[squid-users] Socket failure

hello list,

i have a Squid-2.5S2, on a FreeBSD-4.7. it works good. when i
installed, i set up many kernel-parameter for better performance,
and for minimal error.

in machine is 512MB RAM, and 512MB swap.
and any additional info:

root@proxy:# sysctl -a | grep maxfile
kern.maxfiles: 4096
kern.maxfilesperproc: 4096

usually, when i see the actual state of buffers, i get this result:

root@proxy:# netstat -nm
237/16832/65536 mbufs in use (current/peak/max):
        237 mbufs allocated to data
231/16384/16384 mbuf clusters in use (current/peak/max)
36976 Kbytes allocated to network (75% of mb_map in use)
17844 requests for memory denied
0 requests for memory delayed
0 calls to protocol drain routines

and so, there is istalled cachemgr - i usually see that.
in filedescriptors of cachemgr menu, there are just 4-500 open
filedescriptors, no more.

i think, the OS was configured at correctly.

even so, yesterday i get this message in cache.log (and of
course, squid terminated abnormally...):

2003/05/16 07:28:37| comm_open: socket failure: (23) Too many open files in system
2003/05/16 07:28:37| Reserved FD adjusted from 100 to 209 due to failures
2003/05/16 07:28:37| comm_open: socket failure: (23) Too many open files in system
.
.
2003/05/16 07:29:11| WARNING! Your cache is running out of filedescriptors

and many-many of these lines:

2003/05/16 07:44:37| comm_accept: FD 8: (53) Software caused connection abort
2003/05/16 07:44:37| httpAccept: FD 8: accept failure: (53) Software caused connection abort

may be it was some network problem?

Squid is in a DMZ, and clients are in all country, and connected
trough 64k/2M line to squid.

Thanky you:
air
Received on Sat May 17 2003 - 02:05:27 MDT