Early on in the project, we've discussed our CI needs and concluded to use ASF-hosted Jenkins as our preferred tool of choice. We've also enabled Travis-CI, which covered some scenarios that Jenkins couldn't do at the time, but with the idea to transition to Jenkins eventually.
Over the last few months, Travis-CI has been broken consistently, and several different kinds of infrastructure breakages have been added, one on top of another. This has caused plenty of cost and confusion. In particular, contributors often get confused as to which signal they should care about. At the same time, Jenkins capabilities have improved greatly: multiple parallel precommits are now supported, checked-in DSL support, pipelined matrix builds, Google's donation of Jenkins executors more than doubled, and others. So, based on the previous consensus and the fact the signal was broken for a long time, Jason and I went and asked Infra to disable Travis-CI on our code repository. (Website repository was disabled months ago.) I believe there should be minimal impact of this. The only two elements of the Travis matrix that were passing (still) are Python SDK on the Linux & Mac. Linux one can be trivially moved to Jenkins -- and I know Jason is looking at that. Mac coverage is the only loss at the moment, but is something we can likely address in the (near) future. I'm excited that we finally managed to unify our CI tooling, and can make efforts on improving and maintaining one system as opposed to two. That said, please comment if you have any worries about this or ideas for further CI improvements ;-) Davor
