I'm on the side of freezing/branching earlier so we can really start the QA
process, but I do understand the concerns.

As Kurt alluded to previously, given our current release velocity, 4.1/5.0
will likely be some time away after 4.0. If we manage to release two high
quality stable major versions back to back in a span of say 12 months, that
would actually be pretty awesome. The upgrade cycle will be a minor
complaint for just those major versions as the community settles on a
better cadence as we learn and go through it.

Both Kurt and Jeff have advocated for key features that should be part of
the next major update which seems to be a major part of the desire to push
back against an early feature freeze. Interestingly most of these
contribute to the theme of the release (stability) even though they are
large changes. Particularly:

https://issues.apache.org/jira/browse/CASSANDRA-9754  - "Birch" (changes
file format)
https://issues.apache.org/jira/browse/CASSANDRA-10540 - RangeAwareCompaction
https://issues.apache.org/jira/browse/CASSANDRA-13426 - Idemponent schema
changes

We haven't seen any actual binding -1s yet on June 1, despite obvious
concerns and plenty of +1s

Having said that, the above issues are the only ones people have identified
as:

   - the issues is a desired feature,
   - the issue has clear progress on the tickets,
   - the issues fits the theme of the release, and
   - there is some concern about the issue making the June 1 deadline.

I would invite those working on / reviewing these tickets to comment (Michael
Kjellman, Marcus Eriksson, Aleksey Yeschenko) specifically about inclusion
into 4.0 and June 1.

If we want to delay the feature freeze for these, either a June 1 freeze
with a carve-out/exception for those three (they can get committed after
June 1 to 4.0) or a moderate push back of the freeze date (e.g. July 1) may
be an appropriate compromise.

The carve-out/exception however is messy and opens a can of worms on the
proposed testing process for a 4.0 branch, but it is an option.

I know this list doesn't include changes like transient replicas and
strongly consistent schema changes (previously mentioned), but the state of
the tickets is still in an architectural discussion, so I don't think its
worth making them blockers. Pluggable storage was also raised as something
worth including for 4.0, if someone working on those (Dikang, Blake?) had
an opinion on it regarding 4.0, impact on stability and a June 1 deadline?

Ben


On Wed, Apr 11, 2018 at 11:15 AM Blake Eggleston <beggles...@apple.com>
wrote:

