zookeeper-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Justin Ling Mao" <maoling199210...@sina.com>
Subject Re: Re: Re: Time to think about a 3.6.0 release?
Date Tue, 16 Jul 2019 10:06:10 GMT
















- 1. @Andor Thanks for reformatting my email. I must be careful next time.








- 2. I absolutely agree with what Olivelli said.      --> “I guess that if we start now
we can have a 3.6 in September”.       maybe not need so fast. But having a planning, especially
about what features will be included, what approximate time(A time range) will cut 3.6.0 is
a good idea  drawing from the previous lessons- 3. I have a question: "After 3.6.0 has landed,
the 3.5.* will not be applied any new feature and only fix the bugs and prominent improvements"
Is it right?



----- Original Message -----
From: Enrico Olivelli <eolivelli@gmail.com>
To: DevZooKeeper <dev@zookeeper.apache.org>, maoling199210191@sina.com
Subject: Re: Re: Time to think about a 3.6.0 release?
Date: 2019-07-15 21:55

Justin,
I think that current master has already plenty of new features and it is
worth to start thinking to a release.
The more with add code the more we add risk.
The point of this thread is more about 'stop adding new stuff, complete
ongoing work and start a rampdown phase', it is not "cut a release right
now"
As far as I know current master is very different from branch 3.5,
expecially on server side code, lots of feature came in and stalled on
master branch for months or even years,
so feedback from 3.5 is useful but not so blocker
We should make current master stable
IMHO the recipe for a great release is:
1) enough stuff committed to the release branch sothat it is worth to cut a
release
2) code in good shape: tests are passing, automatic checks are passing
(spotbugs, rat...)
3) licensing stuff is okay
4)  upgrade instructions and changelog about breaking changes/new
beheaviours are complete
5) CI is doing well
6) consensus of the community about the release
Hopefully now that we got out of the 3.5-BETA  problem and we are stable we
can think about a time based release schedule, if a feature can't be
delivered on a release it won't pass so much time for a new release, say
3-4 months
I don't know how many companies are using "current master" (or something
like that) in production, I feel that running 3.5 does not tell very much
about the stability of current 3.6
So my plan would be:
- merge pending pull requests that are ready
- stabilize the codebase  (no more BLOCKER issues for the release)
- start release process
I guess that if we start now we can have a 3.6 in September
Enrico
Il lun 15 lug 2019, 11:55 Justin Ling Mao <maoling199210191@sina.com> ha
scritto:
> - 1.Since the 3.5.5 has just released in May. we still need some time to
> collect the users' feedback.we cannot make sure the release time of 3.6.0?
> Giving the experience from the previous release history:)- 2.please Let me
> share some my thoughts, and the work in progress will be arriving into
> 3.6.0. Plz correct me if I got something wrong.
>  ------------------------------------------P0----------------------------------------------------------------
>    - Support the backend store engine:LMDB. this work needs a very detailed
> proposal which I will send to the community for being discussed fully.
>  - Add a complete backup mechanism for zookeeper internal(PR-917) which I
> will sharp it this week.     - A very powerful benchmark tool(PR-1011)
> which will be available within these two week.     - improve the
> performance of read/write to have the distinct advantages compared to etcd
> v3.4 which will be released soon.     - To strengthen the quota
> feature(PR-934,PR-936,PR-938) and implement the throughout quota.     - To
> strengthen the implements of TTL node(PR-1010)     - Add some new very
> useful CLIs: quorumInfo, watch .etc     - Observe and strengthen the new
> metric system continuously.
>  ------------------------------------------P1----------------------------------------------------------------
>    - strength the docs, especially about the c client, local session,
> security(TLS),ZAB protocol .etc     - introduce some chaos, fuzzy tests and
> tools to hit and check the zk.     - Clean up the all the checkstyle
> violations in the zookeeper-server module(ZOOKEEPER-3431)
