Hi Arun, thanks for the reply. I hope that I've come across as an earnest
and motivated contributor who wants to help make high-quality releases.
Volunteering myself as RM and pushing for a 2.4 was not a reflection on
your stewardship of 2.x so far. I know it's a lot of work.

I didn't mean to suggest 12 releases per year. I was trying to couch my
terminology with "monthly-ish" and "regular cadence". I must have missed
the email thread where we decided on this, but 6-8 weeks per cycle is 100%
fine with me. I obviously don't want meaningless releases just for the sake
of releases, but I think HDFS-2832, HDFS-4949, and all the other
improvements in branch-2 add enough value that cutting something is not
meaningless.

My impression of the two YARN JIRAs is that they're still roughly a month
from being prod-ready, which is why I figured they could wait 6 weeks for
the next 2.x. That was also not a slight on the YARN developers, it's that
I think we should follow through if we're serious about a regular release
cadence. As you allude to, we need to get used to slipping features to the
next release. Symlinks one such example that I've personally been involved
with, so I do understand.

As an aside, I think this means that we should be more judicious about
branch-2 merges in the future, since a regular cadence means branch-2
should stay in a releasable state. I hope we can avoid burdonsome code
divergence through "hard-disable" patches (e.g. symlinks) or pushing down
just refactors (sort of like HDFS-2832), but that remains to be seen.

I can abandon the 2.3rc, and then release current branch-2 as 2.3. Would
> that be better?
>

That'd be great. All I'm after here is a regular release cadence with good
integration testing, so if you're willing to RM it, awesome. As I stated
way back in the thread, feel free to email me if you think I can help with
the release process.

Thanks,
Andrew

Reply via email to