Lustre discuss - Sep 2009 - Issue with 1.8.1 patchless client and 1.6.7.2 server

Specs:
All OSSes and MDSs are 1.6.7.2 (Centos 5.2) 2.6.18-92.1.26.el5_lustre.1.6.7.2smp
Client is Ubuntu Jaunty running vanilla kernel 2.6.27.34 and patchless
lustre 1.8.1

Issue:
The client mounts the filesystem just fine without errors. When I try
to copy a file from lustre to a local partition on the client I get a
whole slew of errors:
See http://pastie.org/622029.txt

The only message I see on the MDS and OSSes are:
LustreError: 6028:0:(socklnd_cb.c:2228:ksocknal_recv_hello()) Unknown
protocol version (2.x expected) from 10.2.1.176
but that''s expected since the 1.8.1 client tries to use the new
protocol, but then drops back down to the old one.

I''m not sure if it''s related to Bug 19561
Is there any other data that I can provide to help figure out what''s
going on?

Thanks,
Scott Barber
Senior Systems Administrator
iMemories.com

On Sep 18, 2009  11:15 -0700, Scott Barber wrote:
> All OSSes and MDSs are 1.6.7.2 (Centos 5.2)
2.6.18-92.1.26.el5_lustre.1.6.7.2smp
> Client is Ubuntu Jaunty running vanilla kernel 2.6.27.34 and patchless
> lustre 1.8.1
> 
> Issue:
> The client mounts the filesystem just fine without errors. When I try
> to copy a file from lustre to a local partition on the client I get a
> whole slew of errors:
> See http://pastie.org/622029.txt
We haven''t done any testing with Ubuntu, and past experience tells me
that they have some funky kernel patches that are interacting badly
with the Lustre code.  The only 2.6.27 kernel that we test is from
SLES11.
> The only message I see on the MDS and OSSes are:
> LustreError: 6028:0:(socklnd_cb.c:2228:ksocknal_recv_hello()) Unknown
> protocol version (2.x expected) from 10.2.1.176
> but that''s expected since the 1.8.1 client tries to use the new
> protocol, but then drops back down to the old one.
Right.
> I''m not sure if it''s related to Bug 19561
> Is there any other data that I can provide to help figure out
what''s going on?
Yes, it definitely looks like the same problem.  You could
try the SLES11 kernel (also 2.6.27) and see if that fixes
the problem for you.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.

On Fri, 2009-09-18 at 14:56 -0600, Andreas Dilger wrote:
> 
> We haven''t done any testing with Ubuntu,
That said, and to be sure, my individual use is not at all an
endorsement for Lustre on Ubuntu, but I''ve been building patchless
clients on Ubuntu kernels up to and including 2.6.24-19-generic.  I tend
to feel the latest Lustre releases might even support the 2.6.27 kernels
in Jaunty, but I have just not had the time to test it yet.
> and past experience tells me
> that they have some funky kernel patches that are interacting badly
> with the Lustre code.
Patchless client even?

b.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
Url :
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20090918/da1b5275/attachment.bin

On Sep 18, 2009  17:03 -0400, Brian J. Murrell wrote:
> That said, and to be sure, my individual use is not at all an
> endorsement for Lustre on Ubuntu, but I''ve been building patchless
> clients on Ubuntu kernels up to and including 2.6.24-19-generic.  I tend
> to feel the latest Lustre releases might even support the 2.6.27 kernels
> in Jaunty, but I have just not had the time to test it yet.
I thought your report in bug 19561 was specifically 2.6.27?
> > and past experience tells me
> > that they have some funky kernel patches that are interacting badly
> > with the Lustre code.
> 
> Patchless client even?
What I mean is UBUNTU patching their kernel, and causing grief for the
Lustre code, not Lustre patches...

That said, it could just be something that was backported to the
2.6.27.stable tree that is giving Lustre heartburn...

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.

On Fri, 2009-09-18 at 15:19 -0600, Andreas Dilger wrote:
> 
> I thought your report in bug 19561 was specifically 2.6.27?
Yeah, I forgot about that bug and the round of Ubuntu 2.6.27 testing I
did which revealed that.

b.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
Url :
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20090918/b8cc203a/attachment.bin

Actually - to be clear I downloaded the kernel from kernel.org and
just copied over the running config from Ubuntu before I compiled it:
cp /boot/config-2.6.28-15-server /usr/src/linux-2.6.27.34/.config
cd /usr/src/linux-2.6.27.34/
make menuconfig (then just save the config)
make-kpkg clean
make-kpkg --initrd  binary kernel_image

and then I just installed the newly created kernel debs and rebooted
to use them.

-Scott


On Fri, Sep 18, 2009 at 2:54 PM, Brian J. Murrell <Brian.Murrell at
sun.com> wrote:
> On Fri, 2009-09-18 at 15:19 -0600, Andreas Dilger wrote:
>>
>> I thought your report in bug 19561 was specifically 2.6.27?
>
> Yeah, I forgot about that bug and the round of Ubuntu 2.6.27 testing I
> did which revealed that.
>
> b.
>
>
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>
>