On Tue, Mar 10, 2020 at 7:56 PM Olivier Lamy <[hidden email]> wrote:
> Thanks Robert
> Anyway as the build only fail for jdk7 with mvn 3.2.5 on a particular node
> of the asf machines.
> I build that perfectly with same jdk/mvn....
> So we cannot really consider this as blocker...
I absolutely do consider that a blocker. Even if it's something as
idiosyncratic as we need to turn one build machine off and on again,
this needs to be repaired before a release. Even when the failure mode
is not a true bug, a failing build trains developers to ignore
failures, which lets much more serious bugs creep in unnoticed.
If you disable one failing test because of a build machine issue, what
happens when that test starts failing everywhere because of an
I've been there before and it's not pretty. Reliable, non-flaky test
suites are critical for software development. CI systems should be as
automatic as possible, and should not be routed around.