[opam-devel] Travis CI failures

[Anil Madhavapeddy]

On 28 Sep 2014, at 16:38, Thomas Gazagnaire <thomas at gazagnaire.org> wrote:

> Hi all,
> 
> Recently I saw Travis CI to be quite unreliable. It took sometimes hours to start and very often I need to manually re-lauch some tests because of connectivity errors. I don't complain too loudly because it's a free (and very useful) service, but I am wondering if we should not upgrade to something a bit more reliable.
> 
> Looking at https://travis-ci.com/plans they seem to have few paying offer which might be interesting, although I am not sure to understand if this will solve the current issues. And it looks a little bit expensive, but we rely on the free service a lot already. 
> 
> Also, ideally we should use Travis CI to only coordinate the build on the Rackspace VMs but we don't have that infrastructure in place yet, so  maybe we should switch to a paying service for few months.

My understanding is that the paid Travis service doesn't entirely help with the existing issues, which are infrastructure overload on their part.  The paid service increases the number of parallel builds, but still (e.g.) enforces the per-job 50 minute limit.

I'll prioritise the Rackspace VM work to let us unlock more builds.  We're seeing an increasing number of repository failures due to the lack of reverse-transitive checking of builds, and that should be very easy to do in parallel on VM infrastructure.

(I may be wrong about the latest in paid Travis -- if anyone wants to enquire with them, please do let me know.  I'm happy to pay for the service if it would solve any problems.)

-anil