[cdwg] [wc-discuss] Lustre 2.2.0 released

[Andreas Dilger]

On 2012-06-18, at 9:26 AM, James A Simmons wrote:
> Seeing this email and from the working groups phone call shows a lot of
> confusion over what is supported or maintained. So for the last few days
> I stepped backed and asked why is everyone confused. What this shows is
> lack of explaining to the community what a maintenance and a feature
> release are and how they differ. Also the version scheme is not well
> defined. 

Last week we also updated the download site with an explanation of the difference between a feature branch and a maintenance branch, but I agree we could do more to make this distinction more clear.

> 	First lets define a maintenance branch. The maintenance branch is
> defines as the branch where no new features are added during its life
> time. A features branch is one that new improvements are introduced for
> validation but will not be supported for general production use and bug
> fixes will not be incorporated.
> 	Currently for the 2.X branches we have 
> 
> 2.1.X - maintenance branch
> 2.2   - feature branch
> 2.3   - feature branch
> 
> 	So here is my proposal to end this confusion. For maintenance branches
> we start a new number i.e for the next one it would be 3.0.0.
> At the same time another branch, 3.1 be open for more features. In other
> words 3.0.0 and 3.1 would be branched off the same root branch but
> evolve in different ways. Yes for 2.X is a bit different due changes in
> companies with 2.0 and 2.1.
> 
> 3.0.0   - maintenance branch
>  3.0.1
>  3.0.X
> 3.1.0   - features branch
>  3.1.1
>  3.1.X
> 3.2.0   - features branch
>  3.2.1
>  3.2.X
> 4.0.0   - maintenance branch 
> 
> 	I just hope that the 3.0 branch would open up after the 2.3.

I don't think that changing the numbering will make this any more clear.  To be honest, I suspect people would think, for example, that the "3.0" release is less stable compared to 3.1.  Also, it isn't clear what the distinction in the above numbering is between 3.0.x and 3.1.x?  If 3.1.x is considered a maintenance release for the features of 3.1.0, then the distinction between a maintenance branch and a features branch is again meaningless.

A better scheme (which matches the above and the current practice) might be that 3-digit release numbers are maintenance releases, and 2-digit release numbers are feature releases.  However, what will make the distinction between a maintenance release and a feature release most clear to users is to just state this explicitly in every release announcement, and on the download pages.

I agree with Chris that having some advance discussion and agreement from major Lustre sites/users on what should be the next maintenance release is important.  While everyone can provide input on this decision, the most important factor is who will be doing the most testing and provide good bug reports.  The other issue is that once a branch is chosen for maintenance release, the users need to understand that this will get only bug fixes and updates to track the supported vendor kernels, while new features need to go into the next feature release.

Sometimes it may be that users feel they _have_ to use a feature release at some site for some pressing reason (e.g. must-have feature or performance improvement).  If it is understood that the path forward for that branch is to move to the next maintenance release when it is available (within 12 months, given that the feature release was made at least 6 months after the previous maintenance release), then that site will still converge on the next maintenance release for long-term support.  If they are willing to use a new feature release, they should also be willing to upgrade to the next stable maintenance release as well (possibly after the x.y.1 or x.y.2 bugfix release is made).

Cheers, Andreas
--
Andreas Dilger                       Whamcloud, Inc.
Principal Lustre Engineer            http://www.whamcloud.com/