> On Jul 22, 2017, at 11:31 PM, Brahma Reddy Battula
> <brahmareddy.batt...@hotmail.com> wrote:
>
> AFAIK, Patch install(i mean,mvn install) will not run all the projects hence
> you didn't seen this failure.
Correct. By design, precommit only runs the test suite against the maven
modules that the patch modifies to save (significant!) processing time. The
number of patches that would benefit from touching the entire source base is
relatively small. But they do exist which is why the two follow up jobs are
important:
* the hadoop-trunk-commit, which writes the maven repo status at the
end of every JIRA. It builds (almost) everything but skips unit tests. It runs
every 20 minutes and only if there has been a commit.
* the daily qbt report which builds and runs EVERYTHING: all the
modules, all the native code supported by the docker-ized env, all the unit
tests. (For those curious, it currently clocks in at around 9.5 hours. That's
up by about 1.5 hours since I added it over a year ago.)
In this case, both of these failed and should have been an indication that
something was very, very wrong. Committers should be making it a priority to
watch for both of those reports to confirm that their commit didn't break
anything.
(I'm not sure if we have or ever had any documentation on what happens after a
commit. While these jobs have been rewritten, the process of patch verification
process from a committer view hasn't changed in probably five years or more.
The tribal knowledge is failing to get passed down from one generation to
another.)
---------------------------------------------------------------------
To unsubscribe, e-mail: common-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-dev-h...@hadoop.apache.org
