Hi,

I'm playing around with -CURRENT using a GENERIC.MP with soft RAID
support added in and after a couple of days of uptime I start getting
these messages as more and more services fail (usually squid first, and
if I'm unlucky, sshd):

/bsd: WARNING: mclpool limit reached; increase kern.maxclusters

The default sysctl.conf value for this is 6140 or so. This is far too
low I've found... even 65535 isn't enough with
BIND+Apache+ftp+Squid+sshd+nfs+samba+symon running for a 2 machine local
network (I've been gradually increasing the original default to find a
sane level that doesn't lock up the machine after a few days of uptime).
I'm currently using 128000. Has any one got suggestions for this ? I've
not seen any posts on the subject since this sysctl flag was introduced
to replace nmbclusters. It could well be a memory leak of some kind as
it occurs when memory usage is only at 180-200Mbytes according to top
-S upon crash ( I leave it running on tty0 so I can get a better
indication of crash problems... logs tend to be empty of anything
valuable apart from the above error).

This error did not occur before switching from 3.5 to -CURRENT. I hope
that this error report will help developers.

darknoth

darknoth