Re: Réf. : Re: failover or vrrp

[Glenn Satchell]

>Date: Wed, 31 May 2006 11:23:15 +0100
>To: dhcp-users at isc.org
>From: Simon Hobson <dhcp at thehobsons.co.uk>
>
>JVNC04 Yahoo wrote:
>>for your solution (both), should dhcpd.conf be same on both server ?
>
>DHCP config is unrelated to any vrrp you may be running, all the 
>server sees are requests forwarded to it by one (or more) relay 
>agents. You will need just the normal DHCP Failover configuration - 
>ie all but the peer declarations will be common across both servers.

Quick overview of dhcp failover. The two servers divide each pool in
half and both offer addresses from their pool to clients. If one server
stops, then the other continues to offer addresses from its share of
the pool. No special action needs ot be taken when a failure occurs or
when the failed server returns to service. The downside is that the
failover protocol is complex and sometimes the servers get out of step
with each other. Generally though I find it works really well. The load
is shared between the two servers, which can be in completely different
locations. Ther eis no downtime during failover, the partner server
just keeps on running. For extended outages the remaining server can be
switched to a mode where it will hand out addresses from the full
range.
>
>>what about linux vrrp (vrrpd) ?
>
>It's not something I've ever looked at, so I don't know what the 
>configuration would be.

I haven't used it specifically wither, but I have written a similar
failover application a few years ago.

Essentially one system, is master, the other standby.  When the standby
detects that the master has failed it configures the service's IP
address on it's interface, starts up the appropriate services and
continues on. Hopefully the clients don't even notice that the master
switched to a different host. Downside is that the standby has to
notice the failure, manage any shared storage and start up the
services. The tricky part with these things is that the storage, in
this case the dhcpd.leases file, has to be either copied to the other
server on a regular basis, or a shared storage, eg SAN, has to be
used.  It is possible that there can be some (small) anount of data
lost and there is the time involved to detect a failure. Upside is that
from the application point of view little has to be done, so it's good
for apps that don't have failover builtin.

Which one you use is up to you. If you are using vrrp or similar
already and are comfortable with it, then perhaps it makes sense to
keep using it. ISC dhcpd has a good application specific failover
protocol builtin, and in my mind that causes less hassle then switching
things "under the covers", so to speak.

FWIW I run dhcpd's failover on a couple of sites, and a couple of
smaller ones with no failover at all. I do run failover on my home
network though and that only has 2 clients :)

regards,
-glenn