[Mono-dev] Continuous Integration for Mono

[David Schmitt]

On 24/06/11 23:37, Jonathan Chambers wrote:
> All,
>
> I'm composing this after a few discussions in the #monodev irc channel
> and this recent thread: http://go-mono.com/forums/#nabble-f1517221
>
> A continuous integration server for mono is required IMO, so I'm going
> to put out a few options and see what the community thinks.
>
> 1. Hosting everything - If the community is most willing to support
> monetarily, the entire build farm can be hosted at any number of
> providers. Someone of course still needs to maintain the CI system, but
> there would be no direct cost for hardware or a physical location for
> the machines (up front or long term). Also, there is the question of
> platforms for which hosting is not readily available.
>
> 2. Hosting the CI server - In this case, I envision a hosted CI
> coordinator machine that can manage and collate build/test results. The
> actual build machines could be hosted, or they could be provided by
> community at large and just directed by the coordinator. There's some
> hand waving here, but I think we could remotely execute the builds and
> make the results publicly available on the CI coordinator.
>
> 3. No hosting - Potentially, the CI system and the build machines could
> be located at some mono friendly organization or company. This obviously
> requires machines and a physical location to house them. Obviously,
> someone needs to maintain the hardware in addition to the CI system itself.
>
> My suggestion is to host at least the CI server at some provider, and
> potentially host the platforms for which hosting is readily available
> (Windows/Linux x86/amd64). Some platforms/OSs such as ARM, PowerPC,
> XBOX, PS3, etc. may not be possible to host. I'm guessing orgs with the
> most interest in those platforms will help in maintaining those
> platforms, or help in providing machines that can be used.

Option #2 seems sensible to me. The committers need control over the 
infrastructure, but the do not need to take care for each and every system.

> Further, I'd suggest using as much 'off the shelf' software as possible
> to allow mono community to focus more on mono and less on build
> infrastructure. Unless we need a custom CI system, I propose we at least
> consider TeamCity. A number of contributors to mono use it internally
> with success.

Is there a reason not to continue using monkeywrench? Perhaps enhanced 
with a little bit of gerrit?

> 1. Any estimates on how much money mono community could collect yearly
> for hosting costs/etc?

Potent build-hosts in "the cloud" can be had for 50€/month. I cannot 
imagine that money will be a problem.

> 2. What platforms *require* CI support? Is there an order of importance
> for platform to integrate into CI system?

I guess platforms should be integrated as they become available. The 
more interesting question is which failures are deemed critical by the 
committers ;-)

> 3. How/Where will *official* releases be built? How/where do we expect
> those binaries to be hosted?

I'm particularly partial to "native" repositories. yum/apt/ppa repos 
really reduce the friction to get code out to people to test stuff. Of 
course my opinion is heavily biased towards server usage.



Best Regards, David