CI

Some slaves run CentOS other run Ubuntu and they are randomly picked when running make check on pull requests. This introduces a random factor that makes it very difficult to figure out if a test is stable. It may fail because of an operating system difference of because there is a bug, there is no way to tell. A stable environment is necessary to run and diagnose tests.

It is a good idea to run the tests on various operating systems, but it must be done in a way that is not random. For instance a test can be run on CentOS first and if it succeeds it can then be run on other operating systems.

We are working hard to get a stable make check and this newly introduced randomness may be the cause of a recent mysterious instability reported at http://tracker.ceph.com/issues/18018 ... or not. We have no way to know for sure until the random selection of operating systems has been removed from the equation.