Pierre,
I have the same problems you describe but my hardware environment is a
bit different. I run a cluster of XServers: One
webserver/application
server, One database server and two dedicated WebObjects application
servers.
Before the change to 10.3 I had one application distributed to three
of
the servers running over 20 instances on each server. Now (with the
upgrade to 10.3) I'm running 10 instances on each server.
I have been able to trace the problem down to WOTASKD - over time it
opens too many TCP ports to the running instances and finally gets
stuck with a "Too many files open" error - check your
/var/log/webobjects.log for that message.
I was able to work around the problem by writing a CRON task the
counts
the number of open TCP connections from WOTASKD and restarts WOTASKD
when the number grows larger than 240 (the crash/hang occurs when your
reach 255 TCP connections)
Here is the line I use to identify the open TCP connections (note; you
will have to adjust your port range if you changed the default or you
have more instances running than I do):
sudo lsof -i tcp:2001-2050 -P
Here is the line I use to restart WOTASKD (I had to fix the startup
script too)
sudo SystemStarter restart 'WebObjects Services'
I run this script every 10 min in a CRON job:
#!/bin/sh
#
# Restart WOTASKD when it has too many TCP connections open
#
# by Mark Fisher 11/28/03
# must be run as "root"
#
maxprocs=240
PROCESSLIST=`sudo lsof -i tcp:2001-2050 -P`
i=0
for process in $PROCESSLIST
do
i=$((i+i))
done
if [ $i -gt $maxprocs ] ;then
echo "Restarting WOTASKD"
`SystemStarter restart 'WebObjects Services'`
fi
echo ""
# end shell script
This solves the problem for now but I really, really wish I knew the
cause of the problem - Is WOTASKD *really* that broken in 10.3? I
wonder how many others out there are having the same trouble getting
10.3 Server & WebObjects working reliably.
If *ANYONE* (Apple are you listening) has any more information on this
problem I would like to hear from you.
Thanks in advance,
Mark FIsher
Chief Software Engineer,
Learning Quest, Inc.
_______________________________________________
WebObjects-dev mailing list
email@hidden
http://www.omnigroup.com/mailman/listinfo/webobjects-dev