On Thu, 22 Aug 2002, Manuel Bouyer wrote:
> On Thu, Aug 22, 2002 at 01:44:54PM -0500, Lista de NetBSD current wrote:
> >
> > This server works very fine with this values of httpd.conf:
> >
> > MinSpareServers 5
> > MaxSpareServers 16 <- too low for num of users
> >
> > When I increase MaxSpareServers to 32 and restart apache,
> > and the load of users begins, I have hundreds of messages
> > in /var/log/httpd/error_log:
> >
> > httpd[Thu Aug 22 09:17:02 2002] [notice] child pid 6676 exit signal Segmentation fault (11)
> > [Thu Aug 22 09:20:32 2002] [notice] child pid 6671 exit signal Segmentation fault (11)
> > httpd in malloc(): warning: recursive call.
> > FATAL:  emalloc():  Unable to allocate 2 bytes
> > httpd in malloc(): warning: recursive call.
> > FATAL:  emalloc():  Unable to allocate 2 bytes
> > httpd in free(): warning: chunk is already free.
> > httpd in free(): warning: chunk is already free. <---- hundreds like this
> > httpd in free(): warning: chunk is already free.
> > [Thu Aug 22 09:26:56 2002] [notice] child pid 6673 exit signal Segmentation fault (11)
> >
> > I modified the startup scripts of apache and mysql with
> > this line: "ulimit -n 2048"
>
> Maybe you also need to bump maxproc and/or datasize for apache.

THANKS!!!!!
I added the line "ulimit -p 512" to the apache startup script
and all is working well now. See the difference:

with sysctl -a
proc.curproc.rlimit.maxproc.soft = 80
proc.curproc.rlimit.maxproc.hard = 1044
proc.curproc.rlimit.descriptors.soft = 64
proc.curproc.rlimit.descriptors.hard = 3404

with sysctl proc.pid
proc.7591.rlimit.maxproc.soft = 512
proc.7591.rlimit.maxproc.hard = 512
proc.7591.rlimit.descriptors.soft = 2048
proc.7591.rlimit.descriptors.hard = 2048
(the proc 7591 is father of all httpd's and owned by root)

Experience 1
To compile the kernel with "exclusively the hardware found"
is not enough. Is necessary read and understand the options
of the kernel config file. I would like to know the impact
of changing maxusers to 128 or 256, or understand if NMBCLUSTERS
are applicable for this case, etc. All this for have a
"special kernel" for apache+php+mysql and taking into
acount the hardware, num of tables in db, num of httpd's,
etc.

Experience 2
In this server php, imp/horde are configured to send the
messages via local1 of syslogd. It has been working well
except for 3 or 4 messages that were sent to my screen
(ssh session) instead the log file. It was in the 3 days
of the problem. I dont know if it is a problem of syslogd
or php, but under our circunstances, had an erratic way
of work.

Thanks again

Heron Gallegos