----- Original Message -----
> On 12/5/2012 10:11 AM, Tim Nelson wrote:
> > I've successfully setup 2x pfSense boxen with CARP. It is working
> > properly, with ~1 second failover. The following test scenarios
> > work well:
> > 
> > -Unplugging a link (WAN, LAN, etc)
> > -Causing system crash (kill -9 1)
> > -Unplugging both SAS HDDs (actually carp doesn't come into play,
> > system keeps routing traffic happily, even though system errors
> > are flying by on the console)
> > 
> > I am finding a situation however where failover is not happening,
> > but I'm not sure if it's *supposed* to in this case:
> > 
> > CARP seems to fail over upon NIC link change (down). If I move the
> > WAN (on my primary system)with a CARP IP from a 'live' switch with
> > connectivity to the outside world to a different switch without
> > connectivity to the outside world, I lose all connectivity. I
> > guess my assumption was the loss of *routing* connectivity would
> > trigger CARP, but it appears this isn't the case. To make matters
> > even more confusing, during this time, both my primary and backup
> > system list the WAN CARP IP as Master.
> > 
> > Does that make sense? Is this expected behavior or am I missing
> > something?
> 
> CARP works based on whether or not each node can see the heartbeats
> from
> the other node. Routing doesn't have anything to do with it, it's all
> layer 1/2.
> 
> The backup will try to take over since it would no longer see
> advertisements from the master, but that just means you'd be dual
> master
> on some VIPs and probably not functional.
> 
> The primary will never demote itself unless it loses link on an
> interface. If the link is still up, the primary will keep going
> master
> on all VIPs.
> 
> That's a bit of an edge case we've been trying to come up with a
> nice/elegant solution for.
> 
> Mostly that bites people using it in ESX where the vswitch doesn't
> lose
> link if the physical NIC dies. You can work around that with some ESX
> mojo to bring down the vswitch if the physical link goes away though.
> 

Ah, this does indeed make sense. I suppose CARP was intended to handle link 
failure of a specific type (sounds like L1/L2), but not *all* failure types.

Thanks Jim!

--Tim
_______________________________________________
List mailing list
[email protected]
http://lists.pfsense.org/mailman/listinfo/list

Reply via email to