring the state of a MariaDB master-slave cluster but is also
capable of performing failover and switchover. In addition, in some
circumstances it is capable of rejoining a master that has gone down and
later reappears.
Note that the failover (and switchover and rejoin) functionality is only
supported in conjunction with GTID-based replication and initially only
for simple topologies, that is, 1 master and several slaves.
The failover, switchover and rejoin functionality are inherent parts of
the MariaDB Monitor, but neither automatic failover nor automatic rejoin
are enabled by default.
The following examples have been written with the assumption that there
are four servers - server1, server2, server3 and server4 - of
which server1 is the initial master and the other servers are slaves.
In addition there is a monitor called TheMonitor that monitors those
servers.
Somewhat simplified, the MaxScale configuration file would look like:
Manual Failover
If everything is in order, the state of the cluster will look something
like this:
If the master now for any reason goes down, then the cluster state will
look like this:
Note that the status for server1 is Down.
Since failover is by default not enabled, the failover mechanism must be
invoked manually:
There are quite a few arguments, so let's look at each one separately_call command indicates that it is a module command that is to be_
&#xNAN;invoked, mariadbmon indicates the module whose command we want to invoke (that
is the MariaDB Monitor),failover is the command we want to invoke, and TheMonitor is the first and only argument to that command, the name of
the monitor as specified in the configuration file.
The MariaDB Monitor will now autonomously deduce which slave is the most
appropriate one to be promoted to master, promote it to master and modify
the other slaves accordingly.
If we now check the cluster state we will see that one of the remaining
slaves has been made into master.
If server1 now reappears, it will not be rejoined to the cluster, as
shown by the following output:
Had auto_rejoin=true been specified in the monitor section, then an
attempt to rejoin server1 would have been made.
In MaxScale 2.2.1, rejoining cannot be initiated manually, but in a
subsequent version a command to that effect will be provided.
Automatic Failover
To enable automatic failover, simply add auto_failover=true to the
monitor section in the configuration file.
When everything is running fine, the cluster state looks like follows:
If server1 now goes down, failover will automatically be performed and
an existing slave promoted to new master.
If you are continuously monitoring the server states, you may notice for a
brief period that the state of server1 is Down and the state ofserver2 is still Slave, Running.
To enable automatic rejoin, simply add auto_rejoin=true to the
monitor secion in the configuration file.
When automatic rejoin is enabled, the MariaDB Monitor will attempt to
rejoin a failed master as a slave, if it reappears.
When everything is running fine, the cluster state looks like follows:
Assuming auto_failover=true has been specified in the configuration
file, when server1 goes down for some reason, failover will be performed
and we end up with the following cluster state:
If server1 now reappears, the MariaDB Monitor will detect that and
attempt to rejoin the old master as a slave.
Whether rejoining will succeed depends upon the actual state of the old
master. For instance, if the old master was modified and the changes had
not been replicated to the new master, before the old master went down,
then automatic rejoin will not be possible.
If rejoining can be performed, then the cluster state will end up looking
like:
Switchover is for cases when you explicitly want to move the master
role from one server to another.
If we continue from the cluster state at the end of the previous example
and want to make server1 master again, then we must issue the following
command:
There are quite a few arguments, so let's look at each one separately_call command indicates that it is a module command that is to be_
&#xNAN;invoked, mariadbmon indicates the module whose command we want to invoke,switchover is the command we want to invoke, and TheMonitor is the first argument to the command, the name of the monitor
as specified in the configuration file,server1 is the second argument to the command, the name of the server we
&#xNAN;want to make into master, and server2 is the third argument to the command, the name of the currentmaster.
If the command executes successfully, we will end up with the following
cluster state:
CC BY-SA / Gnu FDL