Re: Failover/load-balancing with dhcp client bug on “seconds elapsed” field

[Glenn Satchell]

Hi Daniel

"peer holds all free leases" usually indicates that the two dhcp servers 
dhcpd.leases files are out of synchronisation. You might find that 
fixing that problem also solves this issue. It seems that dhcp0 thinks 
the other server should handle this client, but the second server has no 
leases to allocate.

regards,
-glenn

On 04/15/10 08:11, Daniel Duarte wrote:
> Hello all,
>
> We’re having a strange behavior on our network preventing several
> clients from being served by a pair of servers working in failover. It
> seems that the problem is related to a bug on the client that fails to
> increment the “seconds elapsed” field on “request” messages. It only
> increments on the “discovers”. I was hoping that you could suggest some
> workaround or configuration change for this issue, because the update of
> all clients’ firmware might not be an immediate option.
>
> This is our scenario:
>
> - We need to give a static IP to our clients based on the option 82
> location info received by the server;
>
> - to do so we created matching classes, and pools with only one IP
> available. On each on those pools we accept only one mach class;
>
> - when the client sends the first Discover, dhcp0 ignores (I believe he
> thinks it will be handled by its peer), but dhcp1 writes on the log
> “peer holds all free leases” (I believe dhcp1 isn’t controlling this
> specific lease);
>
> - The client keeps trying and incrementing the “seconds elapsed” field
> on the “discovers”. When it becomes greater than our configured “load
> balance max seconds”, dhcp0 answers the request with the usual “offer”;
>
> - The problem is when the client sends the “request”, because it sends
> it with “seconds elapsed” = 0! The transaction id is the same, and all
> the other parameters seem to be correct except this one.
>
> - Dhcp0 ignores the request even that on the options the client
> specifically states that he’s answering his offer.
>
> To me it seems clear that the problem is the bug on the dhcp client,
> however it looks like that the failover / loadbalancing is a lot
> dependent on the good implementation of all the clients. Without
> failover this would work perfectly, ignoring this bug.
>
> Do you think there could be any workaround to solve this problem in the
> server side?
>
> Is dhcp0 working correctly when it ignores the “request” with
> seconds-elapsed = 0?
>
> Thanks,
>
> Daniel Duarte