[kleung@diversa.com: RE: [PVFS2-users] Pvfs2-server dies]

neillm at mcs.anl.gov neillm at mcs.anl.gov
Fri Sep 3 16:58:59 EDT 2004


----- Forwarded message from Kenneth Leung <kleung at diversa.com> -----

From: Kenneth Leung <kleung at diversa.com>
Subject: RE: [PVFS2-users] Pvfs2-server dies
To: neillm at mcs.anl.gov
Date: Fri, 3 Sep 2004 13:45:51 -0700
X-Spam-Status: No, hits=-1.0 required=5.0
	tests=EMAIL_ATTRIBUTION,QUOTED_EMAIL_TEXT
	version=2.55

Thanks guys. I will test this out but it certainly makes sense. There
are lots of other apps that had the same problem when Redhat 9.0 came
out. Redhat 9.0 introduced new threading libraries as Neill said, and
FWIK apps that weren't written to strict POSIX standards would seg
fault.

Ken 

-----Original Message-----
From: neillm at mcs.anl.gov [mailto:neillm at mcs.anl.gov] 
Sent: Friday, September 03, 2004 12:36 PM
To: Kenneth Leung
Cc: pvfs2-users at beowulf-underground.org
Subject: Re: [PVFS2-users] Pvfs2-server dies

On Fri, Sep 03, 2004 at 10:07:02AM -0700, Kenneth Leung wrote:
> I ran the server in the foreground and it looks like a pvfs2-cp 
> command generates a segmentation fault. Here is what was displayed by 
> the
> server:
> 
> About to retrieve attributes for handle 613566760
> (0x80de238) io (prelude sm) state: perm_check
> (0x80de238) io state: send_ack
> (0x80de238) io state: start_flow
> (0x80de238) io state: send_completion_ack
> (0x80de238) io state: release
> (0x80de238) io state: cleanup
> Segmentation fault
> 
> We have 5 I/O servers and the same thing seems to happen on all. Is 
> are there any other troubleshooting steps that can be done?

Hello Ken,

Nathan, thanks for steering Ken in the right direction, but I've been
able reproduce this on Redhat 9.0 just as described and it's an _odd_
bug.  We don't ordinarily develop on the Redhat distro, so it's hard to
notice some quirks like this.  It appears to be a broken NPTL threading
library, or else we're somehow using it incorrectly (as it differs from
other popular distros where our usage is ok).  For now, there's only one
workaround that I know about, and that is unfortunately to disable the
use of the NTPL threading libraries on Redhat 9.0.

To do this, you can add the following line to your server start scripts
just before running the pvfs2-server:

[ for the bash shell ]
export LD_ASSUME_KERNEL=2.2.5

[ for other shells ]
set LD_ASSUME_KERNEL 2.2.5

For more information on what setting this environment variable does,
please see http://people.redhat.com/drepper/assumekernel.html


Best regards,
-Neill.





----- End forwarded message -----


More information about the PVFS2-users mailing list