On 12/30/2011 04:25 PM, Johnny Hughes wrote:
> We have a custom build system.  It uses a beanstalkd workqueue to submit
> packages to mock and can scale up or down as required and submit

Just to add a few more bits : the queue system is meant to extend in
both directions ( import and tracking of sources on one end - and
testing + release management at the other end ). At the moment there is
a bunch of band-aid and Johnny and me typing up stuff to bridge those
bits, but I hope to have the entire setup in place end to end by end of
March 2012. Apart from the cli, some of the other interesting interfaces
that are semi functional at this point include an irc-bot, a rest api
and a publicly available http callback server that allows arbitrary
subscription.

There is, by design, no web interface. The plan is to focus on the
mechanics and make the entire system work, with enough hooks that allows
anyone / everyone to do the client side work. Eg. being able to write a
yum plugin that would harvest update metadata from the buildsystem and
be able to report on 'upcoming updates' ( a PoC for this is functional
now.. )

The entire system is called 'reimzul'

As soon as I have got the basic functional stuff in place, and some docs
around it I'll have it up in a git repo where people can clone and join
the fun. 

And yes, this is very  custom done for the CentOS ecosystem and
address's the problems and hurdles that we run into with the localised
process. Koji is a great system, just like OBS is - but they solve
different problems and do so in a way that makes both of them really bad
solutions for the distro-rebuild and tracking process's.

In other interesting artifacts ( and mostly from free win's ) include
being able to test external third party repo's for pre-updates-release
breakage's.... Another feature is the constant compose ( the
next-release-tree is constantly composed on every update build, so were
pretty much ready to go for the next release every night.. a feature
that came in handy for the 6.2 release.)

again, just to be clear - this system works for c6 now, and will for c7
etc ( unless we come up with something else ). We are very much
committed to using plague for c4/c5 for the time being. The pain
associated with migrating and then proving the system for c5 would be
more than is justifiable at this point.

Bit of a brain dump, but there you have it.