Le 19 déc. 2017 à 10:50, Esteban Lorenzano <estebanlm@gmail.com> a écrit :
but this is because we still need to stabilise the CI: what happens now is that we know some tests are failing randomly then we do not trust the CI to tell us everything is ok. Then we accept PRs in red state. Of course, this will fail eventually, and thatâs what we are seeing.
Most important thing, IMO is to fix the CI: those tests that fails randomly (I think most of them are networking problems) needs to be fixed or removed.
I think we should not rely on external services for our tests. We should mock every network service call to have a local version (still using the network). Then, you will not have failing tests because of timeout or network problem. With Zinc or Teapot itâs quite easy to set up a network service on a local port of the machine running tests.