[#LU-64] lfs getstripe output inconsistent for directory default value

[LU-64] lfs getstripe output inconsistent for directory default value Created: 07/Feb/11 Updated: 26/Nov/13 Resolved: 13/Jun/11
Status:	Resolved
Project:	Lustre
Component/s:	None
Affects Version/s:	Lustre 1.8.6
Fix Version/s:	Lustre 2.1.0, Lustre 1.8.7

Type:

Bug

Priority:

Minor

Reporter:

Prakash Surya (Inactive)

Assignee:

Lai Siyao

Resolution:

Fixed

Votes:

Labels:

None

Environment:

Linux x86_64

Issue Links:

Related
is related to	~~LU-995~~	1.8 <-> 2.1.54 ost-pools 24	Resolved

Severity:

Bugzilla ID:

23,802

Rank (Obsolete):

4990

Description

lfs getstripe is printing inconsistent values for default striping values for directories.

Our filesystem default values for striping are: count=2, size=1048576, offset=-1.

If I create a new directory and look at its stripe values I see:

$ mkdir testdir
$ lfs getstripe testdir
testdir
stripe_count: 2 stripe_size: 1048576 stripe_offset: -1

That is perfectly reasonable. But when I set any one of those values, it changes the way that the
other default values are printed:

$ lfs setstripe --count 4 testdir
$ lfs getstripe testdir
testdir
stripe_count: 4 stripe_size: 0 stripe_offset: -1

A user looks at that and says "oh no, it changed the size to zero!". Sure, essentially they are
the same, since 0 in the size field means "use the default".

If I recollect correctly from my experience with bug 19102 (and we are carrying the patch from that
bug), the problem is that when the EA is not set ("lfs setstripe -c 0 -s 0 -o -1" clears the EA),
it looks up the fs defaults and prints them. But when the EA IS set, the command is not looking up
the global defaults for the fields that mean "use the default".

This really needs to be consistent.

See Also: Bugzilla #23802 https://bugzilla.lustre.org/show_bug.cgi?id=23802

Comments

Comment by Prakash Surya (Inactive) [ 08/Feb/11 ]

Revised Patchset:
http://review.whamcloud.com/#change,299
http://review.whamcloud.com/#change,352

Comment by Peter Jones [ 09/Feb/11 ]

Lai will be working on this one

Comment by Peter Jones [ 14/Feb/11 ]

Lai,

I suggest that you get one of Oleg, Andreas or Johann to provide an additional inspection for the first patch.

Could you please clarify whether you are working on revising the second patch to align better with your expectations?

Regards

Peter

Comment by Lai Siyao [ 14/Feb/11 ]

Peter, okay, I will get Andreas involved; As for the second patch, I have commented there(which is needless IMHO), and will wait for the comments from Prakash and others. Cheers!

Comment by Build Master (Inactive) [ 04/Mar/11 ]

Integrated in reviews-centos5 #387
~~LU-64~~ Make "lfs getstripe" directory output consistent.

Prakash Surya : eccf32d8abb506951716f38f2d7eb4225d3cd19d
Files :

lustre/include/lustre/liblustreapi.h
lustre/utils/liblustreapi.c
lustre/doc/lfs.1
lustre/tests/sanity.sh
lustre/utils/lfs.c

Comment by Build Master (Inactive) [ 08/Mar/11 ]

Integrated in reviews-centos5 #412
~~LU-64~~ Make "lfs getstripe" directory output consistent.

Prakash Surya : 5c628406edee9c9824d42cf182107f71fb82a89b
Files :

lustre/include/lustre/liblustreapi.h
lustre/utils/lfs.c
lustre/tests/sanity.sh
lustre/doc/lfs.1
lustre/utils/liblustreapi.c

Comment by Christopher Morrone [ 14/Mar/11 ]

We need Andreas to respond to Prakash's response in:

http://review.whamcloud.com/299

Comment by Build Master (Inactive) [ 22/Mar/11 ]

Integrated in reviews-centos5 #534
~~LU-64~~ Modify the behavior of ioctl on directories without EA set.

Prakash Surya : c958dee42b89304b5f3d545e4109548b80f069b6
Files :

lustre/mds/handler.c
lustre/tests/sanity.sh
lustre/utils/liblustreapi.c

Comment by Prakash Surya (Inactive) [ 29/Mar/11 ]

The updated patch still needs to be reviewed: http://review.whamcloud.com/352

Comment by Peter Jones [ 04/Apr/11 ]

Prakash

Andreas has been on vacation and Johann is at a Linux filesystem conference this week. With LUG next week it will quite likely not be until the following week that they will get to this review. If this is a matter of operational urgency then let me know and I will reassign but otherwise I think that it would be best to wait for Johann and Andreas

Regards

Peter

Comment by Prakash Surya (Inactive) [ 04/Apr/11 ]

Peter, Thanks for the update.

Unless Chris tells me otherwise, that is not a problem. In the mean time I will work on rebasing this patchset onto the master branch.

Thanks!