dnsserver / redirector "memory leak" with Linux SMP

Hi,

We're running a mixture of Squid releases on a number of different Linux
boxes (Slackware 3.5, Slackware 3.6, RedHat 5.2, RedHat 6.0). They all use
the same in-house redirector, and all code is built on the target machine.

Up until recently we used single CPU x86 machines and had no problems.

Recently we introduced dual CPU machines and are running one with the
2.0.36 SMP kernel (RedHat 5.2 install) and two with the 2.2.12 SMP kernel
(RedHat 6.0 install). On these three machines the memory usage of the
dnsserver and redirector processes increases over time. It appears that the
increase occurs each time Squid hits the helper with a request. The scale
of the problem is such that the redirectors consume 2MB when started, but
by the end of the day the first redirector is consuming in excess of 128MB!
Rebooting these machines with the non-SMP version of the same kernel fixes
the problem, but obviously only one CPU is used.

Has anyone experienced similar problems? Can any Squid hackers offer any
insights? Is there a fix?

Thanks,
Simon.
Received on Tue Sep 28 1999 - 08:49:07 MDT