Lustre discuss - May 2006 - How to restrict a stripe to a subset of the available OSTs?

Lee Ward wrote:
> Hi,
> 
> Is there any way for an application to set which particular OSTs should
> be involved in a stripe? I''ve found how to set size, count, and
offset.
> It appears, though, that the ll_lov_setstripe routine in
> lustre/llite/file.c will only copy in the bytes for the lov_user_md
> structure and not any lov_user_ost_data records. Am I reading the code
> correctly? Is there, perhaps, some other way to accomplish this?
> 
> Thanks in advance.
> 
In userspace, ''lfs setstripe  <filename|dirname> 
<stripe-size>
<start-ost> <stripe-cnt>''
will do what you want. That routine uses
llapi_file_create() from lustre/utils/liblustreapi.c

Hope this helps
cliffw
> 		--Lee 
> 
> 
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss@clusterfs.com
> https://mail.clusterfs.com/mailman/listinfo/lustre-discuss

LOL. You read the subject but not the body of the letter, didn''t you?

		--Lee


On Thu, 2006-04-20 at 09:34 -0700, Cliff White wrote:
> Lee Ward wrote:
> > Hi,
> > 
> > Is there any way for an application to set which particular OSTs
should
> > be involved in a stripe? I''ve found how to set size, count,
and offset.
> > It appears, though, that the ll_lov_setstripe routine in
> > lustre/llite/file.c will only copy in the bytes for the lov_user_md
> > structure and not any lov_user_ost_data records. Am I reading the code
> > correctly? Is there, perhaps, some other way to accomplish this?
> > 
> > Thanks in advance.
> > 
> In userspace, ''lfs setstripe  <filename|dirname> 
<stripe-size>
> <start-ost> <stripe-cnt>''
> will do what you want. That routine uses
> llapi_file_create() from lustre/utils/liblustreapi.c
> 
> Hope this helps
> cliffw
> 
> > 		--Lee 
> 
> > 
> > 
> > _______________________________________________
> > Lustre-discuss mailing list
> > Lustre-discuss@clusterfs.com
> > https://mail.clusterfs.com/mailman/listinfo/lustre-discuss
> 
>

> Date: Thu, 20 Apr 2006 10:27:08 -0600
> From: Lee Ward <lee@sandia.gov>
> Subject: [Lustre-discuss] How to restrict a stripe to a subset of the
> 	available OSTs?
> To: lustre-discuss@clusterfs.com
>
> Hi,
>
> Is there any way for an application to set which particular OSTs  
> should
> be involved in a stripe? I''ve found how to set size, count, and  
> offset.
> It appears, though, that the ll_lov_setstripe routine in
> lustre/llite/file.c will only copy in the bytes for the lov_user_md
> structure and not any lov_user_ost_data records. Am I reading the code
> correctly? Is there, perhaps, some other way to accomplish this?
>
> Thanks in advance.
Lee,

This is not currently supported by Lustre, as you determined through  
the thread with Cliff yesterday. It was, however, the subject of some  
interesting discussion at the LUG yesterday. We discussed with  
participants the idea of a "Pools" feature which, among other things,
will be used to specify a collection of OSTs to be involved in a stripe.

The voting process we went through for future features actually  
ranked OST/stripe management as #2 on the priority list. Peter Braam  
advised that there are some short-term and some longer term areas in  
which we could address this requirement (in order): 1) Pools; 2) Join  
files; 3) background migration.

All of these are already represented in the current Lustre roadmap,  
but the delivery may be moved up in response to community input.  
There is also the possibility that another member of the Lustre  
community may step up and volunteer to implement one of these features.

Cheers,
Peter

Lee Ward wrote:
> LOL. You read the subject but not the body of the letter, didn''t
you?
> 
No, actually I read ''start'' instead of
''involved'' , my bad.
Sorry.
cliffw
> 		--Lee
> 
> 
> On Thu, 2006-04-20 at 09:34 -0700, Cliff White wrote:
> 
>>Lee Ward wrote:
>>
>>>Hi,
>>>
>>>Is there any way for an application to set which particular OSTs
should
>>>be involved in a stripe? I''ve found how to set size, count,
and offset.
>>>It appears, though, that the ll_lov_setstripe routine in
>>>lustre/llite/file.c will only copy in the bytes for the lov_user_md
>>>structure and not any lov_user_ost_data records. Am I reading the
code
>>>correctly? Is there, perhaps, some other way to accomplish this?
>>>
>>>Thanks in advance.
>>>
>>
>>In userspace, ''lfs setstripe  <filename|dirname> 
<stripe-size>
>><start-ost> <stripe-cnt>''
>>will do what you want. That routine uses
>>llapi_file_create() from lustre/utils/liblustreapi.c
>>
>>Hope this helps
>>cliffw
>>
>>
>>>		--Lee 
>>
>>>
>>>_______________________________________________
>>>Lustre-discuss mailing list
>>>Lustre-discuss@clusterfs.com
>>>https://mail.clusterfs.com/mailman/listinfo/lustre-discuss
>>
>>
> 
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss@clusterfs.com
> https://mail.clusterfs.com/mailman/listinfo/lustre-discuss

Thank Peter. Appreciate the status and discussion.

		--Lee

On Fri, 2006-04-21 at 08:23 -0400, Peter Bojanic wrote:
> > Date: Thu, 20 Apr 2006 10:27:08 -0600
> > From: Lee Ward <lee@sandia.gov>
> > Subject: [Lustre-discuss] How to restrict a stripe to a subset of the
> > 	available OSTs?
> > To: lustre-discuss@clusterfs.com
> >
> > Hi,
> >
> > Is there any way for an application to set which particular OSTs  
> > should
> > be involved in a stripe? I''ve found how to set size, count,
and
> > offset.
> > It appears, though, that the ll_lov_setstripe routine in
> > lustre/llite/file.c will only copy in the bytes for the lov_user_md
> > structure and not any lov_user_ost_data records. Am I reading the code
> > correctly? Is there, perhaps, some other way to accomplish this?
> >
> > Thanks in advance.
> 
> Lee,
> 
> This is not currently supported by Lustre, as you determined through  
> the thread with Cliff yesterday. It was, however, the subject of some  
> interesting discussion at the LUG yesterday. We discussed with  
> participants the idea of a "Pools" feature which, among other
things,
> will be used to specify a collection of OSTs to be involved in a stripe.
> 
> The voting process we went through for future features actually  
> ranked OST/stripe management as #2 on the priority list. Peter Braam  
> advised that there are some short-term and some longer term areas in  
> which we could address this requirement (in order): 1) Pools; 2) Join  
> files; 3) background migration.
> 
> All of these are already represented in the current Lustre roadmap,  
> but the delivery may be moved up in response to community input.  
> There is also the possibility that another member of the Lustre  
> community may step up and volunteer to implement one of these features.
> 
> Cheers,
> Peter
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss@clusterfs.com
> https://mail.clusterfs.com/mailman/listinfo/lustre-discuss
>

Hi,

Is there any way for an application to set which particular OSTs should
be involved in a stripe? I''ve found how to set size, count, and offset.
It appears, though, that the ll_lov_setstripe routine in
lustre/llite/file.c will only copy in the bytes for the lov_user_md
structure and not any lov_user_ost_data records. Am I reading the code
correctly? Is there, perhaps, some other way to accomplish this?

Thanks in advance.

		--Lee