Lustre devel - Dec 2006 - [Bug 11486] Improve message: OBD class driver Build Version: 1.4.6.8...

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=11486

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED


This message is printed by the obdclass module when it is first loaded.  It is
the closest thing we have to a Lustre startup message.

My thoughts on this message:

1. If we consider it a startup message, it should really appear when any part of
Lustre is loaded.  Right now, a node that is only an lnet router will not print
this message.  Since the "root" of the Lustre module dependency tree
is libcfs,
I propose moving it there.

2. The version printed (BUILD_VERSION) is too confusing and unlikely to be
useful to anyone.  I propose using the standard Lustre version, in this case
1.4.6.8, instead.  Also, we should eliminate BUILD_VERSION from the few places
it currently exists.

3. Of course, the message needs to follow our standard format (currently a work
in progress) and include a message ID.

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=11486



Created an attachment (id=9237)
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
 --> (https://bugzilla.lustre.org/attachment.cgi?id=9237&action=view)
Move startup message to libcfs; use LUSTRE_VERSION_STRING instead of
BUILD_VERSION.

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=11486

           What    |Removed                     |Added
----------------------------------------------------------------------------
Attachment #9237 is|0                           |1
           obsolete|                            |


Created an attachment (id=9238)
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
 --> (https://bugzilla.lustre.org/attachment.cgi?id=9238&action=view)
Move startup message to libcfs; use LUSTRE_VERSION_STRING instead of
BUILD_VERSION.

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=11486



(In reply to comment #6)
> (This is a diff from b1_5, not b1_4)
Yes.  All new work needs to be done on b1_5 first and then merged back if
appropriate.
> In obd_proc_read_version, is there a need to remove the BUILD_VERSION?	
It''s not strictly necessary, but I''d like BUILD_VERSION to die
and some of the
rest of the patch removes the code that produces it.  Do you think it''s
useful?
 AFAIK it''s never been useful to the support team (but I''ll
ask the rest of them
to be sure.)
> init_obdclass uses CDEBUG for the liblustre case - does init_libcfs_module
need
> to as well?
If you look at the RC check just below my new printk(), printk() is used there
to print an error if necessary.  Therefore I believe printk() is fine anywhere
in init_libcfs_module().

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=11486



This comment may be out of line, but I''m not sure it''s a good
idea to stop using
BUILD_VERSION.  While very verbose & sometimes confusing, we very frequently
build multiple versions of lustre from the same source base.  These mostly vary
in small ways, which patches for bugfixes are added, etc.  In such cases the
LUSTRE_VERSION_STRING doesn''t change & the BUILD_VERSION appearing
at runtime in
server logs in startup msgs is the only way to identify precisely what is
running.

Absolutely not out of line; we''re sending all this stuff to
lustre-devel
expressly
so that anyone can comment.

bogl@cray.com wrote:
> Please don''t reply to lustre-devel. Instead, comment in Bugzilla
by using the following link:
> https://bugzilla.lustre.org/show_bug.cgi?id=11486
>
>
>
> This comment may be out of line, but I''m not sure it''s a
good idea to stop using
> BUILD_VERSION.  While very verbose & sometimes confusing, we very
frequently
> build multiple versions of lustre from the same source base.  These mostly
vary
> in small ways, which patches for bugfixes are added, etc.  In such cases
the
> LUSTRE_VERSION_STRING doesn''t change & the BUILD_VERSION
appearing at runtime in
> server logs in startup msgs is the only way to identify precisely what is
running.
>
> _______________________________________________
> Lustre-devel mailing list
> Lustre-devel@clusterfs.com
> https://mail.clusterfs.com/mailman/listinfo/lustre-devel
>
>

I think that when (sic) we switch to an scm system that has absolute, 
easily identifiable versions, such as svn, we can omit this.  As long as 
multiple builds of things with the same tag (or lack thereof) are 
possible this tag isn''t so helpful. 

We nearly have automatic builds going - I believe that these build 
branches not tags and that we need some kind of identifier.  Andy - is 
that right?

