flink-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Robert Metzger <rmetz...@apache.org>
Subject [CANCELLED] [VOTE] Release Apache Flink 1.3.1
Date Mon, 19 Jun 2017 12:56:58 GMT
Thanks everybody for the testing of the RC1!

I'm hereby cancelling the first RC due to a critical bug in the serializer
stack, reported by Till.

I'll create the next RC once the fix is in.

On Mon, Jun 19, 2017 at 2:20 PM, Tzu-Li (Gordon) Tai <tzulitai@apache.org>
wrote:

> I looked at the changes in the PRs [1, 2].
> I agree that it would be better to include the fixes in 1.3.1.
>
> If we agree to cancel RC1 for RC2 (with shorter voting period), I’ll start
> massaging the PRs to be merged for 1.3.1 (currently the changes in the PRs
> are targeted for 1.3.2).
> They should be ready before tomorrow, so that we can start RC2 soon.
>
>
> On 19 June 2017 at 8:06:11 PM, Robert Metzger (rmetzger@apache.org) wrote:
>
> If we really have to introduce a special path in the serializer config for
> 1.3.1 to 1.3.2, then I would indeed suggest to cancel the RC1.
>
> If only this one commit is different between RC1 and RC2, we can do a
> reduced voting period for RC2.
>
> On Mon, Jun 19, 2017 at 2:02 PM, Till Rohrmann <trohrmann@apache.org>
> wrote:
>
> > I think this actually not true, since in 1.3.0 the field `enumConstants`
> in
> > `ScalaEnumSerializerConfigSnapshot` was always serialized as `null`.
> Thus,
> > due to this we wouldn't have to bump the format if we included FLINK-6948
> > in 1.3.1. If this weren't the case, we would have to bump it also for
> 1.3.1
> > (excluding FLINK-6948).
> >
> > Cheers,
> > Till
> >
> > On Mon, Jun 19, 2017 at 1:39 PM, Tzu-Li (Gordon) Tai <
> tzulitai@apache.org>
> > wrote:
> >
> > > No, I meant that if we include FLINK-6921 / FLINK-6948 in 1.3.1, we
> also
> > > still need to bump the version due to changes in FLINK-6948.
> > > So, that the version needs to be bumped would not be a reason to block
> > > 1.3.1 on it, because we have to do it either way.
> > >
> > > On 19 June 2017 at 7:33:25 PM, Till Rohrmann (trohrmann@apache.org)
> > wrote:
> > >
> > > Do you mean that we have to bump the version also without including
> > > FLINK-6921 and FLINK-6948? Wouldn't that be a release blocker then?
> > >
> > > I think that we actually introduced FLINK-6921 with [1]. Thus, the
> > > `ArrayIndexOutOfBoundsException` is specific to this release.
> However, I
> > > agree that the serializer was broken before as well, however, in a
> > > different way.
> > >
> > > [1]
> > > https://github.com/apache/flink/commit/5281dd6598f17c8dfe0c7b091c90c8
> > > 721d305375
> > >
> > > Cheers,
> > > Till
> > >
> > > On Mon, Jun 19, 2017 at 1:22 PM, Tzu-Li (Gordon) Tai <
> > tzulitai@apache.org>
> > > wrote:
> > >
> > > > The ScalaEnumSerializerConfigSnapshot would need a version bump
> > > > regardless of whether or not the fixes are included in 1.3.1.
> > > > In other words, we still need to bump the version if we include it
> for
> > > > 1.3.1.
> > > >
> > > > I’m not against including FLINK-6921 and FLINK-6948 in for 1.3.1, but
> > > then
> > > > as usual the argument would be that the problem had always been there
> > and
> > > > is not specific to this release.
> > > > I’m personally usually favorable of delaying the release a bit more
> to
> > > get
> > > > fixes for issues we know of in.
> > > >
> > > > I’ll look at the PRs for FLINK-6921 and FLINK-6948 now, and merge
> them
> > > > soon. We could probably have a RC2 with a shorter vote duration?
> > > >
> > > > Best,
> > > > Gordon
> > > >
> > > > On 19 June 2017 at 7:10:11 PM, Till Rohrmann (trohrmann@apache.org)
> > > wrote:
> > > >
> > > > I think the EnumValueSerializer [1, 2] is broken in the current RC.
> > This
> > > > basically means that Flink programs won’t properly notice that state
> > > > migration is required and or fail with obscure exceptions at
> migration
> > > > check or runtime.
