Re: 1.2.20beta crash

From: Henrik Nordstrom <hno@dont-contact.us>
Date: Mon, 11 May 1998 22:42:45 +0200

Dirk Vleugels wrote:

> Well, i've played with 1.2.20 a little more and tried to raise
> NUMTHREADS to 32 in aiops.c. Launching this binary results in a
> nearly deadlock of the whole machine (load goes above 30, until my
> CRTL-C goes through).

There is a race condition that could cause the load to skyrocket this
way if the main thread god suspended for more than 3 seconds. A
preleminary fix is available at
http://hem.passagen.se/hno/squid/squid-1.2.beta20-1.asyncio.patch

> Anyone running the 1.2.20 threaded with snmp support? I added the
> snmp_conf options D. Wessels posted to this list. This, and NUMTHREADS
> is the only difference to the proxy described below.

I haven't tried it with SNMP, but I frequently run a test squid with a
large number of threads on Linux (100 threads).. No problem with the
load here after fixing the race condition.

> FATAL: You've run out of swap file numbers.

What kind of objects did you test with? This can happen if your test
data has a average object size less than 7KB and more than 15000 cached
objects. (It shouldn't be a fatal error, but it is).

I assume you haven't changed store_avg_object_size. The default 13 KB is
a safe bet in most situations.

---
Henrik Nordström
Sparetime Squid Hacker
Received on Mon May 11 1998 - 15:18:59 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:40:07 MST