Today I discovered that an old, old code change (circa 2012) caused
certain maven modules to be skipped during the precommit testing. That code
had been carried forward through all the rewrites, bug fixes, etc, over the
years likely because it seemed the correct thing to do. It is clearly not,
since at least hadoop-pipes was getting ignored. I have not evaluated the
impact of the bug on other parts of the Hadoop code base.
I’ve got sitting in the Hadoop beta test branch of Yetus what I think
is a potential fix to that bug. Initial testing shows all systems are go.
Additionally:
* After several weeks, Yetus being used for Hadoop Common has been more
or as stable as trunk’s, but testing significantly more parts.
* People are still reporting bugs on trunk’s test-patch that have been
fixed for months in Yetus.
* People are still confused as to which version is running where,
despite the email thread just a few days ago. (*exasperated sigh here*)
So I’m going to turn on Yetus for *ALL* Hadoop precommit jobs later
tonight. (Given how backed up Jenkins is at the moment, there is plenty of
time. haha) Anyway, if you see “Powered by Yetus” in the Hadoop QA posts,
you’ve got Yetus. If you don’t see it, it ran on trunk’s test-patch.
(… and now begin the threads on everyone freaking out and/or not
reading the above …)
