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.
[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