[LU-4665] utils: lfs setstripe to specify OSTs Created: 25/Feb/14  Updated: 06/Nov/18  Due: 31/Mar/16  Resolved: 20/Jan/15

Status: Resolved
Project: Lustre
Component/s: None
Affects Version/s: None
Fix Version/s: Lustre 2.7.0

Type: New Feature Priority: Critical
Reporter: Jinshan Xiong (Inactive) Assignee: Emoly Liu
Resolution: Fixed Votes: 1
Labels: llnl

Issue Links:
Related
is related to LU-3840 llapi_layout API design discussion Closed
is related to LU-6135 improved support for selecting specif... Resolved
is related to LU-5170 lfs usability Open
is related to LU-5878 lfs migrate sends file to stdout Closed
is related to LU-8417 setstripe -o does not work on directo... Open
is related to LU-9392 lfs migrate -o and lfs setstripe -o s... Open
Severity: 3
Rank (Obsolete): 12785

 Description   

#5: Sparse OST indicies
#6: Setstripe for specified OSTs.



 Comments   
Comment by Jinshan Xiong (Inactive) [ 25/Feb/14 ]

patch is at: http://review.whamcloud.com/#/c/9383

Comment by James A Simmons [ 25/Feb/14 ]

People have at ORNL have been asking for #6 for a very long time. You made many people happy.

Comment by Christopher Morrone [ 04/Mar/14 ]

Ditto. LLNL folks have been asking for that a long time as well.

Comment by Jian Yu [ 25/Mar/14 ]

Patch to improve Lustre test framework to support specifying arbitrary OST indices: http://review.whamcloud.com/9722
Patch to add regression test cases into conf-sanity.sh: http://review.whamcloud.com/9775

Comment by James A Simmons [ 07/Jul/14 ]

I like to suggest that we rebase this work on top of LU-2182. The LU-2182 work is in pretty good shape.

Comment by Christopher Morrone [ 07/Jul/14 ]

I agree with James.

Comment by James A Simmons [ 07/Jul/14 ]

If you don't have the time Jinshan I will gladly do the work.

Comment by Jinshan Xiong (Inactive) [ 08/Jul/14 ]

Hi James,

Please feel free to pick it up and move it forward. Thanks

Jinshan

Comment by Shuichi Ihara (Inactive) [ 08/Jul/14 ]

Is this still for regular file only? What's other porject or dependency to support specific OST selection for directory?

Comment by James A Simmons [ 08/Jul/14 ]

Started integrating this work with LU-2182.

Comment by Jinshan Xiong (Inactive) [ 05/Aug/14 ]

Hi James,

Are you still working on this? We're going to land this patch in 2.7.

Comment by James A Simmons [ 05/Aug/14 ]

Yes. In fact I'm debugging it right now.

Comment by Jinshan Xiong (Inactive) [ 05/Aug/14 ]

Is there any problems with the patch?

Comment by James A Simmons [ 08/Aug/14 ]

Not with the patch directly. I did see failures due to the new layout api using fsetxattr instead of the ioctl for setting stripes. There was code in ll_setxattr that if sent in stripe_offset was set to zero it was set to -1 which meant a random OST was picked instead of OST 0. The base LU-2182 patch which this work depends on should now behave properly. We will see once Maloo is done.

Comment by Peter Jones [ 09/Sep/14 ]

Emoly

Could you please assist with reviewing these patches?

Thanks

Peter

Comment by Oleg Drokin [ 01/Nov/14 ]

This feature totally needs a way to specify ost_count in addition to OST list that way we do not force user to do their own load balancing.
Say I have a set of 100 OSTs that are local enough for my set of nodes, but I only need a striping of 4...
We seriously should consider adding this as a follow on patch.

Comment by Oleg Drokin [ 01/Nov/14 ]

Ah, I also noticed that the current patch returns error if ldo_def_stripe_offset is not in the list of OSTs, but failed to consider that -1 should mean "use something that makes sense". This too needs to be addressed. Esp. in light of the previous comment

Comment by Andreas Dilger [ 07/Nov/14 ]

Is it possible to access this functionality via the new llapi_layout_* API added in LU-3840? It doesn't seem like it looking at the code. I've got a partially completed patch to add this, will link it here when I push the patch.

Comment by James A Simmons [ 07/Nov/14 ]

No. The llapi_layout api is not mature enough to be used. I tried using it as a back end but it just broke lots of test.

Comment by Andreas Dilger [ 10/Nov/14 ]

I've pushed http://review.whamcloud.com/12663 but it is a best-effort patch for now. It doesn't have a test case yet, and has had no testing at all.

Comment by Andreas Dilger [ 04/Dec/14 ]

Emoly, could you please refresh the 12663 patch according to the review comments, and add a test case for this to the llapi_layout_ test program.

Comment by Emoly Liu [ 05/Dec/14 ]

Andreas, sure.

Comment by Jinshan Xiong (Inactive) [ 16/Jan/15 ]

Do we need a new ticket # for http://review.whamcloud.com/#/c/12275?

Comment by Andreas Dilger [ 19/Jan/15 ]

I opened LU-6135 for the follow-on patches.

Comment by Andreas Dilger [ 20/Jan/15 ]

Have moved the other patches over to LU-6135.

Generated at Sat Feb 10 01:44:47 UTC 2024 using Jira 9.4.14#940014-sha1:734e6822bbf0d45eff9af51f82432957f73aa32c.