[Pacemaker] Best way to split a resource across two locations, four nodes

[Andrew Beekhof]

On Tue, Jan 18, 2011 at 7:15 AM, Andy Smith <andy at strugglers.net> wrote:
> Hi,
>
> After having only used heartbeat a few years ago I'm now starting to
> look at the newer version, with pacemaker. I'm running the 3.0.3-2
> Debian packages on either lenny-backports or squeeze.
>
> This is all new to me so I'd be grateful if you could check some of
> my assumptions here.
>
> For now I'm interested in clustering DNS resolvers. Anycast might be
> technically better for this but the environment I'm working with
> won't support that for some time yet.
>
> I have hardware in two different suites of a datacentre and I would
> like to offer two different resolver IPs to clients, so I'm thinking
> the best way to do this would be (at least) four nodes, two in each
> suite:
>
> node a, suite 1
> node b, suite 2
> node c, suite 1
> node d, suite 2
>
> I would then have a ResolverIP1 resource that could run on nodes "a"
> or "b" but prefers node "a", and a ResolverIP2 resource that could run
> on nodes "c" or "d" but prefers node "d". That way under ideal
> circumstances ResolverIP1 I believe would exist on some node in
> suite 1 whilst ResolverIP2 would exist on some node in suite 2,
> right?
>
> Should an entire suite go dark (e.g. power fail of whole room),

How do you know "they" had a power failure and "we" didn't loose our
own connectivity?

As with all split site clusters they key is knowing when to take over.
A third site that doesn't run services can help, but finding some way
to provide reliable fencing (usually by sticking a human in the loop)
is critical - otherwise you cant be sure the other side isn't serving
the same services as you.

> I
> would still like for the less-preferred layout of ("a" and "c") or
> ("b" and "d") to happen.
>
> Is this the simplest way to achieve what I want?
>
> Since DNS is largely stateless I suppose I could do something with a
> cloned resource, but I do want two different IP addresses to be
> presented so it seems to me that the above way is simplest.
>
> If I wanted the clusters to reconfigure resources back to preferred
> layout after a suite came back, would I just need to make sure to
> have the location score higher than the stickiness score?
>
> Cheers,
> Andy
>
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
>