[Linux-HA] "MySQL Replication/DRBD (was:Shutdown delayed until current resource activity finishes)"

[Clint Byrum]

On Tue, 2004-03-23 at 11:16, Alan Robertson wrote:
> ha at freightgate.com wrote:
> > Hi,
> > 
> > I am still trying to figure out this mysql setup.
<snip>
> Mysql replication is currently asynchronous - so it's a bit risky to use it 
> in a failover situation.  I'd recommend looking at DRBD - which can be 
> configured for synchronous replication.  Then, you have fewer worries.
> 

I think I would disagree about having fewer worries.

With MySQL replication, you just have to rely on your application to
keep things transaction complete. If the transaction finishes, it will
try to replicate. If, halfway through said replication, the master dies,
the transaction never makes it to the endpoint.

Your endpoint still maintains transactional integrity, albeit on a
slightly older version of the data.

With drbd, you are relying on the fact that mysqld thought it wrote the
data. If mysqld is halfway through a big write when the master dies, the
other side will be in a similar pickle as with replication. It will see
that it has an unfinished transaction, and roll back to the last known
good state. Here you're just relying on mysqld to recover from a
corrupted data storage situation.

Is there something else I'm missing?