[squid-users] Re: Server crashes

Sorry for posting in the wrong mailing list.

I checked my cache.log (openbsd 3.1, sparc64):

2002/11/25 07:40:30| WARNING! Your cache is running out of
filedescriptors
2002/11/25 07:40:32| comm_accept: FD 8: (53) Software caused connection
abort
2002/11/25 07:40:32| httpAccept: FD 8: accept failure: (53) Software
caused conn
ection abort
2002/11/25 07:40:34| comm_accept: FD 8: (53) Software caused connection
abort
2002/11/25 07:40:34| httpAccept: FD 8: accept failure: (53) Software
caused conn
ection abort
2002/11/25 07:40:34| comm_accept: FD 8: (53) Software caused connection
abort
2002/11/25 07:40:34| httpAccept: FD 8: accept failure: (53) Software
caused conn
ection abort
2002/11/25 07:40:34| comm_accept: FD 8: (53) Software caused connection
abort
2002/11/25 07:40:34| httpAccept: FD 8: accept failure: (53) Software
caused conn
ection abort
2002/11/25 07:40:36| sslWriteServer: FD 618: write failure: (32) Broken
pipe.
2002/11/25 07:40:37| comm_accept: FD 8: (53) Software caused connection
abort
2002/11/25 07:40:37| httpAccept: FD 8: accept failure: (53) Software
caused conn
ection abort
2002/11/25 07:40:37| sslReadServer: FD 870: read failure: (54)
Connection reset
by peer
2002/11/25 07:40:38| comm_accept: FD 8: (53) Software caused connection
abort
2002/11/25 07:40:38| httpAccept: FD 8: accept failure: (53) Software
caused conn
ection abort

# sysctl -a | grep maxfiles
kern.maxfiles = 3404

I couldn't find in OpenBSD kern.maxfilesperproc option

# ulimit -a
time(cpu-seconds) unlimited
file(blocks) unlimited
coredump(blocks) unlimited
data(kbytes) 1048576
stack(kbytes) 8192
lockedmem(kbytes) 218541
memory(kbytes) 655624
nofiles(descriptors) 128
processes 1044

I also checked my cachemgr (general runtime information)
Number of HTTP request received: 371321
Page faults with physical i/o: 111

Thank you

On Monday, November 25, 2002, at 03:08 AM, Henrik Nordstrom wrote:

> Please use the squid-users mailinglist for Squid usage and
> configuration
> questions.
>
> If you do not see the swap beeing used and the CPU is not fully
> utilized
> then there most likely is a TCP/IP bottleneck.
>
> Things to check:
>
> * Your cache.log to see if there is any hints
> * Your system messages file to see if there is any hints
> * TCP/IP SYN backlog size and use
> * Kernel memory usage for network buffers
>
> Then depending on the hints found in the log files you might need to
> investigate
>
> * Filedescriptor usage
> * TCP/IP unbound ports availability
>
> And if no other hints are found
>
> * Kernel TCP/IP settings
>
> Regards
> Henrik
>
>
> Monah Baki wrote:
>>
>> Hi,
>>
>> I have a sparc server running squid-2.4-stable7. Its running for an
>> ISP
>> that has 5000 users. The alteon switch is configured for 1300
>> concurrent connection, the server has 3 harddisks 15K rpm and 784MB
>> RAM
>> (soon to be upgraded to 2GB RAM).
>>
>> In squid.conf,
>> cache_mem is 96MB
>> cache_swap_low 85
>> cache_swap_high 100
>> maximum_object_size 2048KB
>>
>> cache_dir ufs /cache1/cache 6000 14 256
>> cache_dir ufs /cache2/cache 6000 14 256
>> cache_dir ufs /cache3/cache 6000 14 256
>>
>> The server after 4 days of uptime, either response becomes too slow,
>> which the ISP has been telling us, or the server crashes. I read FAQ
>> 10.18, but wanted to make sure before I cahnged anything.
>>
>> Can squid handle 1300 concurrent connections or that's too high?
>> If I issue top command, I see squid running at 80+%, I've never seen
>> my
>> swap being used at all.
>> Memory: Real: 402M/659M act/tot Free: 218M Swap: 0K/2047M used/tot
>>
>> Am I doing something wrong? Could it be the OpenBSD has to be tweaked
>> and not squid?
>>
>> Thank you for your help.
>
Received on Mon Nov 25 2002 - 09:40:53 MST