Lustre discuss - Dec 2008 - Bug when using /dev/cciss/c0d2 as mdt/ost

I am trying to build lustre-1.6.6 against the pre-patched kernel downloaded
from SUN.
But as written in Operations manual, it creates rpms for
2.6.18-92.1.10.el5_lustrecustom. Is there a way to ask it not to append
custom as extraversion.
Running kernel is 2.6.18-92.1.10.el5_lustre.1.6.6smp.
-- 
Regards--
Rishi Pathak
National PARAM Supercomputing Facility
Center for Development of Advanced Computing(C-DAC)
Pune University Campus,Ganesh Khind Road
Pune-Maharastra
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20081224/15fc3c3e/attachment-0001.html

In this setup there are 15 OSS each having 840GB targets and one MDS with
840GB mdt. All nodes are HP DL580G5 servers. These servers come with HP
Smart Array P400.Raid volumes are visible as /dev/cciss/[c0d0,c0d1].
Creation of back end fs used to fail giving errors

mkfs.lustre FATAL: failed to write local files
mkfs.lustre: exiting with 12 (Cannot allocate memory)

. A search on lustre-dicuss showed that this was a bug in ldisk module. See
the below link for more clarification:
http://www.mail-archive.com/lustre-discuss at lists.lustre.org/msg01713.html

The solution was to fix the code and rebuild it against the pre-patched
lustre-kernel source. I hope I have made myself clear :)

On Wed, Dec 24, 2008 at 9:59 PM, Brian J. Murrell <Brian.Murrell at
sun.com>wrote:
> On Wed, 2008-12-24 at 11:02 +0530, rishi pathak wrote:
> > I am trying to build lustre-1.6.6 against the pre-patched kernel
> > downloaded from SUN.
>
> Can I ask, just out of curiosity, why you are trying to build and
don''t
> just use the provided RPMs?
>
> > But as written in Operations manual, it creates rpms for
> > 2.6.18-92.1.10.el5_lustrecustom. Is there a way to ask it not to
> > append custom as extraversion.
>
> If I understand your problem correctly, you can edit the Makefile in the
> kernel source and remove the "custom" from the EXTRAVERSION.
I tried that but it seems like it is hard coded in each and every
autoMakefile & Makefile.
I accepted defeat because of deadline and rebuild kernel and lustre again.
Setup is over IB and is working fine (yet to benchmark it though).
>
>
> b.
>
>
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>


-- 
Regards--
Rishi Pathak
National PARAM Supercomputing Facility
Center for Development of Advanced Computing(C-DAC)
Pune University Campus,Ganesh Khind Road
Pune-Maharastra
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20081225/15f4e395/attachment.html

Interesting. Just for the record, we have a P800 controller on DL380,
and we get no such problems. Running on 1.6.5

Has this problem been addressed? Because we are planning to go to
1.6.6, but I don''t want to face problems like this :-)

TIA



On Thu, Dec 25, 2008 at 1:09 AM, rishi pathak <mailmaverick666 at
gmail.com> wrote:
> In this setup there are 15 OSS each having 840GB targets and one MDS with
> 840GB mdt. All nodes are HP DL580G5 servers. These servers come with HP
> Smart Array P400.Raid volumes are visible as /dev/cciss/[c0d0,c0d1].
> Creation of back end fs used to fail giving errors
>
> mkfs.lustre FATAL: failed to write local files
> mkfs.lustre: exiting with 12 (Cannot allocate memory)
>
> . A search on lustre-dicuss showed that this was a bug in ldisk module. See
> the below link for more clarification:
> http://www.mail-archive.com/lustre-discuss at
lists.lustre.org/msg01713.html
>
> The solution was to fix the code and rebuild it against the pre-patched
> lustre-kernel source. I hope I have made myself clear :)
>
> On Wed, Dec 24, 2008 at 9:59 PM, Brian J. Murrell <Brian.Murrell at
sun.com>
> wrote:
>>
>> On Wed, 2008-12-24 at 11:02 +0530, rishi pathak wrote:
>> > I am trying to build lustre-1.6.6 against the pre-patched kernel
>> > downloaded from SUN.
>>
>> Can I ask, just out of curiosity, why you are trying to build and
don''t
>> just use the provided RPMs?
>>
>> > But as written in Operations manual, it creates rpms for
>> > 2.6.18-92.1.10.el5_lustrecustom. Is there a way to ask it not to
>> > append custom as extraversion.
>>
>> If I understand your problem correctly, you can edit the Makefile in
the
>> kernel source and remove the "custom" from the EXTRAVERSION.
>
> I tried that but it seems like it is hard coded in each and every
> autoMakefile & Makefile.
> I accepted defeat because of deadline and rebuild kernel and lustre again.
> Setup is over IB and is working fine (yet to benchmark it though).
>>
>>
>> b.
>>
>>
>> _______________________________________________
>> Lustre-discuss mailing list
>> Lustre-discuss at lists.lustre.org
>> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>
>
>
> --
> Regards--
> Rishi Pathak
> National PARAM Supercomputing Facility
> Center for Development of Advanced Computing(C-DAC)
> Pune University Campus,Ganesh Khind Road
> Pune-Maharastra
>
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>
>

On Mon, Dec 29, 2008 at 11:18:01AM -0500, Mag Gam wrote:
> Has this problem been addressed? Because we are planning to go to
> 1.6.6, but I don''t want to face problems like this :-)
FYI, the ENOMEM issue with the cciss driver is addressed in bug 17490 and
will be fixed in 1.6.7.

Cheers,
Johann

On Wed, 2008-12-24 at 11:02 +0530, rishi pathak wrote:
> I am trying to build lustre-1.6.6 against the pre-patched kernel
> downloaded from SUN. 
Can I ask, just out of curiosity, why you are trying to build and don''t
just use the provided RPMs?
> But as written in Operations manual, it creates rpms for
> 2.6.18-92.1.10.el5_lustrecustom. Is there a way to ask it not to
> append custom as extraversion.
If I understand your problem correctly, you can edit the Makefile in the
kernel source and remove the "custom" from the EXTRAVERSION.

b.

On Monday 05 January 2009, Johann Lombardi wrote:
> On Mon, Dec 29, 2008 at 11:18:01AM -0500, Mag Gam wrote:
> > Has this problem been addressed? Because we are planning to go to
> > 1.6.6, but I don''t want to face problems like this :-)
>
> FYI, the ENOMEM issue with the cciss driver is addressed in bug 17490 and
> will be fixed in 1.6.7.
As I too have a bunch of HP machines (with cciss storage) in the pipe for 
lustre installation. How far along is 1.6.7 or will I have to patch+build 
1.6.6?

Tia,
 Peter
> Cheers,
> Johann
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part.
Url :
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20090120/317c3f54/attachment.bin

On Tue, Jan 20, 2009 at 05:48:41PM +0100, Peter Kjellstrom
wrote:
> As I too have a bunch of HP machines (with cciss storage) in the pipe for 
> lustre installation. How far along is 1.6.7 or will I have to patch+build 
> 1.6.6?
We are about to tag 1.6.7-rc3 which should hopefully be the last release
candidate.

Johann