On 2014-06-06 00:19:15 -0400 (-0400), Kevin Benton wrote:
> So if I'm understanding you correctly, a job can fail and will be
> held onto in case one of the parent jobs causes a reset in which
> case it will be retested?
Correct. If change E suffers a job failure and so does change B
ahead of it, then it's assumed that change B may be at fault and
change E gets retested without B. That was one of the primary design
requirements (and arguably THE defining characteristic) of Zuul's
speculative parallelization.
http://ci.openstack.org/zuul/gating.html#testing-in-parallel
http://docs.openstack.org/infra/publications/zuul/#(18)
--
Jeremy Stanley