DHCP failover doesn't receive DHCP requests in secondary server

Simon Hobson dhcp1 at thehobsons.co.uk
Thu Aug 16 13:11:39 UTC 2018

perl-list <perl-list at network1.net> wrote:

> The problem isn't with the secondary server.  The DHCP packets must always go to both servers.  If the packets are relayed from a relay agent, then the relay agent must be configured to send to both of the DHCP servers (On a Cisco router, that would involve having ip helper-address listed twice, for example).
> If no relay, then the DHCP servers must both be on the same LAN with the clients.

Just to clarify as the above might give the wrong impression, the only requirement is that broadcast reach both servers and unicast packets reach the IP they are addressed to. It doesn't mean that both must be relayed, or both must be on the same LAN. It's perfectly OK for the servers to be on different networks, even geographically dispersed, provided that all the routing and relay agents are in place.

For example, it would be OK to have one central server, and failover peers in each satellite office of a multi-site business. The central server would take over if the on-site server failed, and the on-site server could continue working if comms to the main site went down. In this setup, clients at the satellite office would access the local server by directly (either broadcast or unicast), and the central server by relay agent (broadcast packets) or normal routing (unicast).

But back to the OPs problem. A bit of history might help, and config files, and information about network topology (clients on same network as servers, or relay agents involved).
If there was originally one server and it's been made into half of a pair, then it's possible that clients already have leases from the original server and so aren't having to broadcast looking for other servers - which means the second server won't see many (if any) requests until some clients are unable to renew by unicast with the original server for some reason.
It will also vary between different sorts of clients - devices that are never shut down might never* need to broadcast a request, while road warrior laptops might need to every time they join the network.

* For some approximation of "never" meaning "between significant server outages".

More information about the dhcp-users mailing list