[VOTE] Release Apache Flink 0.9.0 (release-0.9.0-rc3)

[mxm]

Dear community!

After careful testing and three release candidates, I take a chance and
hereby start a vote.

Please vote on releasing the following candidate as Apache Flink version
0.9.0.

-------------------------------------------------------------
The commit to be voted on:
e4643d248350c5c2727d93aafd227e544f7d5690

Branch:
release-0.9.0-rc3 (
https://git1-us-west.apache.org/repos/asf/flink/?p=flink.git;a=shortlog;h=refs/heads/release-0.9.0-rc3
)

The release artifacts to be voted on can be found at:
http://people.apache.org/~mxm/flink-0.9.0-rc3/

Release artifacts are signed with the key with fingerprint C2909CBF:
http://www.apache.org/dist/flink/KEYS

The staging repository for this release can be found at:
https://repository.apache.org/content/repositories/orgapacheflink-1041
-------------------------------------------------------------

Please vote on releasing this package as Apache Flink 0.9.0.

The vote is open for the next 72 hours and passes if a majority of at least
three +1 PMC votes are cast.

The vote ends on Saturday (June 20, 2015).

[ ] +1 Release this package as Apache Flink 0.9.0
[ ] -1 Do not release this package because ...


The following commits have been added since the last release candidate
(release-0.9.0-rc2):

0914a7d [FLINK-2210] Table API support for aggregation on columns with null
values
adc3e0e [FLINK-2203] handling null values for RowSerializer
fa02e5f [FLINK-2174] allow comments in 'slaves' file
44b969e [FLINK-2226][YARN] fail application on failed single-job cluster job
57810b5 [build] merge transitive notice files to shaded notices
57955d9 [FLINK-2120][runtime] rename AbstractJobVertex to JobVertex
6a9782a [legal] Updates LICENSE/NOTICE file of source and binary
distribution
d690633 [FLINK-2225] [scheduler] Excludes static code paths from
co-location constraint to avoid scheduling problems
0a8142a [hotfix] Removed execute() that followed print() in quickstart
wordcount jobs
2359b49 [docs] Update obsolate cluster execution guide
c2b1e12 [FLINK-2209] [docs] Document linking with jars not in the binary
dist
ccaa0b0 [FLINK-2221] [docs] Docs for not using local filesystem on the
cluster as state backup
d147c71 [hotfix] Some small fixups in README.md
a300eec [FLINK-2219] [webfrontend] Fix IllegalArgumentException and
decrease log level
c4f3f48 [FLINK-2224] Log error cause in JobStatusChange
8232809 [FLINK-2216] Exclude javadoc jar from examples
b342cf7 [readme] Synchronize tagline with intro, fix typos

[Márton Balassi]

+1

Verified checksums and signatures
Built from source
Run bundled batch examples on a local setup
Run streaming example with checkpointed operator state on cluster, killed
taskmanagers underneath

On Wed, Jun 17, 2015 at 2:35 AM, Maximilian Michels <[hidden email]> wrote:

> Dear community!
>
> After careful testing and three release candidates, I take a chance and
> hereby start a vote.
>
> Please vote on releasing the following candidate as Apache Flink version
> 0.9.0.
>
> -------------------------------------------------------------
> The commit to be voted on:
> e4643d248350c5c2727d93aafd227e544f7d5690
>
> Branch:
> release-0.9.0-rc3 (
>
> https://git1-us-west.apache.org/repos/asf/flink/?p=flink.git;a=shortlog;h=refs/heads/release-0.9.0-rc3
> )
>
> The release artifacts to be voted on can be found at:
> http://people.apache.org/~mxm/flink-0.9.0-rc3/
>
> Release artifacts are signed with the key with fingerprint C2909CBF:
> http://www.apache.org/dist/flink/KEYS
>
> The staging repository for this release can be found at:
> https://repository.apache.org/content/repositories/orgapacheflink-1041
> -------------------------------------------------------------
>
> Please vote on releasing this package as Apache Flink 0.9.0.
>
> The vote is open for the next 72 hours and passes if a majority of at least
> three +1 PMC votes are cast.
>
> The vote ends on Saturday (June 20, 2015).
>
> [ ] +1 Release this package as Apache Flink 0.9.0
> [ ] -1 Do not release this package because ...
>
>
> The following commits have been added since the last release candidate
> (release-0.9.0-rc2):
>
> 0914a7d [FLINK-2210] Table API support for aggregation on columns with null
> values
> adc3e0e [FLINK-2203] handling null values for RowSerializer
> fa02e5f [FLINK-2174] allow comments in 'slaves' file
> 44b969e [FLINK-2226][YARN] fail application on failed single-job cluster
> job
> 57810b5 [build] merge transitive notice files to shaded notices
> 57955d9 [FLINK-2120][runtime] rename AbstractJobVertex to JobVertex
> 6a9782a [legal] Updates LICENSE/NOTICE file of source and binary
> distribution
> d690633 [FLINK-2225] [scheduler] Excludes static code paths from
> co-location constraint to avoid scheduling problems
> 0a8142a [hotfix] Removed execute() that followed print() in quickstart
> wordcount jobs
> 2359b49 [docs] Update obsolate cluster execution guide
> c2b1e12 [FLINK-2209] [docs] Document linking with jars not in the binary
> dist
> ccaa0b0 [FLINK-2221] [docs] Docs for not using local filesystem on the
> cluster as state backup
> d147c71 [hotfix] Some small fixups in README.md
> a300eec [FLINK-2219] [webfrontend] Fix IllegalArgumentException and
> decrease log level
> c4f3f48 [FLINK-2224] Log error cause in JobStatusChange
> 8232809 [FLINK-2216] Exclude javadoc jar from examples
> b342cf7 [readme] Synchronize tagline with intro, fix typos
>

