DHCP Relay agent not forwarding messages to the client

[Peter Rathlev]

On Tue, 2015-06-16 at 12:16 -0700, Sean McMurray wrote:
> I have zero experience with dhcp failover, but I agree with Gero. They
> should not both make offers.
> 
> If they are both making offers, what is the point of configuring them
> in failover? You might as well set them up independent of each other
> and have them serve different pools.
> 
> What am I missing?

The point in failover is at least a) that the still online server can
ACK renew requests from clients that got their lease from the other
server and b) that the two servers make sure the pool of free leases are
balanced between them as long as they can communicate so that the
remaining server can serve at least "pool-free-leases / 2" new clients
from when the other crashed.

This means that with just one server online you can survive for a longer
period than if you had started with two separate servers with half the
pool each.

The failover communication also means that you can lose one server
completely and still recover gracefully. Even with the local lease file
gone a new server would just download a copy from the partner. This is
probably less important as most people have ample redundancy in their
disk subsystems but worth mentioning.

Regarding your observation: It is possible that the "secs" parameter
from the client is larger than the configured "load balance max seconds"
value. If it is then load balancing is supposed to be disabled and both
servers should answer.

Taking a quick look at a capture on both our servers I see many clients
that send "secs" larger than the 3 that we have configured. The ones I
watched that had seconds less than 3 (typically 0) were only answered by
one of the servers.

The "secs" parameters should not be mangled by relay agents so you
should not see a difference between relayed and locally broadcast
discover messages.

If both servers answer queries that have "secs" less than your
configured "load balance max seconds" then that does sound like a bug of
some kind. I haven't thoroughly checked if our servers do that since it
doesn't really bother me. :-)

-- 
Peter