flink-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Vasiliki Kalavri <vasilikikala...@gmail.com>
Subject Re: Planning the 0.9 Release
Date Fri, 05 Jun 2015 14:13:23 GMT
If you want my personal opinion, I'd say yes.

I am currently using Gelly for all my projects. For one of them, we have
been running experiments over the last 4 months and we'll be deploying it
in production very soon :)

Gelly did not change any internals or runtime features; it simply builds on
the DataSet API and the delta iterations and these are stable features.

The motivation for deprecating Spargel and creating the migration guide was
that we don't want to confuse people by having two Graph APIs. If you feel
uncomfortable with a deprecated API and one in beta, then go ahead and
change that. It's no problem for me.

Cheers,
-V.

On 5 June 2015 at 15:51, Stephan Ewen <sewen@apache.org> wrote:

> Fair enough about including the issues into 0.9.1
>
> Concerning Gelly, would you recommend people to use that in production
> today? If not, it would be nice to have some non-deprecated code where we
> are confident about that.
>
> On Fri, Jun 5, 2015 at 2:08 PM, Vasiliki Kalavri <
> vasilikikalavri@gmail.com>
> wrote:
>
> > Hi,
> >
> > let me clarify:
> >
> > FLINK-1252 is missing a test for PageRank (which might not even be
> needed,
> > since the implementation is basically identical to the existing Spargel
> > one) and a test for MusicProfiles, which is basically using
> LabelProgation
> > (and we have a separate test for this).
> >
> > FLINK-1943 is about implementing a compiler and translation test for the
> > recently added Gather-Sum-Apply iteration.
> >
> > IMO, the second would be nice to have, but not a blocker.
> > I could work on it after my paper deadline, in a week. But since I see
> > you're eager to have the release today, we could include this is the
> first
> > bugfix of 0.9.
> >
> > -Vasia.
> >
> >
> > On 5 June 2015 at 13:59, Andra Lungu <lungu.andra@gmail.com> wrote:
> >
> > > The Pregel-like, vertex-cenric part of Gelly is as stable as it will
> ever
> > > be.
> > > I vote for deprecating Spargel in this release, but keep in mind that
> > this
> > > is just an opinion :)
> > >
> > > On Fri, Jun 5, 2015 at 1:50 PM, Stephan Ewen <sewen@apache.org> wrote:
> > >
> > > > Okay, I was not aware it is only two missing tests. That is not that
> > big
> > > a
> > > > deal.
> > > >
> > > > I am not very attached to the Spargel Stuff, I just want to make sure
> > we
> > > do
> > > > not deprecate something that works well for something that is still
> > work
> > > in
> > > > progress.
> > > >
> > > >
> > > >
> > > >
> > > > On Fri, Jun 5, 2015 at 1:46 PM, Andra Lungu <lungu.andra@gmail.com>
> > > wrote:
> > > >
> > > > > Hi Stephan,
> > > > >
> > > > > I don't know if I have a saying in this, but I will give it a go
:)
> > > > >
> > > > > The two unsolved issues don't affect the functionality at all.
> > > > > Gelly can, at the moment, support anything Spargel could. There is
> a
> > > > guide
> > > > > in the documentation explaining how to migrate Spargel code to
> > Gelly. I
> > > > > don't see why Spargel should not be deprecated yet. Just because
of
> > > those
> > > > > missing tests?
> > > > >
> > > > > On Fri, Jun 5, 2015 at 1:40 PM, Stephan Ewen <sewen@apache.org>
> > wrote:
> > > > >
> > > > > > Thanks Vasia!
> > > > > >
> > > > > > We should clearly label Gelly as Work in Progress and at Beta
> > status,
> > > > > then
> > > > > > it should be okay. This is very fair, it is the first version,
> > people
> > > > > > understand that.
> > > > > >
> > > > > > In that sense, let us not call Spargel deprecated in favor of
> Gelly
> > > > yet,
> > > > > > but do that for the next release when Gelly is ready.
> > > > > >
> > > > > > Stephan
> > > > > >
> > > > > >
> > > > > > On Fri, Jun 5, 2015 at 12:10 PM, Vasiliki Kalavri <
> > > > > > vasilikikalavri@gmail.com
> > > > > > > wrote:
> > > > > >
> > > > > > > Hi all,
> > > > > > >
> > > > > > > regarding the 2 gelly issues, I'm sorry but I haven't had
time
> to
> > > > work
> > > > > on
> > > > > > > these.
