DHCP don't acknowledges more than 80 users??

[John Wobus]

>>
> John, thank you for your detailed expanation  - what I don't 
> understand is that infact the real primary
> server is servering all the clients without any interaction of me.
>
> I did the tests last wednesday by setting shutting down both (3.0.3 
> and 3.0.5) makeing the
> 3.0.5 that was secondary the primary one (by simply coping the config 
> of the primary 3.0.3
> and changing the IP address)  and start only this dhcpd in the night 
> to see if
> it works alone next day - what did not succeed.
> Now I shuted down the 3.0.5 and started the formerly primary (3.0.3) 
> again (with the
> same configuration that had the 3.0.5 for testing before) ... and it 
> works - without any
> partner down setting and without seeing the 3.0.5 any time.
>
> Is there a difference in the behaviour between 3.0.3 and 3.0.5 or is 
> the Suse Linux Enterprise dhcpd 3.0.3
> patched somehow that it really could do a failover without manual 
> "partner down"?

Both servers know about all existing leases other than any for which 
the inter-server communcation
may not have completed.  I admit I'm vague on what the servers do with 
this info in communications
interrupted mode: my guess is that either server will renew the lease 
but not reuse the IP, thus exhibits
some behavior that might be called "taking over", but not all.  I 
invite others to chime in.

I don't know specific differences between 3.0.3 and 3.0.5 or Suse Linux 
Enterprise dhcpd.  What I _can_
say is that state changes are logged, and it should, at minimum, be 
easy to trace which state-behavior the
server is intending to carry out.  For your analysis, besides tracing 
an affected IP in the log, it can also be
helpful to save some copies of the lease files while reproducing the 
problem, and trace the IP in them
during analysis.

I also recall problems with dhcpd (3.0.3, I believe) that take IPs out 
of service.  It has a problem
with synchronizing the two daemons' state for the specific IP, and an 
IP could become stuck in a state
where neither daemon would lease.  This made pools act like they were 
smaller than they really were.
We run 3.0.4 now, which mitigates the problem as exhibited at our site 
to tolerability and I believe
subsequent dhcpd versions have improved things even more.  The problem 
appeared to be
exacerbated by lots of dhcpd restarts, something we do much much more 
than many sites.  The issue
was extremely clear once I began looking at individual IPs in the lease 
file.

John Wobus