Lustre discuss - Apr 2009 - Patched e2fsprogs source

I am trying to figure out an easy way (without turning on quota) the usage
of a Lustre file system. I was going to modify lfs find to print out the
size as well as the name and just run a lfs /mount find --user <username>
--print and get it to write out file sizes.

Is this the right track to go on? Or has someone already solved this
problem?

Thanks very much.
Suvendra.

On Apr 05, 2009  23:08 -0400, Suvendra Nath Dutta wrote:
> I am trying to figure out an easy way (without turning on quota) the usage
> of a Lustre file system. I was going to modify lfs find to print out the
> size as well as the name and just run a lfs /mount find --user
<username>
> --print and get it to write out file sizes.
Doing this will make "lfs find" no faster than just using
"find", so
it is fairly pointless.  Secondly, just using the file size is not
going to be accurate, because of sparse files.
> Is this the right track to go on? Or has someone already solved this
> problem?
I _think_ it is possible to enable Lustre quota w/o quota enforcement
(i.e. do not set any quota limits for users) and that would get you
what you want, but I''m not a Lustre quota expert and I could tell you
the commands to do this.

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

Hello Andreas,

Thank you for your response. This is exactly what we do on one of our file
systems. We are not able to do this on this other file system for the
following reason:
[root at iliadaccess04 ~]# lfs quotacheck -ug /n/scratch
quotacheck failed: Input/output error
[root at iliadaccess04 ~]# tail /var/log/messages
Apr  6 10:27:58 iliadaccess04 kernel: LustreError:
24683:0:(quota_check.c:246:lov_quota_check()) lov idx
 34 inactive
[root at iliadaccess04 ~]# lctl dl
  ...
 43 UP mgc MGC10.242.62.210 at tcp 1144636d-657d-d672-cb82-b7ed47117624 5
 44 UP lov aegalfs-clilov-ffff81042de30c00
6102ae5d-0599-15dd-7bfc-3d971d20b099 4
 45 UP mdc aegalfs-MDT0000-mdc-ffff81042de30c00
6102ae5d-0599-15dd-7bfc-3d971d20b099 5
 46 UP osc aegalfs-OST002a-osc-ffff81042de30c00
6102ae5d-0599-15dd-7bfc-3d971d20b099 5
 47 UP osc aegalfs-OST0000-osc-ffff81042de30c00
6102ae5d-0599-15dd-7bfc-3d971d20b099 5
 ...
 80 UP osc aegalfs-OST0021-osc-ffff81042de30c00
6102ae5d-0599-15dd-7bfc-3d971d20b099 5

For some reason when the file system was constructed that one OST was built
with a peculiar ID. And as far as we know, once an OST, always an OST. And
we figured that since this was a live file system, we were stuck with it
being for ever without quota. If there is a way around this, that would be
ideal!

Thanks.
Suvendra.



On 4/6/09 1:46 AM, "Andreas Dilger" <adilger at sun.com> wrote:
> On Apr 05, 2009  23:08 -0400, Suvendra Nath Dutta wrote:
>> I am trying to figure out an easy way (without turning on quota) the
usage
>> of a Lustre file system. I was going to modify lfs find to print out
the
>> size as well as the name and just run a lfs /mount find --user
<username>
>> --print and get it to write out file sizes.
> 
> Doing this will make "lfs find" no faster than just using
"find", so
> it is fairly pointless.  Secondly, just using the file size is not
> going to be accurate, because of sparse files.
> 
>> Is this the right track to go on? Or has someone already solved this
>> problem?
> 
> I _think_ it is possible to enable Lustre quota w/o quota enforcement
> (i.e. do not set any quota limits for users) and that would get you
> what you want, but I''m not a Lustre quota expert and I could tell
you
> the commands to do this.
> 
> Cheers, Andreas
> --
> Andreas Dilger
> Sr. Staff Engineer, Lustre Group
> Sun Microsystems of Canada, Inc.
>