[Aljoscha Krettek-2]

-1

There is a bug in the newly introduced Null-Value support in RowSerializer:
The serializer was changed to write booleans that signify if a field is
null. For comparison this still uses the TupleComparatorBase (via
CaseClassComparator) which is not aware of these changes.

The reason why no Unit-Test found this problem is that it only occurs if
very long keys are used that exceed the normalised-key length. Only then do
we actually have to compare the binary data.

I see three options:
 - Revert the relevant Table API changes
 - Create a new RowComparator that does not derive from CaseClassComparator
but basically copies almost all the code
 - Add support for null-values in Tuples and Case classes as well, thereby
bringing all composite types in sync regarding null-values.


On Wed, 17 Jun 2015 at 16:30 Márton Balassi <[hidden email]>
wrote:

> +1
>
> Verified checksums and signatures
> Built from source
> Run bundled batch examples on a local setup
> Run streaming example with checkpointed operator state on cluster, killed
> taskmanagers underneath
>
> On Wed, Jun 17, 2015 at 2:35 AM, Maximilian Michels <[hidden email]>
> wrote:
>
> > Dear community!
> >
> > After careful testing and three release candidates, I take a chance and
> > hereby start a vote.
> >
> > Please vote on releasing the following candidate as Apache Flink version
> > 0.9.0.
> >
> > -------------------------------------------------------------
> > The commit to be voted on:
> > e4643d248350c5c2727d93aafd227e544f7d5690
> >
> > Branch:
> > release-0.9.0-rc3 (
> >
> >
> https://git1-us-west.apache.org/repos/asf/flink/?p=flink.git;a=shortlog;h=refs/heads/release-0.9.0-rc3
> > )
> >
> > The release artifacts to be voted on can be found at:
> > http://people.apache.org/~mxm/flink-0.9.0-rc3/
> >
> > Release artifacts are signed with the key with fingerprint C2909CBF:
> > http://www.apache.org/dist/flink/KEYS
> >
> > The staging repository for this release can be found at:
> > https://repository.apache.org/content/repositories/orgapacheflink-1041
> > -------------------------------------------------------------
> >
> > Please vote on releasing this package as Apache Flink 0.9.0.
> >
> > The vote is open for the next 72 hours and passes if a majority of at
> least
> > three +1 PMC votes are cast.
> >
> > The vote ends on Saturday (June 20, 2015).
> >
> > [ ] +1 Release this package as Apache Flink 0.9.0
> > [ ] -1 Do not release this package because ...
> >
> >
> > The following commits have been added since the last release candidate
> > (release-0.9.0-rc2):
> >
> > 0914a7d [FLINK-2210] Table API support for aggregation on columns with
> null
> > values
> > adc3e0e [FLINK-2203] handling null values for RowSerializer
> > fa02e5f [FLINK-2174] allow comments in 'slaves' file
> > 44b969e [FLINK-2226][YARN] fail application on failed single-job cluster
> > job
> > 57810b5 [build] merge transitive notice files to shaded notices
> > 57955d9 [FLINK-2120][runtime] rename AbstractJobVertex to JobVertex
> > 6a9782a [legal] Updates LICENSE/NOTICE file of source and binary
> > distribution
> > d690633 [FLINK-2225] [scheduler] Excludes static code paths from
> > co-location constraint to avoid scheduling problems
> > 0a8142a [hotfix] Removed execute() that followed print() in quickstart
> > wordcount jobs
> > 2359b49 [docs] Update obsolate cluster execution guide
> > c2b1e12 [FLINK-2209] [docs] Document linking with jars not in the binary
> > dist
> > ccaa0b0 [FLINK-2221] [docs] Docs for not using local filesystem on the
> > cluster as state backup
> > d147c71 [hotfix] Some small fixups in README.md
> > a300eec [FLINK-2219] [webfrontend] Fix IllegalArgumentException and
> > decrease log level
> > c4f3f48 [FLINK-2224] Log error cause in JobStatusChange
> > 8232809 [FLINK-2216] Exclude javadoc jar from examples
> > b342cf7 [readme] Synchronize tagline with intro, fix typos
> >
>