- Peter -

Nathaniel Rutman wrote:
> Absolutely not out of line; we''re sending all this stuff to 
> lustre-devel expressly
> so that anyone can comment.
>
> bogl@cray.com wrote:
>> Please don''t reply to lustre-devel. Instead, comment in
Bugzilla by
>> using the following link:
>> https://bugzilla.lustre.org/show_bug.cgi?id=11486
>>
>>
>>
>> This comment may be out of line, but I''m not sure
it''s a good idea to
>> stop using
>> BUILD_VERSION.  While very verbose & sometimes confusing, we very 
>> frequently
>> build multiple versions of lustre from the same source base.  These 
>> mostly vary
>> in small ways, which patches for bugfixes are added, etc.  In such 
>> cases the
>> LUSTRE_VERSION_STRING doesn''t change & the BUILD_VERSION
appearing at
>> runtime in
>> server logs in startup msgs is the only way to identify precisely 
>> what is running.
>>
>> _______________________________________________
>> Lustre-devel mailing list
>> Lustre-devel@clusterfs.com
>> https://mail.clusterfs.com/mailman/listinfo/lustre-devel
>>
>>   
>
> _______________________________________________
> Lustre-devel mailing list
> Lustre-devel@clusterfs.com
> https://mail.clusterfs.com/mailman/listinfo/lustre-devel
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://mail.clusterfs.com/pipermail/lustre-devel/attachments/20070102/6b18bf63/attachment.html

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=11486



(In reply to comment #11)
> I think bogl''s comment #9 is a good enough reason to leave the
build ver, and I
> don''t see a very strong argument for getting rid of it.  I think
it should only
> show up in the proc and at module load time, be clearly labelled as the
build
> ver and not the lustre ver, and I don''t think anyone will be
especially confused
> by it.  My two cents.      
A customer _has_ expressed confusion with this message, which is the reason
I''m
addressing it.

Perhaps including the Lustre version as well, plus clearly marking this version
as a build version, will make it clear enough.  I''ll prepare a revised
patch for
comment.

The new build system can do a build given either a branch or a tag, like the
previous system.  But moving to a model where we tag regular "builds"
might
address the issue at hand, so a specific build might be called something
like v1_4_8_build2 and there would be a tracker bug which shows exactly
what was included in 1.4.8 build 2.

-andy

Peter Braam wrote:
> I think that when (sic) we switch to an scm system that has absolute, 
> easily identifiable versions, such as svn, we can omit this.  As long as 
> multiple builds of things with the same tag (or lack thereof) are 
> possible this tag isn''t so helpful. 
> 
> We nearly have automatic builds going - I believe that these build 
> branches not tags and that we need some kind of identifier.  Andy - is 
> that right?
> 
> - Peter -
> 
> Nathaniel Rutman wrote:
>> Absolutely not out of line; we''re sending all this stuff to 
>> lustre-devel expressly
>> so that anyone can comment.
>>
>> bogl@cray.com wrote:
>>> Please don''t reply to lustre-devel. Instead, comment in
Bugzilla by
>>> using the following link:
>>> https://bugzilla.lustre.org/show_bug.cgi?id=11486
>>>
>>>
>>>
>>> This comment may be out of line, but I''m not sure
it''s a good idea to
>>> stop using
>>> BUILD_VERSION.  While very verbose & sometimes confusing, we
very
>>> frequently
>>> build multiple versions of lustre from the same source base.  These
>>> mostly vary
>>> in small ways, which patches for bugfixes are added, etc.  In such 
>>> cases the
>>> LUSTRE_VERSION_STRING doesn''t change & the
BUILD_VERSION appearing at
>>> runtime in
>>> server logs in startup msgs is the only way to identify precisely 
>>> what is running.
>>>
>>> _______________________________________________
>>> Lustre-devel mailing list
>>> Lustre-devel@clusterfs.com
>>> https://mail.clusterfs.com/mailman/listinfo/lustre-devel
>>>
>>>   
>>
>> _______________________________________________
>> Lustre-devel mailing list
>> Lustre-devel@clusterfs.com
>> https://mail.clusterfs.com/mailman/listinfo/lustre-devel