[Gluster-devel] Architecture advice

[Gordan Bobic]

On Mon 12/01/09 13:32 , Daniel Maher <dma+gluster at witbe.net> wrote:
> Gordan Bobic wrote:
> >> How does the HA translator choose a node, exactly ?  Does it
> randomly 
> >> select one from the list of available subvolumes, or does it pick
> them 
> >> in the order they're specified in the config file, or some other
> way 
> >> entirely ?
> > 
> > Not sure what the default is, but you can specify a preferred read
> 
> > server (writes have to go to all servers regardless) with
> > "option read-subvolume" in the volume spec.
> option read-subvolume is available only in the AFR translator, and
> if 
> AFR is being done on server-side (the scenario noted previously in
> this 
> thread), then there would be no AFR declaration in the client spec
> file, 
> and thus no way to set the read-subvolume option.

Not on the client, anyway. But if you're AFR-ing on server side, then your client always talks to one server anyway. The traditional way to handle server failure in that case is to set up Heartbeat or RHCS to fail over the IP address resource to the surviving server.

The TCP connection will reset when the fail-over occurs - I'm not sure how gracefully/transparently GlusterFS reconnects.

> The HA translator is not that useful in the context of a
> straightforward 
> client-side AFR setup in any case, since (as i understand it) if a
> given 
> server dies, the client continues to interact with the remaining
> defined 
> servers seamlessly.  This, of course, is part of the attraction of 
> defining AFR on the cleint side ; but in cases where it is not
> feasible 
> (overhead concerns, etc...), using HA with server-side AFR is a very
> interesting prospect.

I wasn't aware of there being a HA translator built into GlusterFS, but unless you have proper fencing in place, failing over IP addresses won't work. Without proper cluster fencing in place you can easily find yourself in a split-brain situation where both servers think they have the same IP address and neither can talk to any of the clients.

> Some way to extend simplistic load-balancing features to the HA 
> translator (or, perhaps, another translator working in concert)
> would therefore be quite ideal.

I think you're trying to use a wrong tool for the job. Look into something like RedHat Cluster Server.

Gordan
---- Msg sent via @Mail - http://atmail.com/