Brainstorming for UDS-Q - Build Prioritization

[Scott Ritchie]

On 4/13/12 7:22 AM, Scott Kitterman wrote:
> I won't be at UDS, so if this is of interest to someone, I would really like
> it if someone who is would take the ball and run with it.
>
> Currently build sequence on the buildds is determined (primarily) by a
> combination of pocket and component:
>
> https://help.launchpad.net/Packaging/BuildScores
>
> In my opinion, one of the major reasons to prioritize builds is to help keep
> ISOs buildable.  During non-freeze periods it is not at all rare for flavors
> built out of Universe to go days without a successful image build on some
> architectures due to archive skew.
>
> I think we should change the scoring model in Launchpad (yes, this would be an
> LP change, but I think it should be driven by Ubuntu) to include a
> seeded/unseeded element as well.  This would give us more images available for
> testing on a regular basis without having to get special respins.
>
> This could either be done by dropping the current component based score
> changes and substituting a seeded/unseeded differential or by adding an
> additional score differential for seeded-in-ubuntu packages on top of the
> existing model.
>
> Scott K
>

Archive skew is even more pernicious: it can make packages with 
cross-architecture dependencies (notably Wine) completely uninstallable. 
  Now that multiarch is a thing this will be an increasing problem for 
anyone testing a devel system with any foreign packages installed.

Could I propose, perhaps radically, that we instead consider not 
publishing packages for _any_ arch until they have zero build 
failures/in progress for all their build targets?

Less radically, might I propose we special case i386 and amd64 to work 
this way (ie, i386 failures/in progress block amd64 publishing)?

Thanks,
Scott Ritchie