On Mon, Jul 16, 2012 at 6:09 AM, lars hofhansl <[email protected]> wrote:
> It happened again, and two more runs with test failures.
> In seems that unless we can accept a build not created by the build machines, 
> but instead locally on my machine, we're at the moment unable to produce a 
> release.

I used to always make releases by building locally.  Its only in the
last six months or year that I figured I could take what was made by
jenkins.


> I can easily do that if people are willing to accept an artifact created that 
> way. Please let me know.
> I would also support the stance that we should not release another version 
> before we sorted that out. Opinions are welcome.
>

I think it'd be fine making the 0.94.1RC0.

> It's interesting that the security build has none of the problems. There at 
> least 1/2 of the builds succeed.
>

Yeah.  Thats interesting.  As Andrew speculates, its different order
of unit tests or extra friction makes tests pass.

>
> Do you know how to get access to the Jenkins machines?
>

It seemed "difficult" last time I asked:
https://issues.apache.org/jira/browse/INFRA-4055

> What we need to do is (1) analyze the tests and see why the fail/hang and (2) 
> what the heck is up with the Jenkins machines.
>

On 1., yeah, its an ongoing job that we've been slacking on.  We could
organize a swat team to dig in.    It'll take a bit of work.  Seems
like a bunch of flakey tests and then there are the times it passes
where a test is hung (but it looks like all passed).

> Jean-Marc said many tests fail as part of a bigger run and than pass when run 
> in isolation. That is still not acceptable, IMHO.
>

Agree.

St.Ack

Reply via email to