On Thu, 2006-10-05 at 04:24 +0000, Sebastian Reitenbach wrote:
> Hi,
>
> Bruno Wolff III <bruno@wolff.to> wrote:
> > On Wed, Oct 04, 2006 at 11:23:45 -0000,
> > Sebastian Reitenbach <itlistuser@rapideye.de> wrote:
> > >
> > > something I thought that might work:
> > > is there sth. that will repair an inconsisten postgresql datastore? e.g.
> the
> > > master database died, the slave will mount the storage, then repair it in
> a
> > > reasonable time, and then start to work.
> >
> > You can just start the slave server and it will recover using th WAL files
> > and should do so in a short time. However, you MUST be certain that the
> > master is no longer running when the slave starts or your database WILL BE
> > TRASHED.
> that sounds great, that I can just start the slave, and it will repair
> whatever broke. I am aware of that I have to make sure that the master is
> really dead.
Be absolutely sure that you test, test, test this. And then test it
some more. There are all sorts of nasty surprises that can happen in an
HA config that can leave you with a corrupt database. As mentioned, two
postmasters writing on the same directory is certainly one.
But you have to watch for problems in the HA config can as well.
What happens if the load on the master shoots up to the point where it
can't release the disk resources in time? Will the other box try and
take over an scribble on your file system?
What happens when you pull the plug on the master while under load
(either physically, or virtually with a halt -q)?
How does the HA solution play with your file system of choice?
Trust me, you want to know the answers to all of these questions and
more before you even consider turning an HA solution on in your
production environment. Don't just rely on the marketing material and
what sales drones tell you. If not, make sure you have good backups.
Brad.