On Tue, Apr 14, 2015 at 03:56:05PM -0700, David Fetter wrote:
> 
> Naively, a simple way to do this would be to have >1 machine, each
> running all the slons for a cluster, replacing any machines that fail.
> 
> Would Bad Things™ happen as a consequence?

I seem to recall doing this by accident some years ago, and getting a
lot of deadlocks (and resulting rollbacks).  I know the whole system
is carefully designed for safety, so I don't think it'll break
anything, but I think you'll get a lot of non-optimal locking that
will block stuff.  Also, your troubleshooting will be a nightmare.

I suspect you'd be much better off to run some sort of watchdog across
machines and start in the event you can't reach through.  If you have
a network problem between the nodes, you still shouldn't break
anything, but it's more likely to work smoothly, I think.

A

-- 
Andrew Sullivan
ajs at crankycanuck.ca