Broken Failovers?

[Jonathan Brockmeier]

Ok, but that would not explain why hosts that have MAC address that match
the hardware address and the start time after the end time were having
problems until the lease file was re-written...both server were operating
fine for renewals and some discoveries...just some clients that seemed to
match as being off for awhile and then coming back and had their lease start
time > lease ends...

Jonathan Brockmeier

On Wed, Apr 2, 2008 at 6:44 AM, Glenn Satchell <Glenn.Satchell at uniq.com.au>
wrote:

>
> >Date: Tue, 1 Apr 2008 20:58:53 -0400
> >From: "Jonathan Brockmeier" <brockj at hope.edu>
> >To: dhcp-users at isc.org
> >Subject: Broken Failovers?
> >
> >I think I have found out that I have been running with broken failover
> for
> >over a year:
> >
> >From Server A:
> >failover peer "PEERNAME" state {
> >  my state normal at 4 2008/03/20 12:49:19;
> >  partner state normal at 1 2007/02/12 19:41:32;
> >}
> >
> >From Server B:
> >failover peer "PEERNAME" state {
> >  my state normal at 4 2008/03/20 12:49:28;
> >  partner state normal at 1 2007/02/12 19:41:48;
> >  mclt 600;
> >}
>
> This is a feature :) The partner state date is filled in the first time
> the lease file is written, and then nothing ever refers to it again. So
> your failover isn't broken. I'm sure that if you search the archives
> for "partner state normal" you'll find a few references.
>
> If you look in the syslog file (varies with different operating
> systems) you'll see what the current state is.
>
> >Also it looks like when I get "pool request" messages the leases that
> seemed
> >to be moved go into a state that they are locked for awhile (same on both
> >servers)
> >lease aaa.bbb.ccc.ddd {
> >  starts 2 2008/04/01 20:08:04;
> >  ends 2 2008/04/01 18:23:13;
> >  tstp 2 2008/04/01 18:23:13;
> >  tsfp 2 2008/04/01 20:08:04;
> >  cltt 2 2008/04/01 18:13:13;
> >  binding state backup;
> >  hardware ethernet 00:c0:17:31:38:8b;
> >  uid "\001\000\300\02718\213";
> >}
> >
> >I would assume due to state 4 or is this related to some of the failover
> bug
> >fixes since 3.0.3?
> >
> >If this is related to failover recovery not working for over a year, what
> is
> >the best way to recover since the servers are "working" for most clients
> >most of the time.
>
> binding state backup means that the lease is available to be allocated
> by the secondary. That's quite normal. Different releases have juggled
> this around a bit, but since about 3.0.5 (I think) it tries to leave
> leases with the one host, rather than bouncong them between the two
> servers.
>
> The dhcpd.leases man page explains what all the fields are.
>
> regards,
> -glenn
>
>
>


-- 
Jonathan Brockmeier, CIT
Hope College
616-395-7670
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.isc.org/pipermail/dhcp-users/attachments/20080402/30ebecb6/attachment.html>