Server instability

[Phil Mayers]

On Mon, 2007-09-24 at 15:39 -0400, Nathan Hay wrote:
> I am a newbie, running 3 (for redundancy) FreeRadius servers (1.1.7)
> on SUSE 10 SP1 (32-bit) to authenticate our wireless clients (PEAP
> MSCHAPv2) to our eDirectory via LDAP.  We average 800-900 simultaneous
> wireless clients (need to support a potential 4K in the future).
>  
> The setup works well and authenticates users very quickly, but every
> couple days, the radiusd process will either blow up and start
> consuming 99% of the CPU or die altogether.  More often it blows up.
> We had stability problems initially, even when the process was
> running, so I took everything out of the config that we didn't need
> and that seemed to help.

First question; are you HUPing the daemon? If so, don't - it won't work
well.

Second question; if this happens reliably can you recompile from
scratch:

./configure --enable-developer
make
make install

...and when it happens do this:

gdb /usr/local/sbin/radiusd
set pagination off
set logging file /root/radiusd-wireless.txt
set logging on
attach $PID
thread apply all bt full

...that'll give some details as to what the server is doing when it
pegs the CPU. Other options are strace or (if your Linux system has it)
SystemTap. The aim being to determine what it's doing when it goes wrong.

>