Re: [squid-users] COSS crashes :(

From: Adrian Chadd <adrian@dont-contact.us>
Date: Thu, 14 Dec 2006 02:27:43 +0800

On Wed, Dec 13, 2006, Wojciech Puchar wrote:

> 2006/12/13 18:55:25| assertion failed: coss/async_io.c:203: "reterr == 0"
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

That means that the aio_* routines returned an error. Grr.

> then config was
>
> cache_dir coss /dev/ad4d 9765 max-size=131072 block-size=1024
> cache_dir coss /dev/ad6d 9765 max-size=131072 block-size=1024
> cache_dir diskd /squid/0b 55000 4 256
> cache_dir diskd /squid/1b 55000 4 256
>
> /dev/ad4d and /dev/ad6d are both 20000000 sectors long

Is this under FreeBSD-6? If so, recompile without the --enable-coss-aio-ops and
use --enable-storeio="ufs null coss aufs" rather than diskd.

FreeBSD-5 and FreeBSD-6 have somewhat better threading models than FreeBSD-4
so diskd and the Posix AIO code isn't strictly needed. diskd is buggy under
high load and the AIO stuff available there is pretty naive under the high
loads people want to run COSS under nowdays (read: I wrote it something like
five years ago under a much smaller workload.)

Adrian

-- 
- Xenion - http://www.xenion.com.au/ - VPS Hosting - Commercial Squid Support -
Received on Wed Dec 13 2006 - 11:25:59 MST

This archive was generated by hypermail pre-2.1.9 : Mon Jan 01 2007 - 12:00:01 MST