> I agree that not releasing semi-regularly is not good for the project. I
> think our habit of releasing half working software is much worse though.
> Our testing/stability story is not iron clad. I really think the bar for
> releasing 4.0 should be that the people in this thread are running the code
> in production, recommending their customers run it in production, or
> offering and supporting it as part of their cloud service.
>
> In that context, the argument for waiting for some features is less about
> trying to do all the things and more about making 4.0 something worth the
> time and expense of validating for production.
>
> ´╗┐On 4/11/18, 1:06 AM, "Sylvain Lebresne" <lebre...@gmail.com> wrote:
>
>     On Wed, Apr 11, 2018 at 12:35 AM Jeff Jirsa <jji...@gmail.com> wrote:
>
>     > Seriously, what's the rush to branch? Do we all love merging so much
> we
>     > want to do a few more times just for the sake of merging? If nothing
>     > diverges, there's nothing gained from the branch, and if it did
> diverge, we
>     > add work for no real gain.
>     >
>
>     Again, to me, the "rush" is that 1) there is tons of changes sitting in
>     trunk
>     that some user (_not all_, granted)[1], especially new ones, would
> likely
>     benefits, and sooner for those is better than later, 2) we want to
> favor
>     release stability and we *know* from years of experience (and frankly,
>     common
>     sense) that the bigger the release is, the harder it is to test
> it/ensuring
>     overall stability[2] and 3) not having major releases for years[3] is
>     impacting at least the perceived dynamism/liveness of the project to
>     external
>     actors (prospective new user come in mind here, but not only) and
> that's
>     simply bad for the project.
>
>     And having listed arguments for a soon freeze/not accumulating much
> more
>     before release, I'd like to reverse the question to you: what are the
> big
>     downsides of not doing that? Are we really that hung up on our own
>     developers
>     comfort that the annoyance of a bit more merging trumps the arguments
> above?
>
>     Anyway, the reasons above make me thing that it's better _for the
> project_
>     to freeze 4.0 soon, which doesn't exclude a "short" cycle for the
> following
>     major (where my definition of short here is something like 6-8
> months), and
>     I'm happy to decide to make 4.0 a non-mandatory upgrade to whatever
>     comes next so that folks that prefer upgrading rarely can simply skip
> it and
>     go to the next one. Likely nobody will die if we wait more though, and
> it's
>     clear it will make a few people here more happy if we do, but I
> believe the
>     project as a whole will be a bit worst off, that's all.
>
>     --
>     Sylvain
>
>
>     [1]: I'll note that I don't deny upgrading is huge deal for some
> users, but
>     let's not skew arguments too much based on any one user interest. For
> many
>     users, upgrading even every year to get improvements is still
> considered as
>     a
>     good deal, and that's not counting new users for which it's super
>     frustrating
>     to miss out on improvements because we release major only every 2+
> years.
>     [2]: I'll be clear: I will simply not buy anyone argument that "we'll
> do
>     so much better testing this time" on face value. Not anymore. If you
> want to
>     use that argument to sell having bigger releases, then prove it first.
> Let's
>     do reasonably sized 4.0 and 4.1/5.0 and prove that our
> testing/stability
>     story
>     is iron clad now, and then for 4.2/6.0 I'll be willing to agree that
> making
>     bigger release may not impact stability too much.
>     [3]: Conservative estimate, if we do care about stable releases as we
> all
>     seem
>     to, even if we were to freeze June 1, we will almost surely not release
>     before
>     October/November, which will be ~1.3 year since the last major release
>     (again,
>     that's the conservative estimate). If we push a few months to get some
> big
>     complex feature in, not only this push the freeze of those few months,
> but
>     will also require more testing, so we're looking at 2+ years, with a
>     possibly
>     large '+'.
>
>
>
>
>     >
>     > Beyond that, I still don't like June 1. Validating releases is hard.
> It
>     > sounds easy to drop a 4.1 and ask people to validate again, but it's
> a hell
>     > of a lot harder than it sounds. I'm not saying I'm a hard -1, but I
> really
>     > think it's too soon. 50'ish days is too short to draw a line in the
> sand,
>     > especially as people balance work obligations with Cassandra feature
>     > development.
>     >
>     >
>     >
>     >
>     > On Tue, Apr 10, 2018 at 3:18 PM, Nate McCall <zznat...@gmail.com>
> wrote:
>     >
>     > > A lot of good points and everyone's input is really appreciated.
>     > >
>     > > So it sounds like we are building consensus towards June 1 for 4.0
>     > > branch point/feature freeze and the goal is stability. (No one has
>     > > come with a hard NO anyway).
>     > >
>     > > I want to reiterate Sylvain's point that we can do whatever we
> want in
>     > > terms of dropping a new feature 4.1/5.0 (or whatev.) whenever we
> want.
>     > >
>     > > In thinking about this, what is stopping us from branching 4.0 a
> lot
>     > > sooner? Like now-ish? This will let folks start hacking on trunk
> with
>     > > new stuff, and things we've gotten close on can still go in 4.0
>     > > (Virtual tables). I guess I'm asking here if we want to
> disambiguate
>     > > "feature freeze" from "branch point?" I feel like this makes sense.
>     > >
>     > >
> ---------------------------------------------------------------------
>     > > To unsubscribe, e-mail: dev-unsubscr...@cassandra.apache.org
>     > > For additional commands, e-mail: dev-h...@cassandra.apache.org
>     > >
>     > >
>     >
>
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscr...@cassandra.apache.org
> For additional commands, e-mail: dev-h...@cassandra.apache.org
>
> --
Ben Bromhead
CTO | Instaclustr <https://www.instaclustr.com/>
+1 650 284 9692
Reliability at Scale
Cassandra, Spark, Elasticsearch on AWS, Azure, GCP and Softlayer

Reply via email to