flink-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Robert Metzger <rmetz...@apache.org>
Subject Re: Planning Flink release 0.7-incubating
Date Tue, 07 Oct 2014 11:53:36 GMT
As an update for everyone: My POJO feature is finished, including
documentation.
Aljoscha is currently adopting the Scala API to have support for (nested)
POJOs as well.
Once that is done, I'll merge everything and create a first candidate that
we can use for testing.

On Tue, Sep 30, 2014 at 10:49 PM, Fabian Hueske <fhueske@apache.org> wrote:

> I just checked the "Run Example" quickstart and it needs a bit of work.
>
> 2014-09-30 22:41 GMT+02:00 Robert Metzger <rmetzger@apache.org>:
>
> > I'm working hard on getting the POJOs ready.
> >
> > We also should do a pass over our documentation, the quickstarts and the
> > website to see if everything is in a good state (for example the
> > collection-based execution needs documentation as well). We should also
> > finally document the hadoop-input format wrappers (I think Timo is
> working
> > on a pull request for that).
> > This page mentions the LocalDistributedExecutor and contains some (most
> > likely outdated) scala code:
> >
> http://flink.incubator.apache.org/docs/0.7-incubating/local_execution.html
> >
> > We also need to deprecate the old record api (
> > https://issues.apache.org/jira/browse/FLINK-1106).
> >
> > On Tue, Sep 30, 2014 at 9:17 PM, Stephan Ewen <sewen@apache.org> wrote:
> >
> > > I think we are approaching ready state.
> > > Last issues are going in and we started working on dependencies and
> test
> > > platform diversity in order to make stabilizing phase.
> > >
> > > We should have an official feature freeze soon and fork the 0.7-release
> > > branch. I personally vote to include the POJO support (I think Robert
> is
> > > sorting that one out and is close to completion), and I want to add the
> > > collection-based execution (today or tomorrow). Till has the BLOB
> manager
> > > ready, which would be good to include (better support large libraries
> or
> > > fat jars).
> > >
> > > After that, I vote to freeze.
> > >
> > > On Tue, Sep 30, 2014 at 9:12 PM, Gyula Fora <gyfora@apache.org> wrote:
> > >
> > > > Hey,
> > > >
> > > > So what is the current decision regarding the time of the upcoming
> > > release?
> > > >
> > > > As for the streaming component, we included all the features we
> wanted,
> > > we
> > > > will start to test everything tomorrow, making sure that all works as
> > > > intended.
> > > > We are also almost finished with cleaning up the connector
> dependencies
> > > > that Stephan pointed out, should be finished by tomorrow.
> > > >
> > > > Gyula
> > > >
> > > > On 26 Sep 2014, at 10:49, Fabian Hueske <fhueske@apache.org> wrote:
> > > >
> > > > > +1 to manage this on JIRA (if possible)
> > > > >
> > > > > 2014-09-26 10:40 GMT+02:00 Aljoscha Krettek <aljoscha@apache.org>:
> > > > >
> > > > >> Can we not manage this stuff on Jira?
> > > > >>
> > > > >> On Fri, Sep 26, 2014 at 10:16 AM, Stephan Ewen <sewen@apache.org>
> > > > wrote:
> > > > >>> I personally like the idea of SOFT time-based feature freezes.
> > > > Otherwise,
> > > > >>> releases will get delayed again and again, because of features
> that
> > > we
> > > > >> try
> > > > >>> to squeeze in.
> > > > >>>
> > > > >>> Having reached the freeze point already, we could still include
> the
> > > > >>> features that are pending ready state in the next days
> (streaming,
> > > blob
> > > > >>> Manager, POJOs), but otherwise head for a release state.
> > > > >>>
> > > > >>> We had a mail listing the issues to include into 0.7, but
a wiki
> > page
> > > > >> would
> > > > >>> probably be better. In that sense, we could start collecting
the
> > > issues
> > > > >> for
> > > > >>> the next release from now on.
