[CentOS] Clustering MySQL

J Potter jpotter-centos at codepuppy.com
Wed Dec 12 14:50:44 UTC 2007


> After all the discussions regarding MySQL-style clustering (multi- 
> master etc), what about a "classic" HA cluster for MySQL? Since the  
> OP mentioned high availability, wouldn't the simplest solution be  
> failover clustering (ie. single master with failover, shared  
> storage, fenced nodes etc) via Centos CS?

We have this running in one setup, and it's been working (mostly) fine:

	- master-master setup
	- heartbeat creating a virtual IP
	- all mysql clients use the virtual IP

So, effectively, it's a master-master setup where only 1 master is  
ever receiving traffic, and if that master fails, it'll automatically  
fail-over to the standby master.

The benefit of doing master-master in this scenario is that there's no  
real recovery process needed for restoring redundancy -- when the  
failed master comes back online, it catches up with the current  
master. (Make sure auto-fallback is off in heartbeat.)

The only problem I've seen is that a crashed node may not be able to  
replicate correctly, if its on-disk log position gets out of sync with  
what the other node has. It seems if this happens one has to do a real  
sync (lock tables, lvm snapshot, unlock tables, if you're willing to  
give up the storage needed for the lvm snapshot; or rsync, shutdown  
and re-rsync, startup).

best,
Jeff



More information about the CentOS mailing list