> > > > > > > And most certainly I won't be able to work on these today
:S
> > > > > > >
> > > > > > > In any case, I wouldn't consider them blocker issues, so
if you
> > > > agree,
> > > > > > > please go ahead with the release candidate.
> > > > > > >
> > > > > > > -Vasia.
> > > > > > >
> > > > > > > On 5 June 2015 at 11:46, Robert Metzger <rmetzger@apache.org>
> > > wrote:
> > > > > > >
> > > > > > > > I'll address the remaining documentation issues today.
> > > > > > > >
> > > > > > > > What about
> > > > > > > >
> > > > > > > >    -
> > > > > > > >
> > > > > > > >    Sync Streaming Java/Scala API
> > > > > > > >    - Consolidate names across batch/streaming (discussion)
> > > > > > > >    - Merge static code analysis
> > > > > > > >
> > > > > > > > and the gelly TODOs
> > > > > > > >
> > > > > > > >    -
> > > > > > > >
> > > > > > > >    FLINK-1522 Add tests for the library methods and
examples
> > > > > > > >    - FLINK-1943 Add Gelly-GSA compiler and translation
tests
> > > > > > > >
> > > > > > > >
> > > > > > > > They seem both unresolved.
> > > > > > > >
> > > > > > > > Other than that it seems we are good to go.
> > > > > > > > Maybe we can manage to get the first release candidate
out
> > today?
> > > > > > > >
> > > > > > > > I'm uncertain whether we should fork off the "release-0.9"
> > branch
> > > > as
> > > > > > part
> > > > > > > > of the RC creation or whether we should wait a bit
with that.
> > > > > > > >
> > > > > > > > Forking it of would allow us to merge some of the
pull
> requests
> > > > > (storm
> > > > > > > > compat), but we would need to apply many patches to
two
> > branches.
> > > > > > > >
> > > > > > > >
> > > > > > > > On Fri, Jun 5, 2015 at 10:56 AM, Stephan Ewen <
> > sewen@apache.org>
> > > > > > wrote:
> > > > > > > >
> > > > > > > > > Yes, Till, Max, Ufuk, and me have reached consensus
that we
> > > will
> > > > > > > postpone
> > > > > > > > > the batch scheduling to 0.10. It is a crucial
feature, but
> we
> > > > would
> > > > > > not
> > > > > > > > > like to have it in a highly pre-mature version
in there.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > What came up a lot is that there are a good set
of
> additions
> > > that
> > > > > are
> > > > > > > > close
> > > > > > > > > to completion. We will probably bring up the
suggestion of
> a
> > > 0.10
> > > > > > > release
> > > > > > > > > very soon after the 0.9 release.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > On Fri, Jun 5, 2015 at 10:50 AM, Ufuk Celebi
<
> uce@apache.org
> > >
> > > > > wrote:
> > > > > > > > >
> > > > > > > > > > Ping.
> > > > > > > > > >
> > > > > > > > > > On 04 Jun 2015, at 14:11, Ufuk Celebi <uce@apache.org>
> > > wrote:
> > > > > > > > > >
> > > > > > > > > > > Critical issues:
> > > > > > > > > > >
> > > > > > > > > > > - Skipped buffer (
> > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
> http://apache-flink-mailing-list-archive.1008284.n3.nabble.com/Buffer-re-ordering-problem-td6009.html
> > > > > > > > > )
> > > > > > > > > > (I'm on it)
> > > > > > > > > > > - Execution graph deadlock (FLINK-2133)
> > > > > > > > > >
> > > > > > > > > > I'm investigating the skipped buffer. I
vote to not block
> > the
> > > > > > release
> > > > > > > > on
> > > > > > > > > > this though if I can't reproduce it in the
next days.
> > > > > > > > > >
> > > > > > > > > > What about FLINK-2133?
> > > > > > > > > >
> > > > > > > > > > > Big open issue:
> > > > > > > > > > > - There has been consensus to merge
the static code
> > > analysis
> > > > PR
> > > > > > for
> > > > > > > > > this
> > > > > > > > > > release, which needs some further testing
(I'm on it)
> > > > > > > > > > > - Batch scheduling. I think Till and
Stephan are
> working
> > on
> > > > > this.
> > > > > > > Can
> > > > > > > > > > you guys give an estimate whether we will
be able to have
> > it
> > > in
> > > > > the
> > > > > > > > > release?
> > > > > > > > > >
> > > > > > > > > > @Till, Stephan: Will batch scheduling make
it into the
> > > release?
> > > > > :)
> > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
>

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