[users@httpd] Apache taking 99% CPU load

[Nico Sabbi <ns...@tiscali.it>]

Hi,
I have an apache-1.3.31 web server on a Mandrake 10.0 system, statically 
compiled with mod_perl,
loading all the other modules as DSO:
|*
$ /usr/sbin/httpd-perl -v
Server version: Apache-AdvancedExtranetServer/1.3.31 (Mandrakelinux/4mdk)
Server built:   Jul 13 2004 18:34:47
$ /usr/sbin/httpd-perl -l
Compiled-in modules:
  http_core.c
  mod_so.c
  mod_perl.c

The configuration I use as at the bottom of this message.
I can't pass to Apache2 because one of my vhosts is strictly tied to 
mod_perl1.
I'm running this configuration on a Dual P4 2.8 with 512 KB embedded 
cache, 2 GB ram
and Dell Perc 3/DI 72 GB Raid 10 system, and in addition a much less 
powerful
server runs exactly the same configuration and vhosts (a combination of 
perl/php sites).*|m
|*
My problem is that on the primary server, after 10-15 minutes, I can 
always see
some httpd-perl taking 99% cpu usage (although memory usage is not a 
problem),
e.g. something like:

top - 12:51:21 up 33 days, 22:14,  1 user,  load average: 1.47, 1.18, 2.18
Tasks:  74 total,   4 running,  70 sleeping,   0 stopped,   0 zombie
Cpu(s):  15.7% user,   5.9% system,   0.0% nice,  78.4% idle
Mem:   2068140k total,   869616k used,  1198524k free,    68960k buffers
Swap:  2096440k total,     4232k used,  2092208k free,   254172k cached

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
20517 www       25   0 57508  56m  28m R 99.5  2.8   3:27.61 httpd-perl
20680 www       25   0 34028  33m 8188 R 88.4  1.6   0:00.76 httpd-perl
20513 www       25   0 38592  37m  27m R 16.6  1.9   0:01.51 httpd-perl
20692 root      25   0   908  908  700 R 12.9  0.0   0:00.12 top
    1 root      25   0   112   80   64 S  0.0  0.0   0:37.65 init
    2 root      RT   0     0    0    0 S  0.0  0.0   0:00.01 swapper
    3 root      RT   0     0    0    0 S  0.0  0.0   0:00.00 swapper
*|
and so on, but the load average gets as high as 80.

I noticed that MaxRequestsPerChild  doesn't seem to be respected:
when it was set at 5 I could see in the logs at least 15 requests
served by the same pid.

The secondary server, though much less powerful and running exactly the 
same software/library versions,
never shows this behavior. The only significant difference is the 
kernel: 2.4.27 on the "sick" server (I can't replace
it, because of serious bugs in the raid driver that crash the server 
very soon), 2.6.3-7 on the secondary.

Can anyone explain
1) why apache doesn't die after having reached the requests limit
2) why it takes 99% cpu?
3) what other (and better than mod_status/Apache::VMonitor) can I use to 
see what's going on
on the server?

Thanks very much,
       Nico