Lustre discuss - Jun 2007 - oops with Debian Sarge and Lustre 1.5.97

Hi all,

I have now tried to install Lustre following the report
https://mail.clusterfs.com/pipermail/lustre-discuss/2006-October/002311.html
It all works reasonably well until I try to access the Lustre FS  after
I mounted it on a client. A simple "ls" may hang immediately, it may
also work, as well as creation of files, reading, moving and deletion,
but at some point a command will hang and the kern.log says:

kernel: [95414.189214] kernel BUG at fs/dcache.c:862!
kernel: [95414.189229] invalid opcode: 0000 [#2]
kernel: [95414.189245] Modules linked in: osc mgc lustre lov lquota mdc
ksockl
nd ptlrpc obdclass lnet lvfs libcfs ide_generic ipv6 af_packet e1000
piix xfs evdev ide_cd ide_disk a
ta_piix libata tulip de4x5 hp100 dmfe floppy dm_mod ehci_hcd usbmouse
usbhid usbkbd usbcore rtc sr_mo
d cdrom sd_mod sg 3w_9xxx scsi_mod
kernel: [95414.189375] CPU:    0
kernel: [95414.189376] EIP:    0060:[<c017498a>]    Not tainted VLI
kernel: [95414.189377] EFLAGS: 00010246   (2.6.18 #1)
kernel: [95414.189428] EIP is at d_instantiate_unique+0x150/0x167
kernel: [95414.189447] eax: f43d4d14   ebx: f43d4d14   ecx: f43d4d14
edx: f3d481b8
kernel: [95414.189468] esi: f43d4d77   edi: f62f0af7   ebp: f6501748
esp: f6517af4
kernel: [95414.189487] ds: 007b   es: 007b   ss: 0068
kernel: [95414.189505] Process chown (pid: 31020, ti=f6516000
task=f63f2570 task.ti=f6516000)
kernel: [95414.189524] Stack: f43d4d14 f8f69180 f43d4d50 f62f0ad0
00000080 00010000 32b110c4 f62f0af0
kernel: [95414.189568]        00000007 00000000 f62f0a94 f6517f34
f6517c64 f9383de6 f62f0a94 f6501730
kernel: [95414.189609]        00000000 f6d60000 f9384011 f6501730
f62f0a94 00010000 f93a6a53 f939afb7
kernel: [95414.189652] Call Trace:
kernel: [95414.189680]  [<f9383de6>] ll_find_alias+0x1f/0x4b [lustre]
kernel: [95414.189749]  [<f9384011>] lookup_it_finish+0x1ff/0x5a3 [lustre]
...

Anybody seen this bug before?

The failed command, of course, cannot be killed. The Lustre FS may be
unmounted with "umount -lf" and mounted again, usable until the next
command hit this bug.



Btw, the code to set up the Debian package  that is contained in
pkg-lustre on svn.debian.org still uses the Lustre 1.5.97 sources.
However, the code for 1.6.0.1 has already been uploaded to that
repository - any plans to follow up with the Debian code soon?

Many thanks,


Thomas

Hi Thomas,
>
>
> Btw, the code to set up the Debian package  that is contained in
> pkg-lustre on svn.debian.org still uses the Lustre 1.5.97 sources.
> However, the code for 1.6.0.1 has already been uploaded to that
> repository - any plans to follow up with the Debian code soon?
I''ve CC''ed Goswin, I think he has new packages. In the mean
time I recommend
to install the utils and kernel modules seperately, so to keep the debian 
utils and only to update the kernel modules. 
If you want to keep linux-2.6.18 you have to wait for lustre-1.6.1 or you 
manually have to apply patches to be found in bug #11039 
(https://bugzilla.lustre.org/show_bug.cgi?id=11039).

If you can upgrade to 2.6.20, you may take our q-leap release. I''m not
sure
about its 2.6.18 support status, since we now only care about 2.6.20+.

http://www.pci.uni-heidelberg.de/tc/usr/bernd/downloads/lustre/1.6/


Cheers,
Bernd


-- 
Bernd Schubert
Q-Leap Networks GmbH

Thomas Roth <t.roth@gsi.de> writes:
> Btw, the code to set up the Debian package  that is contained in
> pkg-lustre on svn.debian.org still uses the Lustre 1.5.97 sources.
> However, the code for 1.6.0.1 has already been uploaded to that
> repository - any plans to follow up with the Debian code soon?
>
> Many thanks,
>
>
> Thomas
I commited some updates but a few kernel patch updates are out of sync
with our local lustre version so you will probably get some crashes. I
hope I can sync those tomorrow.

MfG
        Goswin

Goswin von Brederlow wrote:
> Thomas Roth <t.roth@gsi.de> writes:
> 
>> Btw, the code to set up the Debian package  that is contained in
>> pkg-lustre on svn.debian.org still uses the Lustre 1.5.97 sources.
>> However, the code for 1.6.0.1 has already been uploaded to that
>> repository - any plans to follow up with the Debian code soon?
>>
>> Many thanks,
>>
>>
>> Thomas
> 
> I commited some updates but a few kernel patch updates are out of sync
> with our local lustre version so you will probably get some crashes. I
> hope I can sync those tomorrow.
> 
> MfG
>         Goswin

Well, I got some. After changing a few ''1.5.97'' strings to
''1.6.0.1''
compilation and commenting out unavailable patches, compilation got as
far as
dh_install -p lustre-utils
which failed with
cp: cannot stat `./lustre-1.6.0.1/lnet/utils/debugctl'': No such file or
directory

A bit strange, though, as this ultility had been compiled, and without
problems the compiler messages would suggest. But it is not there in the
end to be put into the package. Some premature dh_clean somewhere?
Hm, at least I can check what has been built into the lustre-source and
linux-patch packages.

Regards,
Thomas

Hi Thomas,
> A bit strange, though, as this ultility had been compiled, and without
> problems the compiler messages would suggest. But it is not there in the
> end to be put into the package. Some premature dh_clean somewhere?
> Hm, at least I can check what has been built into the lustre-source and
> linux-patch packages.
we recently discovered disk corruption with our ldiskfs patch set, so you 
probably don''t want to use it in its present form. A newer patchset
also
caused problems. So you have two choices, either to remove extents and 
mballoc patches or to wait.

Cheers,
Bernd