> > > > >>> Am 26.09.2014 09:17 schrieb "Daniel Warneke" <warneke@apache.org
> >:
> > > > >>>
> > > > >>>> Hi,
> > > > >>>>
> > > > >>>> I like Fabian's idea. Is there a wiki page (or something
> similar)
> > > > where
> > > > >> we
> > > > >>>> can collect the proposed JIRAs?
> > > > >>>>
> > > > >>>> Best regards,
> > > > >>>>
> > > > >>>>    Daniel
> > > > >>>>
> > > > >>>> Am 24.09.2014 23:03, schrieb Fabian Hueske:
> > > > >>>>
> > > > >>>>> I agree, a hard feature stop deadline might not be
the best
> > > practice.
> > > > >>>>>
> > > > >>>>> How about the following procedure:
> > > > >>>>> We decide two (or three) weeks before a targeted
release date
> > about
> > > > >> which
> > > > >>>>> JIRAs to include. JIRAs that are selected for a release
should
> be
> > > > >>>>> completed
> > > > >>>>> or really close to completion (via progress estimates
in JIRA).
> > > > >>>>> After we decided which JIRAs to include in a release,
we can
> use
> > > JIRA
> > > > >> to
> > > > >>>>> track the progress and dedicate another week exclusively
for
> > > testing
> > > > >> after
> > > > >>>>> the last feature was completed.
> > > > >>>>>
> > > > >>>>>
> > > > >>>>> 2014-09-24 19:10 GMT+02:00 Márton Balassi <
> > > balassi.marton@gmail.com
> > > > >:
> > > > >>>>>
> > > > >>>>> As for the streaming team we're also getting ready
for the
> > release,
> > > > >> but a
> > > > >>>>>> couple of days will be needed to finish the features
that we
> > would
> > > > >> like
> > > > >>>>>> to
> > > > >>>>>> include.
> > > > >>>>>>
> > > > >>>>>>    - A little work is still needed for reduce
operations and
> > > > >>>>>>    groups/connected streams (any comment on Gyula's
recent
> > e-mail
> > > is
> > > > >>>>>> really
> > > > >>>>>>    appreciated :) )
> > > > >>>>>>    - The examples are being updated to match
the standard,
> check
> > > out
> > > > >> the
> > > > >>>>>>    WordCount. (
> > > > >>>>>>
> > > > >>>>>> https://github.com/mbalassi/incubator-flink/blob/
> > > > >>>>>> streaming-new/flink-addons/flink-streaming/flink-
> > > > >>>>>> streaming-examples/src/main/java/org/apache/flink/
> > > > >>>>>> streaming/examples/wordcount/WordCount.java
> > > > >>>>>> )
> > > > >>>>>>    Hopefully it gives you some deja vu. :)
> > > > >>>>>>
> > > > >>>>>>
> > > > >>>>>> On Wed, Sep 24, 2014 at 6:53 PM, Ufuk Celebi
<uce@apache.org>
> > > > wrote:
> > > > >>>>>>
> > > > >>>>>> On 24 Sep 2014, at 18:37, Robert Metzger <rmetzger@apache.org
> >
> > > > >> wrote:
> > > > >>>>>>>
> > > > >>>>>>> Hey guys,
> > > > >>>>>>>>
> > > > >>>>>>>> exactly 3 weeks ago, we discussed to
do a feature freeze for
> > the
> > > > >>>>>>>> 0.7-incubating release today.
> > > > >>>>>>>>
> > > > >>>>>>>> From our initial feature list:
> > > > >>>>>>>> - *Flink Streaming* "Beta Preview". I
would suggest to ship
> > the
> > > > >>>>>>>>
> > > > >>>>>>> streaming,
> > > > >>>>>>>
> > > > >>>>>>>> but clearly mark it as a preview in the
documentation.
