David W. Hankins
David_Hankins at isc.org
Thu Jun 8 15:34:44 UTC 2006
On Thu, Jun 08, 2006 at 02:56:07PM +1200, Eustace, Glen wrote:
> Having spent sometime looking at a couple of packets traces, 1 of a Hard
> Disk boot and one a Net Boot of the same Macintosh, the only difference
> I can see that may effect the initial lease allocation is a field in the
> DISCOVER that Ethereal calls 'Transaction Id'. In the HD boot it is
> 0x00003c3f and in the netboot case it is 0x00003bdd.
Yeah, that's normal, and that shouldn't affect lease allocation.
> The OFFER from the
> dhcp server is x.x.217.189 in the first case and x.x.217.162 in the
> other. Comparing the entries in the leases file, both have the same
> 'uid' but the ddns-txt entry is different first starts with '008c' and
> the other '318c'.
That's odd. The packets I'm looking at (on both sides) have
no client identifier...so both should use the chaddr.
> thus sends out a DISCOVER and gets 2 OFFERs in succession, I presume
> that the result from the clients perspective is to merge both sets of
> attributes. In both cases the client sends a REQUEST for the offered IP
> numbers and in both cases, that request is ACK'ed
That would also be a unusual - in theory the client is only supposed to
use values in the binding it REQUESTs. So only one wins out.
> At this stage I am none the wiser.
It's less fuzzy. It all starts with this:
Seconds elapsed: 4
That's set on both DISCOVERs in both traces.
I'm guessing you have load balance max secs set to less than this
because both of your failover servers are answering.
So this isn't quite so mystifying behaviour anymore.
What's mystifying is how the lease states are conspiring to get
If the lease is ACTIVE (which it should be the second time because
there's only 9 minutes inbetween these two and you gave it a 1-hour
lease) both servers should choose that lease to offer.
If the lease is FREE or BACKUP (does the client release its lease?),
then the behaviour is more reasonable - both servers must answer, and
they can only offer leases in one of those two states.
Best place to look now would be at your failover relationship...if
the lease went active on one server, and that didn't propogate to
the other server, I think it would result in this.
Certainly in 9 minutes the servers should have been able to process
that binding update.
David W. Hankins "If you don't do it right the first time,
Software Engineer you'll just have to do it again."
Internet Systems Consortium, Inc. -- Jack T. Hankins
More information about the dhcp-users