Re(2): Memory leak in 2.2S5 and 2.3S1?

>About a month ago I recompiled squid to add delay pools. Shortly
>afterwords, we started having problems with machine and squid crashes. I
>have noticed that squid will start out using about 100meg of memory at the
>beginning of the day. It will be using about 150meg during the peak time
>of the day. At the end of the day, the memory is at about 190meg and
>climbing at a rate of about 1 meg every few minutes. At this point the
>utilization of the cache is quite low (6 reqs/sec). Something else that
>happens about this time is that the byte hit ratio in cachemgr turns into
>a negative number (might be unrelated).
>
>I updated to version 2.3s1 to see if it would help with the problem. At
>this time I also added async-io and heap-replacement to see if it would
>help performance (it helped a lot). But the memory usage is still
climbing.
>
>This is running Redhat version 6.?, on a Dell Poweredge 2300, 450 Mhz,
>256meg ram, and 10gig cache (which I purged when I upgraded to 2.3). Today
>it processed about 1,000,000 requests.
>
>The original squid was compiled without any special options with the
>exception that our filtering package (smartfilter) was patched in. I
>recompiled with delay pools and the problems seemed to start. Upgrading to
>2.3s1 with the addition of async-io, delay-pools, heap-replacement, and a
>new smartfilter module didn't seem to help.
>
>I would really like to keep the delay-pools if possible because without
>it, internet traffic was consuming all of our bandwidth and causing
>problems for other applications. Any help to resolve this problem is
>greatly appreciated.
>
>Thanks,
>
>Erik
>

The problem seems to be solved now thanks to several suggestions (which
one fixed it I'm not sure), but is what I did:

First, I goofed on the version of redhat, it was version 5.2.

I recompiled first with the patch to show delay-pool memory usage. It
showed that delay-pools were not the source of the problem.

I then recompiled with Henrik's persistant post patch, dl-malloc,
heap-replacement (for lfuda replacement), and async-io (threads=32). This
version so far is not leaking memory, at least not at the large rate as
before.

I increased the IP port range because some brief monitoring showed that I
was getting too close to the default limit for comfort. Hopefully this
fixes the machine crashes.

Thanks to everybody who helped.

Erik
Received on Tue Apr 04 2000 - 10:28:47 MDT