1:16 p.m.
Hi everyone,
After spending the last year plus in a feature-boxed release scheduling
mode, I think it would be both good and feasible to get the WildFly project
back into the roughly time-boxed delivery model that we managed pretty
successfully starting with WildFly 12. This thread is to gather inputs from
the WildFly developer community about doing that over the next year. After
discussing here, my aim is to post our plans to the broader WildFly
community via a wildfly.org news post and a thread on the wildfly google
group.
One question is what length of time boxes we would shoot for.
Personally I thought the roughly quarterly cadence worked well. It seemed
short enough that we didn't find ourselves regularly stressing about
getting a particular feature into a particular release because it would
otherwise have to wait too long. I could imagine trying to go for a shorter
cycle, but I don't think that's really practical right now. Optimizing our
delivery processes to make that more practical would be a good goal for the
next year though.
If we go back to quarterly releases, the next question is what that
schedule would look like. If we assume WildFly 28 goes out as planned[1],
then a quarterly cadence would look like this:
WildFly 29 Beta - Thur Jun 29
WildFly 29 Final – Thur Jul 13
WildFly 30 Beta – Thur Sep 28
WildFly 30 Final – Thur Oct 12
WildFly 31 Beta – Thur Dec 14 [2]
WildFly 31 Final – Thur Jan 11
WildFly 32 Beta – Thur Mar 28
WildFly 32 Final -- Thur Apr 11
Those dates are the Thursdays when the release is announced on and download
links are added to wildfly.org. We'd want PRs ready intended for the
release to be ready to merge 6 days before that date, i.e. the preceding
Friday.
In the past we tried for a March/June/September/December cadence, but,
besides that fact the April WildFly 28 will become a starting point, my
sense is January/April/July/October will work better. A lot of developers
go on holidays in late July and August, which tended to make September
releases difficult. And then aiming for a mid-December Final release was
often a problem, because any delay would push the release into the late
December period when folks are away, driving the actual release into
January. Even if that didn't happen, the possibility made December releases
stressful.
I don't think a strictly quarterly cadence is necessary; e.g. we could make
the Apr-Jul release cycle a couple weeks shorter to get that release out of
July and into June, if devs would find that better for their summer
schedules.
Note that my idea here isn't to come up with a set-in-stone schedule a year
in advance. It's more to come up with general idea that we can promise to
our community (e.g. 4 releases in a year) and a rough prediction when they
will be.
[1]
https://lists.jboss.org/archives/list/wildfly-dev@lists.jboss.org/thread/...
[2] Note that the WildFly 31 Beta date is significantly further ahead of
the Final date than the typical 2 weeks. This is so the Beta is done and
released before so many devs end work for the year.
Best regards,
Brian Stansberry
Project Lead, WildFly