Failover on a large DHCP system

[Glenn Satchell]

>From: John Wobus <jw354 at cornell.edu>
>Subject: Re: Failover on a large DHCP system
>Date: Fri, 8 May 2009 17:17:53 -0400
>To: Users of ISC DHCP <dhcp-users at lists.isc.org>
>X-BeenThere: dhcp-users at lists.isc.org
>
>On May 7, 2009, at 10:30 AM, Nicholas F Miller wrote:
>> We have a large DHCP instance which is currently running 304 
>> shared-network definitions and 677 pools. We would like to implement 
>> DHCP failover but are a little worried about the overhead needed to 
>> implement it on such a large DHCP network. Will the servers be 
>> overwhelmed trying to keep things in sync with so many pools?
>>
>> Also, can we reload configs without restarting DHCP yet?
>> _________________________________________________________
>> Nicholas Miller, ITS, University of Colorado at Boulder
>
>We use failover and avoid using omshell by reconfiguring and restarting 
>the daemon once every 2 minutes
>(if there is a config change). It generally works well (we've been 
>doing it for years), and failover actually aids coverage since we 
>stop/start the
>two servers one after the other.  In general, though, if there are any 
>bugs that somewhat-rarely causes problems when you restart
>the daemon, we see them, given all our restarts.  I believe there are 
>other sites that do pretty much the same thing.
>
>I might consider omshell, but haven't because (1) we've been doing this 
>and it works, i.e., inertia; (2) rumors of its demise;
>and (3) I like having our config all in one file: we use short leases, 
>so our entire lease file is short-term data that we don't
>worry about losing: even if we lost both lease files, it would not be 
>much of a disaster for us.
>

Static, ie hosts configured with a host statement, do not get a dynamic
lease as such, so are not part of dhcpd.leases and do not get
replicated by failover.

If you had a dhcpd configuration that only used static host assignment
you could create a number of dhcpd servers with identical configuration
and no failover. The reason is that all servers would give out the same
answer give out the same answer each time.

Also, if you do use omshell, then you need to add the host to all dhcp
servers, since it does not get replicated by failover.

I have seen setups that used omshell to dynamically add hosts (it's
really quite simple and there is an example on the omshell man page)
and also added a host statement to dhcpd.conf. No restart of dhcpd is
required, but the entry is there to avoid operator confusion, and gets
picked up on the next restart.

I think the rumour of omapi's demise is probably some time off. We
still see questions on this list for 5 year old versions of dhcpd, so
there's still a lot of life left in it I think.

regards,
-glenn
--
Glenn Satchell     mailto:glenn.satchell at uniq.com.au | I telephoned the
Uniq Advances Pty Ltd         http://www.uniq.com.au | swine flu info
PO Box 70 Paddington NSW Australia 2021              | line and all I got
tel:0409-458-580  tel:02-9380-6360  fax:02-9380-6416 | was crackling.