Comments on: Making sense of MySQL HA options

By: Justin Huff

Justin Huff — Mon, 13 Jul 2009 22:06:07 +0000

I like leaving log-slave-updates on even if you don’t loop it. Makes patching things back up a little easier if you have to switch roles.

I don’t let MySQL start slave automatically….just like I don’t let MySQL start on boot. Events of that magnitude need a human touch:)

You might check out: http://code.google.com/p/mysql-master-master/

By: btm

btm — Mon, 13 Jul 2009 20:03:45 +0000

Right. That’s basically what I’ve done except I used heartbeat (was relatively simple actually) and I haven’t added the circular dependency. If I do, it will be after I take down the legacy server. I’ll have to rebuild the slave anyways because I’m currently: legacy -> master -> slave, without log-slave-updates to the slave is likely out of sync already.

My only concern with circular replication is knowing if there was any corruption on the master if it crashes before I bring it back online.

By: Justin Huff

Justin Huff — Mon, 13 Jul 2009 18:51:41 +0000

Yeah, circular replication with two hosts — each host is the replication master for the other. They are generally active/passive. I can (and do) have various background jobs that write to the ‘passive’ host, but this must be done with care.

Since MySQL replication is single threaded, lots of writing to both is sure to cause replication lag, which will make life painful.

By: btm

btm — Sat, 11 Jul 2009 04:40:38 +0000

What do you mean by master master?

Are you using circular replication? Are you writing to the same databases on both servers simultaneously or did you split the databases in a more traditional active/active setup (with two VIPs?) with either server able to back up the other?

By: Justin Huff

Justin Huff — Fri, 10 Jul 2009 23:54:14 +0000

All of our production stuff is master-master MySQL replication. It’s pretty solid, gets you a place to do intense read-only queries, and also allows you to do other nifty stuff like online schema changes. I also like having two totally separate sets of datafiles, lest some corrupted data hork the InnoDB tablespace.

I started with heartbeat, but quickly ran away screaming. Some of it’s more intense features (STONITH) are needed for DRBD, but it’s a downright pain otherwise. That and it bit me in the ass a couple times. Now, I use keepalived to simply manage a VIP.