[Buildbot-devel] RFC: using buildbot to manage crowd-sourced testing.

Dustin J. Mitchell dustin at v.igoro.us
Sat Feb 16 22:02:01 UTC 2013


Yes, this sounds great.  In particular, I think volunteer computing
could be a real strength for Buildbot.

You've suggested a bunch of pretty significant changes to Buildbot's
model, though:
 * slaves pulling work rather than the master pushing it
 * dynamic addition of slaves
 * cryptographic authentication (as Andrii suggests)
 * slave configuration checks

It would also be great to have some way for a single volunteer compute
resource to offer its services to multiple masters.  Maybe this even
occurs through some kind of cloud of "brokers" which can match slave
capabilities and configuration to registered masters.  This starts to
sound more and more like BOINC.

These changes will come with some difficult compatibility challenges
and security concerns, as well as challenging implementation.

Historically, we've failed our GSoC students by asking them to design
something, particularly something new and different.  Students lack
the familiarity with Buildbot, and the deep background in software
development, to be able to create a practical design.  It's also worth
remembering that GSoC is about attracting and retaining contributors,
not about cheap labor.  In almost any case, it will be more efficient
to implement something ourselves than mentor a student's
implementation.

I don't think any of the current core contributors will have the
resources to focus on this, particularly with the work on nine
currently underway.

So, I think the way forward here is for people from the Buildbot
community to do the initial design and plot out the implementation
process.  Then bits of that process can be broken out as GSoC tasks,
as long as someone is wiling to mentor for the task.

Again, this is a great idea, and I hope it proceeds.  I'm only
outlining what I see as the most likely conditions for success.

Dustin




More information about the devel mailing list