[Ufuk Celebi-2]

On 17 Jun 2015, at 18:05, Aljoscha Krettek <[hidden email]> wrote:

> -1
>
> There is a bug in the newly introduced Null-Value support in RowSerializer:
> The serializer was changed to write booleans that signify if a field is
> null. For comparison this still uses the TupleComparatorBase (via
> CaseClassComparator) which is not aware of these changes.
>
> The reason why no Unit-Test found this problem is that it only occurs if
> very long keys are used that exceed the normalised-key length. Only then do
> we actually have to compare the binary data.
>
> I see three options:
> - Revert the relevant Table API changes
> - Create a new RowComparator that does not derive from CaseClassComparator
> but basically copies almost all the code
> - Add support for null-values in Tuples and Case classes as well, thereby
> bringing all composite types in sync regarding null-values.

I vote vor option 1 for now.

[mxm]

I also vote for reverting the Table API changes.

On Wed, Jun 17, 2015 at 6:16 PM, Ufuk Celebi <[hidden email]> wrote:

>
> On 17 Jun 2015, at 18:05, Aljoscha Krettek <[hidden email]> wrote:
>
> > -1
> >
> > There is a bug in the newly introduced Null-Value support in
> RowSerializer:
> > The serializer was changed to write booleans that signify if a field is
> > null. For comparison this still uses the TupleComparatorBase (via
> > CaseClassComparator) which is not aware of these changes.
> >
> > The reason why no Unit-Test found this problem is that it only occurs if
> > very long keys are used that exceed the normalised-key length. Only then
> do
> > we actually have to compare the binary data.
> >
> > I see three options:
> > - Revert the relevant Table API changes
> > - Create a new RowComparator that does not derive from
> CaseClassComparator
> > but basically copies almost all the code
> > - Add support for null-values in Tuples and Case classes as well, thereby
> > bringing all composite types in sync regarding null-values.
>
> I vote vor option 1 for now.
>

[Aljoscha Krettek-2]

+1 I also think it's the cleanest solution for now. The table API still
works, just without support for null values.

On Thu, 18 Jun 2015 at 10:08 Maximilian Michels <[hidden email]> wrote:

> I also vote for reverting the Table API changes.
>
> On Wed, Jun 17, 2015 at 6:16 PM, Ufuk Celebi <[hidden email]> wrote:
>
> >
> > On 17 Jun 2015, at 18:05, Aljoscha Krettek <[hidden email]> wrote:
> >
> > > -1
> > >
> > > There is a bug in the newly introduced Null-Value support in
> > RowSerializer:
> > > The serializer was changed to write booleans that signify if a field is
> > > null. For comparison this still uses the TupleComparatorBase (via
> > > CaseClassComparator) which is not aware of these changes.
> > >
> > > The reason why no Unit-Test found this problem is that it only occurs
> if
> > > very long keys are used that exceed the normalised-key length. Only
> then
> > do
> > > we actually have to compare the binary data.
> > >
> > > I see three options:
> > > - Revert the relevant Table API changes
> > > - Create a new RowComparator that does not derive from
> > CaseClassComparator
> > > but basically copies almost all the code
> > > - Add support for null-values in Tuples and Case classes as well,
> thereby
> > > bringing all composite types in sync regarding null-values.
> >
> > I vote vor option 1 for now.
> >
>

[Till Rohrmann]

+1 for reverting.

On Thu, Jun 18, 2015 at 10:11 AM Aljoscha Krettek <[hidden email]>
wrote:

> +1 I also think it's the cleanest solution for now. The table API still
> works, just without support for null values.
>
> On Thu, 18 Jun 2015 at 10:08 Maximilian Michels <[hidden email]> wrote:
>
> > I also vote for reverting the Table API changes.
> >
> > On Wed, Jun 17, 2015 at 6:16 PM, Ufuk Celebi <[hidden email]> wrote:
> >
> > >
> > > On 17 Jun 2015, at 18:05, Aljoscha Krettek <[hidden email]>
> wrote:
> > >
> > > > -1
> > > >
> > > > There is a bug in the newly introduced Null-Value support in
> > > RowSerializer:
> > > > The serializer was changed to write booleans that signify if a field
> is
> > > > null. For comparison this still uses the TupleComparatorBase (via
> > > > CaseClassComparator) which is not aware of these changes.
> > > >
> > > > The reason why no Unit-Test found this problem is that it only occurs
> > if
> > > > very long keys are used that exceed the normalised-key length. Only
> > then
> > > do
> > > > we actually have to compare the binary data.
> > > >
> > > > I see three options:
> > > > - Revert the relevant Table API changes
> > > > - Create a new RowComparator that does not derive from
> > > CaseClassComparator
> > > > but basically copies almost all the code
> > > > - Add support for null-values in Tuples and Case classes as well,
> > thereby
> > > > bringing all composite types in sync regarding null-values.
> > >
> > > I vote vor option 1 for now.
> > >
> >
>