> > > > >>>>>>>> -* Java API Pojo improvements*: Code
generation, key
> selection
> > > > >> using a
> > > > >>>>>>>> string-expression:
> > > > https://issues.apache.org/jira/browse/FLINK-1032
> > > > >>>>>>>>  - *Reworked Scala API*. Bring the Scala
API in sync with
> the
> > > > >> latest
> > > > >>>>>>>> developments in the Java API:
> > > > >>>>>>>> https://issues.apache.org/jira/browse/FLINK-641
> > > > >>>>>>>>  -* Akka-based RPC service*:
> > > > >>>>>>>> https://issues.apache.org/jira/browse/FLINK-1019
> > > > >>>>>>>>  - *Kryo-based serialization*. This feature
has been
> requested
> > > by
> > > > >> many
> > > > >>>>>>>> users. Mostly because they wanted to
use Collections inside
> > > POJOs:
> > > > >>>>>>>> https://issues.apache.org/jira/browse/FLINK-610
> > > > >>>>>>>> - Rework JobManager internals to support
incremental program
> > > > >> rollout &
> > > > >>>>>>>> execution
> > > > >>>>>>>> - First parts of dynamic memory assignments
> > > > >>>>>>>>
> > > > >>>>>>>> The following features are in the master,
as of today:
> > > > >>>>>>>> - *Flink Streaming*
> > > > >>>>>>>> - *Reworked Scala API*
> > > > >>>>>>>> -* New Scheduler*
> > > > >>>>>>>>
> > > > >>>>>>>> We certainly need some days to test everything
until we can
> > > start
> > > > >> the
> > > > >>>>>>>>
> > > > >>>>>>> vote.
> > > > >>>>>>>
> > > > >>>>>>>> Based on our experience with the last
major release, I would
> > > > really
> > > > >>>>>>>>
> > > > >>>>>>> like
> > > > >>>>>>
> > > > >>>>>>> to
> > > > >>>>>>>
> > > > >>>>>>>> do the testing and bugfixing BEFORE the
first release
> > candidate.
> > > > For
> > > > >>>>>>>>
> > > > >>>>>>> the
> > > > >>>>>>
> > > > >>>>>>> 0.6-incubating release, we had 6  candidates)
> > > > >>>>>>>>
> > > > >>>>>>>> How do you guys feel about this? Should
we wait a few more
> > days
> > > > for
> > > > >> the
> > > > >>>>>>>> release so that a few more features make
it into the
> release?
> > > > >>>>>>>>
> > > > >>>>>>>> I'm undecided on this. On the one hand,
its really nice to
> > > release
> > > > >> on a
> > > > >>>>>>>> regular schedule, but it also eats up
some time and causes
> > > > overhead
> > > > >>>>>>>> (different branches etc.).
> > > > >>>>>>>> I would really like to have the Java
API Pojo improvements
> in
> > > the
> > > > >>>>>>>>
> > > > >>>>>>> release.
> > > > >>>>>>>
> > > > >>>>>>>> I think I can finish it until end of
this week.
> > > > >>>>>>>>
> > > > >>>>>>>> Opinions?
> > > > >>>>>>>>
> > > > >>>>>>> I agree that the finished features (especially
the Scala API)
> > are
> > > > >> nice
> > > > >>>>>>>
> > > > >>>>>> for
> > > > >>>>>>
> > > > >>>>>>> a new release, but still I would like to
wait a few more
> days.
> > > > >>>>>>>
> > > > >>>>>>> Some of the missing features are on the brink
of being
> finished
> > > > (e.g.
> > > > >>>>>>> the
> > > > >>>>>>> Pojo improvements). I wouldn't want to invest
a week in bug
> > > fixing
> > > > >> and
> > > > >>>>>>> doing the release vote, when the new features
are likely to
> be
> > > > >> finished
> > > > >>>>>>> just a few days afterwards. And the upcoming
features will
> > > > >> definitely be
> > > > >>>>>>> worth a release, so users can work with them.
;)
> > > > >>>>>>>
> > > > >>>>>>
> > > > >>>>
> > > > >>
> > > >
> > > >
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message