On 4.2.15 10:03 , Davide Giannella wrote:
We already split the profiles on travis and we can do the same apache
buildbot. The point is that our tests will have to be instructed to be
so. That's where the FixturesHelper come in
(https://issues.apache.org/jira/browse/OAK-1855)
For some reason I had problems with this approach on the Apache
buildbot. First injecting the environment variables through the
configuration didn't work due to a platform issues
(https://issues.apache.org/jira/browse/BUILDS-48). Then the
Assume.assumeTrue() logic didn't catch for whatever reason.
In
https://github.com/apache/jackrabbit-oak/blob/trunk/oak-jcr/src/test/java/org/apache/jackrabbit/oak/jcr/AtomicCounterIT.java
a mixed approach where I instructed the test to run only if the profile
is there and only on the Segment fixture specifically (in case of
multiple fixtures where specified).
I can take care of creating new profiles on travis, one per each
persistence, it should be trivial but then all of us will have to act on
each test instructing them to behave accordingly.
For the apache buildbot it's a bit longer as I didn't look into it
deeply yet.
We also have https://analysis.apache.org/jenkins/job/oak/
I think we should consolidate our CI builds to a single platform. As
otherwise we have to implement individual workarounds for the cartesian
product of the issues of all platforms.
In the end we need to improve the signal to noise ratio. We can't blame
people for ignoring failing CI builds if we switch of email
notifications due to too much noise be spurious failures.
Michael
