I'm gonna jump in with my $0.02 and say that releasing from 'master' is
never a good idea.  It just so happens to work out for us now, but I'm
pretty sure c0sin is a masochist and we should be wary of anything he
suggests. (heart you c0s!)

I, for one, have lofty goals of puppet-4 enablement, but that's clearly a
feature that doesn't belong on the 1.2.x train.  Freezing 'master' for a
1.2.x point release just makes my feature branch more-and-more outdated
during the freeze, so let's not do that.  And let's not worry about which
jiras belong on which branch at potential release time -- it's too error
prone.  Let's instead come up with a release process that works for us.

Full disclosure in case you didn't recognize my email domain -- I work at
Canonical, and the steady Ubuntu release cadence (April / October) has
worked very well for us.  I'm not suggesting we align Bigtop with those
dates, but I *am* suggesting we come up with a regular release cycle that
works for us.

Here's what I have in mind:
- clone 'master' to 'stable' right now; this is the branch from which all
1.2.x releases will be born until 1.3.
- features that break backwards compat go on 'master' (when they're ready,
of course); bug fixes should make their way to 'stable' as appropriate.
- educate committers to be sure of the branch they're committing (i'm
guilty of this, as i do not pay much attention to the destination of things
i commit -- if the patch applies, it gets pushed to master!).  aside: half
of us use patches attached to JIRAa; the other (better) half use github
PRs.  let's pick one way and do that!
- define a release cadence:  quarterly, biannually, annually, whatever
makes sense.  personally, i like quarterly.  newbies to our project get a
stable release that's at-worst 3 months old.  this of course brings the
burden of a release manager; i'm happy to sign up as the 1.2.2 RM.
- integration test strategy.  i know juju can test stacks; i'm sure itest
or some k8s / container strategy can be used as well.  let's build on
Evans' matrix [1] to gate releases.  this needs to be cross-OS and
cross-$arch, as I believe this to be one of the major advantages of Bigtop
-- it just works no matter the underlying infra.

I know this stuff is like software engineering 101, but I think it's
important to keep our band-of-crazies on the same page.  If for no other
reason, let's just agree on something for my sake.  Thoughts?

[1]
https://ci.bigtop.apache.org/view/Provisioner/job/Bigtop-trunk-deployments/

Thanks!
-Kevin

On Tue, Jun 27, 2017 at 1:00 PM, Konstantin Boudnik <[email protected]> wrote:

> Releasing from master is ok but would actually create more work than
> it seems. Basically, we'd need to freeze the master through the whole
> release process, etc. With all 1.2.1 patches on master we can just
> merge them into 1.2 and have the sweet time releasing 1.2.1 without
> blocking anything.
>
> It doesn't mean we have to support 1.2.* unless we want to and have
> resources, but just from the pure technical standpoint it seems
> easier. But I am ok either way.
>
> Cos
>
> --
>   Take care,
> Konstantin (Cos) Boudnik
> 2CAC 8312 4870 D885 8616  6115 220F 6980 1F27 E622
>
> Disclaimer: Opinions expressed in this email are those of the author,
> and do not necessarily represent the views of any company the author
> might be affiliated with at the moment of writing.
>
>
> On Sun, Jun 25, 2017 at 10:54 AM, Evans Ye <[email protected]> wrote:
> > Hi all,
> >
> > Olaf asked on slack that how do we do 1.2.1 release when there's no patch
> > goes in 1.2-branch and most on the patches for 1.2.1 are on master.
> >
> > My thinking is directly release 1.2.1 version from master, with following
> > reasons:
> >
> > 1. We did not bump master to 1.3.0-SNAPSHOT
> > 2. With small number of components upgraded we can say it's a minor
> version
> > release for bigtop. We've discussed this to meet the goal of release
> > earlier, release often.
> > 3. it seems most of the 1.3 fixed issues[1] can also reside in 1.2.1
> except
> > Debian-9 support, which may or may not be excluded because it's quite
> > independent.
> > 4. We don't have too much resource to maintain two release lines. We
> shall
> > better keep it simple as long as it's still a reasonable way.
> >
> > We're close to the end of June so I'll start to prepare the release. For
> > those doc related JIRAs, I'll try to get them polished before the
> release.
> > Please help me to get the others fixed if you've some free cycles :)
> >
> > [1]. https://s.apache.org/byeu
> >
> >
> >
> > 2017-05-29 21:05 GMT+08:00 Arnaud Launay <[email protected]>:
> >
> >> Le Mon, May 29, 2017 at 10:24:34AM +0800, Evans Ye a écrit:
> >> > Surely we can add Solr 6.5.1(BIGTOP-2784) to 1.2.1 release. Do
> >> > you have time to get it to the finish line recently?
> >>
> >> Unfortunately, I don't have enough knowledge about it to really
> >> understand what's needed and what's not, the install.sh I have
> >> edited so far is mostly "err, this file doesn't exist anymore,
> >> let's comment the line to let the compilation continue".
> >>
> >> So I do have a compiling solr 6, but I'm fairly certain the
> >> generated packages are incomplete :/
> >>
> >>         Arnaud.
> >>
>

Reply via email to