> > > >
> > > > This definitely will be enough reason for another bug fix release if
> we
> > > > don’t want to include fixes in 1.3.1. If we include the fixes in
> 1.3.2,
> > > > then this will require a version bump for the
> > > > ScalaEnumSerializerConfigSnapshot because we have to change the
> > format.
> > > > This also entails code for backwards compatibility.
> > > >
> > > > [1] https://issues.apache.org/jira/browse/FLINK-6921
> > > > [2] https://issues.apache.org/jira/browse/FLINK-6948
> > > >
> > > > Cheers,
> > > > Till
> > > > ​
> > > >
> > > > On Mon, Jun 19, 2017 at 10:34 AM, Dawid Wysakowicz <
> > > > wysakowicz.dawid@gmail.com> wrote:
> > > >
> > > > > +1
> > > > >
> > > > > - built from source (2.10, 2.11)
> > > > > - checked aggregate function with AggregateFunction return type
> > > different
> > > > > than stream type
> > > > >
> > > > > Z pozdrowieniami! / Cheers!
> > > > >
> > > > > Dawid Wysakowicz
> > > > >
> > > > > *Data/Software Engineer*
> > > > >
> > > > > Skype: dawid_wys | Twitter: @OneMoreCoder
> > > > >
> > > > > <http://getindata.com/>
> > > > >
> > > > > 2017-06-19 7:15 GMT+02:00 Tzu-Li (Gordon) Tai <tzulitai@apache.org
> >:
> > > > >
> > > > > > +1
> > > > > >
> > > > > > Tested the following blockers of 1.3.1:
> > > > > >
> > > > > > Serializers & checkpointing
> > > > > > - Verified Scala jobs using Scala types as state (Scala
> > collections,
> > > > case
> > > > > > classes, Either, Try, etc.) can restore from savepoints taken
> with
> > > > Flink
> > > > > > 1.2.1 & 1.3.1. Tested with Scala 2.10 & 2.11.
> > > > > > - Tested restore of POJO types as state, behavior & error
> messages
> > > for
> > > > > > changed POJO types consistent across different state backends
> > > > > > - Tested stream join with checkpointing enabled
> > > > > > - Sharing static state descriptor (w/ stateful KryoSerializer)
> > across
> > > > > > tasks did not reveal any issues
> > > > > >
> > > > > > Elasticsearch connector
> > > > > > - ES 5 connector artifacts exists in staging repo
> > > > > > - Tested cluster execution with ES sink (2.3.5, 2.4.1, 5.1.2),
no
> > > > > > dependency conflicts, successful
> > > > > >
> > > > > > Flink CEP
> > > > > > - Out-of-order matched events is now resolved
> > > > > >
> > > > > > - Ran local build + test on MacOS (-Dscala-2.10, -Dscala-2.11),
> > > > > successful
> > > > > > - LICENSES untouched since 1.3.0
> > > > > > - No new dependencies
> > > > > >
> > > > > > Best,
> > > > > > Gordon
> > > > > >
> > > > > > On 14 June 2017 at 10:14:39 PM, Robert Metzger (
> > rmetzger@apache.org)
> > > > > > wrote:
> > > > > >
> > > > > > Dear Flink community,
> > > > > >
> > > > > > Please vote on releasing the following candidate as Apache Flink
> > > > version
> > > > > > 1.3.1.
> > > > > >
> > > > > > The commit to be voted on:
> > > > > > http://git-wip-us.apache.org/repos/asf/flink/commit/7cfe62b9
> > > > > >
> > > > > > Branch:
> > > > > > release-1.3.1-rc1
> > > > > >
> > > > > > The release artifacts to be voted on can be found at:
> > > > > > *http://people.apache.org/~rmetzger/flink-1.3.1-rc1/
> > > > > > <http://people.apache.org/~rmetzger/flink-1.3.1-rc1/>*
> > > > > >
> > > > > > The release artifacts are signed with the key with fingerprint
> > > > D9839159:
> > > > > > http://www.apache.org/dist/flink/KEYS
> > > > > >
> > > > > > The staging repository for this release can be found at:
> > > > > > https://repository.apache.org/content/repositories/
> > > orgapacheflink-1124
> > > > > >
> > > > > >
> > > > > > -------------------------------------------------------------
> > > > > >
> > > > > >
> > > > > > The vote ends on Monday (5pm CEST), June 19rd, 2016.
> > > > > >
> > > > > > [ ] +1 Release this package as Apache Flink 1.3.1
> > > > > > [ ] -1 Do not release this package, because ...
> > > > > >
> > > > >
> > > >
> > >
> >
>

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