>  -----------------------------------------
> P2------------------------------------------------------------------     -
> Debug mode feature. Look at an example of redis     - the tracing
> feature(PR-994). if having another time, integrating with opentracing
> sounds a very good idea.     - replace jute with thrift or PB may be put
> into the 4.0.0 when wanting to break the backward compatibility? And at the
> 4.0.0, implementing the restful api is also a  very good idea.
>
>
>
> ----- Original Message -----
> From: Fangmin Lv <lvfangmin@gmail.com>
> To: dev@zookeeper.apache.org
> Subject: Re: Time to think about a 3.6.0 release?
> Date: 2019-06-26 07:33
>
> It's great to have a 3.6.0 release, currently all the FB contributed
> features has been running inside FB for more than a month, so it
> should be stable enough for community to use.
> Also I agreed with Patrick's point to review all flags and consider to turn
> on by default.
> For the pending PRs, the following might be higher priority and would be
> nice to include in the 3.6.0 release:
> * ZOOKEEPER-3356: Implement advanced Netty flow control based on feedback
> from ZK to avoid OOM issue
> * ZOOKEEPER-3145: Avoid watch missing issue due to stale pzxid when
> replaying CloseSession txn with fuzzy snapshot
> * ZOOKEEPER-3240: Close socket on Learner shutdown to avoid dangling socket
> Thanks,
> Fangmin
> On Sat, Jun 15, 2019 at 9:21 AM Patrick Hunt <phunt@apache.org> wrote:
> > Good idea. Agree on including anything we've postponed to a new cycle -
> the
> > patch from mapr is an obvious one to consider.
> >
> > We should also look at things we've disabled by default and consider
> > whether we can turn them on by default. If not why not, and what can we
> do
> > to fix this in a subsequent release.
> >
> > Have we deprecated anything that we should now remove?
> >
> > Also a good time to review the state of Java versions and make changes
> wrt
> > supported versions and so forth.
> >
> > There was a proposal to remove contribs, or at least consider the ones
> that
> > are still valuable vs moving some out. We should do that as well.
> >
> > Regards,
> >
> > Patrick
> >
> > On Sat, Jun 15, 2019 at 9:02 AM Jordan Zimmerman <
> > jordan@jordanzimmerman.com>
> > wrote:
> >
> > > On Persistent/Recursive watches: I’m willing to rebase, etc if there’s
> > > confidence it will be merged.
> > >
> > > ====================
> > > Jordan Zimmerman
> > >
> > > > On Jun 15, 2019, at 10:59 AM, Andor Molnar
> <andor@cloudera.com.invalid
> > >
> > > wrote:
> > > >
> > > > Hi Enrico!
> > > >
> > > > Very good point, I entirely support the idea.
> > > >
> > > > Question to Friends@Facebook and Twitter contributors: how many
> > > outstanding
> > > > Jiras/PRs do you have which you would like to see in 3.6?
> > > >
> > > > I'd also like to highlight the long outstanding PR from Mapr:
> > > > https://github.com/apache/zookeeper/pull/730
> > > >
> > > > And some great new features which are still looking for to be merged:
> > > > - Persistent recursive watchers:
> > > > https://github.com/apache/zookeeper/pull/136
> > > > - Enforce client auth: https://github.com/apache/zookeeper/pull/118
> > > > - Slow operation log
> > > > - Jetty port unification
> > > >
> > > > Regards,
> > > > Andor
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >> On Sat, Jun 15, 2019 at 1:31 PM Enrico Olivelli <
> eolivelli@gmail.com>
> > > wrote:
> > > >>
> > > >> Hi Zookeepers !
> > > >> I checked on JIRA and it seems that master in good shape, no real
> > > blockers
> > > >> that mine the stability of the code.
> > > >>
> > > >> We have plenty of cool pull requests almost ready to be merged
> (mostly
> > > from
> > > >> Facebook friends and Twitter fork)
> > > >>
> > > >> Current master branch is full of great features in respect to 3.5.
> > > >>
> > > >> AFAIK There is no incompatibility with 3.5 so it is okay to stay
> with
> > > >> 3.6.0, although I think that there is so much stuff to legit a
> switch
> > to
> > > >> 4.0.0 (but we can reserve such bump for the time we will separate
> the
> > > java
> > > >> client and create a minimal compatibility breakage)
> > > >>
> > > >> Thoughts?
> > > >>
> > > >> Enrico
> > > >>
> > >
> >
>
Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message