Re: Crash on 2.5PRE7-CVS

From: Joey Coco <anesthes@dont-contact.us>
Date: Wed, 29 May 2002 20:24:36 -0500 (EST)

Hello,

We're seeing various problems with that release under load as well. It
appears once the squid process grows to around 100MB (which at roughly
300 req/second is about an hour) it becomes useless, almost totally stops
serving connections. Systems run 512MB ram.

It rairly uses more than 40% cpu (Pentium III 600mhz machines, 6 of them
in fact, using accelerated/transparent proxy for 90% of traffic, and
some peering for the remainder).

After our peak time today, I changed the file descriptor max from 1024 ->
4096 to see what happens tomorrow..

The combined download from internet / upload to client is around 12-20
mbit/second during our peak loads.

I've tried turning cache off all together by using null fs, and it makes
no difference at all.

Only errors i've noticed when I run debug to level 9, is some weird
forwarding loop detected messages for some broken urls. Not
enough to break the box I think? Clients with the "code red" virus cause
these same messages, but I used ACL to block those requests.

I'm not really sure what the right combo of servers + hardware is gonna be
to route 100mBit/sec of traffic through squid.
 
-- Joe

On Thu, 30 May 2002, Henrik Nordstrom wrote:

> "Chemolli Francesco (USI)" wrote:
> >
> > I had a crash on a fairly loaded 2.5PRE7-CVS.
> >
> > At the time it was serving about 180 reqs/second (it was not
> > overloaded, CPU was at about 85%), and it died with this error message.
> > At the time it had already successfully served more than 28M requests.
> >
> > 2002/05/29 11:46:13| assertion failed: comm.c:646: "F->flags.open"
> >
> > Only recent config change had been setting the half_open_clients to off.
>
> Could you get a strack trace?
>
> It would be nice to know if this was a client side or server side
> connection..
>
> Regards
> Henrik
>
Received on Wed May 29 2002 - 17:28:09 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:15:31 MST