reef-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Boris Shulman <shulm...@gmail.com>
Subject Re: [VOTE] Release Apache REEF 0.15.0 (rc2)
Date Tue, 24 May 2016 16:27:30 GMT
I would prefer to go on with this release and augmenting it as Markus
suggested. I can be the release manager for the next one (0.15.1) as well.
Should we make 0.15.1 off the 0.15.0 branch and not master?

Another thought is why we rely on the manual testing when we have
completely automated tests on AppVeyor and Travis? I would prefer to rely
more on teh automated testing vs. manual testing on unstable environments.


Boris.

On Tue, May 24, 2016 at 9:06 AM, Markus Weimer <markus@weimo.de> wrote:

> On 2016-05-24 07:16, Yunseong Lee wrote:
>
>> However, I'm a bit concerned that some users may encounter the same
>> problem
>> with the release.
>>
>
> Agreed. I am really conflicted on this release, which is expressed in my
> indecisive `+0` vote. Here is why I think we should release:
>
>   * Since 0.14, a ton of bugs in REEF.NET have been fixed. The release
>     is overall better than any we had before.
>   * The very existence of the tests that now sometimes fail is a
>     massive improvement.
>   * Some of those fixes are crucial to users of REEF.NET who'd like
>     to see them in a release ASAP.
>   * Because of the way the .NET integration tests determine success
>     (timers and file-based synchronization), I don't trust their
>     reports. Neither positive nor negative. So a test failure might not
>     indicate an actual bug, but a bug in the test. This is a terrible
>     situation which we should fix.
>
> Here is why I think we should not release:
>
>   * Failing tests in a release isn't what the Apache Foundation stands
>     for. We make quality software that people can depend on. And a
>     flaky test in a release doesn't exactly scream "quality".
>   * This release vote has seen very few votes. So we don't exactly know
>     how flaky it really is.
>
> That said, the release meets all the paperwork requirements of an Apache
> release (licenses are in order, vote passed) and can go out. We might
> consider augmenting the release notes with "known issues" and a quick
> follow-up release 0.15.1 which rectifies the testing situation.
>
> Markus
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message