On Tue, Nov 9, 2010 at 7:19 PM, Tears ! <unix.co@stripped> wrote:
> The problem here is that during step 1 and step 3 we do not want other
> servers to query for any existing locks to avoid duplicate locks. Since our
> two db servers are running in a master-master configuration on a shared IP
> with heatbeat so it might happen that web1 locks a db table on db1 and web2
> ends up querying db2. That's why we want a table to be locked on both db
> servers.
>
Aha, that's where your problem is, alright. As you're finding, it's not a
good idea to do writes on both masters at once, excactly because of locking
and concurrency issues.
Given that you're also implying a loadbalancing scenario, I assume you're
also running ldirectord with your heartbeat and ipfail ? I
*strongly*recommend you add another virtual IP that you use
specifically for write
operations, and only ever assign that to a single master - that'll neatly
avoid you the asynchronicity issues you're having. Ipfail will then switch
that IP over to the other master if the "primary" fails.
--
Bier met grenadyn
Is als mosterd by den wyn
Sy die't drinkt, is eene kwezel
Hy die't drinkt, is ras een ezel
