Suggestion: Continuous mass rebuild

[Peter Robinson]

On Mon, Jul 30, 2012 at 9:46 PM, Dan Horák <dan at danny.cz> wrote:
> Richard W.M. Jones píše v Ne 29. 07. 2012 v 18:38 +0100:
>> Currently we're doing a mass rebuild about every couple of releases,
>> ie. once a year.
>>
>> Since Dennis Gilmore has written this rebuild script already, why
>> don't we run the script more or less continuously?  Obviously we could
>> pace the builds so they happen for each package about once a month and
>> don't overload Koji.
>>
>> Then we track packages that don't build, say, 3 times in a row, and
>> file FTBFS bugs for them and after that prioritize fixing them or kick
>> them out of the distro.
>
> please keep in mind that the resources available to secondary arches are
> limited and can't undergo the load the mass rebuild generate. They can
> follow the normal load of builds coming from primary for rawhide and
> updates for 2 released versions, but such peaks are way above the
> capacity. What takes 4 days in primary koji with dozens of builders it
> takes 2 or 3 weeks in secondaries (ppc, s390) with other work mostly
> stopped ...

I agree with this as we experience this with ARM as well. That said
the proposed scratch rebuilding in parallel (without committing to
git) is something I applaud from a secondary arch PoV as a lot of the
build issues I experience on secondary are actually also FTBFS on
Primary when I test it further due to some other chance since the
build was done so this will get more attention to those issues and
help secondary arches greatly as it will be easy to tell if there's a
known problem on primary.

Peter