Server lost network connection while dhcp server messages log shows abandoning ip address pinged before offer

Question

Some of my servers lost network connection (ip address disappears in ifconfig). I checked the linux dhcp server's messages log, and I found around the time those servers lost connections, there are messages like below:

DHCPDISCOVER from <mac address of the server> via <port> Abandoning IP Address <IP of the server>: pinged before offer 

I think it is unlikely that my dhcp server's address pool is exhausted (given only very few servers are running and the ip address range is much larger than the number of running servers). I suspect that the same client server that is previously already assigned with the ip is making dhcpdiscover, and while dhcp server tries to assign the client server with that same ip, the dhcp server pings the proposed ip (which the client server is still using it), and eventually this causes the Abandoning IP address message and also then the client server no longer gets the ip address thus loses the network connection.

My questions would be (i) would my hypothesis here be possible? If not, what are some other possibilities (e.g., loop in the network)? (ii) what action can I take so this won't happen in the future?

Answer 1

Are you using DHCP relay? Then pinging the address during discovery might produce a false positive on the duplicate address check. Without relay, an ARP check should be able to match the discoverer's MAC address with the ARPed one.

You should reserve server addresses and make sure that no one else can use them (e.g. strict MAC/IP binding with DHCP snooping on the switch).

Strictly speaking, a discovering DHCP client shouldn't still have an IP address in the lease range, so it shouldn't be pingable. A client trying to extend its lease sends a request only.

Some clients behave that way if their network link is temporarily down and quickly comes up again - they continue using their previous IP address while still using DHCP discover to re-detect the subnet. Accordingly, the effect you're seeing might also indicate cabling problems that you should rule out.