Re: [squid-users] Re: Squid 3.3.2 SMP Problem

From: Amos Jeffries <>
Date: Thu, 31 Oct 2013 20:05:03 +1300

On 31/10/2013 6:02 a.m., Dr.x wrote:
> hi all ,
> ive tried that on kernel of centos 6.4 last one
> but it give me :
> [root_at_squid ~]# sysctl -w net.local.dgram.recvspace=262144
> error: "net.local.dgram.recvspace" is an unknown key
> wt does that mean ?

It means the system control toggle fro datagram packet buffer size is
not called that name (if it exists at all).

> im trying ti use kernel of centos 6.4 without compiling it but i have kid
> registeration time out !!

Kid registration is due two basic causes:
1) the Unix data sockets opened by SHM between the worker kid and
coordinator are somehow broken
  - might show up like this if the coordinator was able to open SHM
properly but the worker has problems.

2) the Worker so overloaded with processing during startup that its
registration packt does not reach the coordinator quickly enough.
  - might happen if a large cache_dir were being scanned/repaired on startup

There could be other reasons, but those are the big ones we see a lot of
reports about.
There are now open bug reports about big rock cache_dir takign too long
to load, AUFS rebuild recovery taking too long to finish, and "squid -z
; squid" too-fast process sequence (eg as seen in many startup scripts)
closing SHM connections midway through startup of the second "squid"

Received on Thu Oct 31 2013 - 07:05:11 MDT

This archive was generated by hypermail 2.2.0 : Thu Oct 31 2013 - 12:00:08 MDT