[OpenAFS] Re: 1.4.x quorum election process?

Andrew Deason adeason@sinenomine.net
Wed, 26 Oct 2011 13:11:27 -0500


On Wed, 26 Oct 2011 13:49:51 -0400
Jeff Blaine <jblaine@kickflop.net> wrote:

> The "lowest IP address" favoritism decision is totally
> arbitrary, no?

I'd say "mostly" rather than "totally". I think it is slightly true that
more important/reliable machines tend to be lower IPs. Or that at least
was true at some point. As far as the protocol goes, though, yes, it's
arbitrary; the factor deciding the "best" site can be anything, as long
as all sites agree who the "best" is.

> We're kind of screwed unless there's a way around it,
> and really would not like to have to apply a local patch
> with every rollout.

So, is the lowest IP in this setup just really unreliable? I'm not aware
of many situations in which this limitation is a show-stopper.

> Would a "favor highest" patch be accepted if it was controlled
> via configure script, defaulting to the traditional behavior?

>From what I remember of long-ago conversations, the most desirable way
to do this is to make the "best" site (or rather, the ranking of sites
from best to worst) runtime configurable and arbitrary. That is, you
specify the list of sites explicitly in order of preference, as opposed
to having several pre-set algorithms or something. I'm not sure if
anyone has code or a more specific design than that.

I'm not sure what degree of sanity checking would be required before
something like this is acceptable... since if the sites disagree about
who is "best", you will be sad iirc, and currently I don't think we have
a way to check that over the wire.

-- 
Andrew Deason
adeason@sinenomine.net