DHCP Failover and duplicate responses

Jesse Norell jesse at kci.net
Fri Sep 7 18:03:33 UTC 2007


On Thu, 2007-09-06 at 13:59 -0700, David W. Hankins wrote:
> 
> What it comes down to is that clients RENEWING will unicast, so if
> a server gets a RENEWING DHCPREQUEST, then only it *can* answer, so
> it must.
> 
> Clients only get to the REBINDING state if RENEWING times out.  So if
> the client is REBINDING, the server also wants to answer on the basis
> that the other server has been lame with renewals, failing to answer.
> 
> So the design decision is to answer all DHCPREQUESTs so long as it is
> legally allowed to. 

  This is likely quite related to an issue I've been trying to track
down/resolve.  I don't mean to hijack a thread here, but it's pretty
well related.

  Anyways, I've setup a dhcp failover pair and I find that with both
running, normal lease renewals work fine, but an explicit release/renew
assignes a different ip address every time.  It's actually the secondary
server that does that (the primary server will try to assign the same ip
address, the secondary tries to change it, and it always ends up
changed).  Is there a way (configuration) to have the same ip address
always assigned?


  A little more info, if helpful:  I've used dhcp3-server 3.0.4-13 from
debian etch and dhcp-3.0.5-5 from fedora fc6 (and mix/matched the two).
This happens with client directly attached to the subnet served by the
servers, and also through dhcp relay.  Using relay, I've tried both a
cisco router and isc dhcrelay, with the same results:  if I only relay
to the primary server's ip addr, release/renew gets the same ip addr
100% of the time; if I relay to only the secondary server, or to both
servers, a release/renew changes the ip addr every time.

  I've verified that both failover servers are talking to each other
(not a firewall problem, acting like the peer is down or anything).  I
also have a friend with a near identical setup (failover pair running
3.0.5 on Fedora FC6, using dhcrelay to relay clients) who says he
doesn't see that problem.  I see it in production and in a lab setup.


-- 
Jesse Norell
Kentec Communications, Inc.
jesse at kci.net


More information about the dhcp-users mailing list