Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 29 Sep 1997 12:01:24 +0100
From:      Alan Judge <Alan.Judge@indigo.ie>
To:        scsi@freebsd.org, stable@freebsd.org
Subject:   Kernel disk hangs under load with 2.2-STABLE
Message-ID:  <199709291101.MAA16750@indigo.ie>

next in thread | raw e-mail | index | archive | help
[Posting to both scsi and stable, since I'm not sure which is at
 fault and if there is a hang bug in 2.2-STABLE, it would be good to
 find it.]

I have a machine running squid (with the data stored on a CCD striped
set of disks).  Since it's a major web cache, the disks get a fairly
steady and sometimes heavy load.  The machine doesn't do anything else
and squid is the only large process (RSS of 120MB on a 256MB machine).

The kernel is 2.2-STABLE, cvsuped last Tuesday; hardware is dual
Adaptec 2940UWs with Quantum Atlas II UW disks.  No unusual config
options other than increases of NBUF and NMBCLUSTERS.

Every so often (about once in two to three days), the squid process
hangs and shows up in D state in ps.  The process cannot be killed and
commands that are disk related sometimes hang.  The only cure is to
reboot, and the reboot is usually not clean (syncing disks,
.... giving up).

Oddly, reducing the size of the web cache (which indirectly reduces
squids RSS and the proportion of the CCD array in use) reduces the
frequency of the hangs.

I know that's probably not enough info to find a bug.

Any suggestions on how to start debugging this?  What commands
should I try?  How do I find what squid is stuck waiting for?
Should I bring up gdb or DDB and look at something?
-- 
Alan Judge                                      Phone: +353-1-6046901
Indigo Network Operations Centre                Fax:   +353-1-6046948




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199709291101.MAA16750>