flink-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Maximilian Michels <...@apache.org>
Subject Re: [DISCUSS] Releasing Flink 1.1.4
Date Wed, 26 Oct 2016 09:21:06 GMT
+1 for a 1.1.4 release

We could backport putting user jars into the system class loader for
per-job Yarn clusters: https://github.com/apache/flink/pull/2692
Arguably, this is somewhat a new feature but it gets rid of duplicate
class loading issues users experienced in practice.

We already have the following commits on the release-1.1 branch:

05a5f46 [FLINK-4862] fix Timer register in ContinuousEventTimeTrigger
5731672 [FLINK-4581] [table] Fix Table API throwing "No suitable driver
found for jdbc:calcite"
9c87f92 [FLINK-4586] [core] Broken AverageAccumulator
210230c [FLINK-4829] snapshot accumulators on a best-effort basis
c1d6b24 [FLINK-4829] protect user accumulators against concurrent updates
fe464b4 [FLINK-4709] [core] Fix resource leak in InputStreamFSInputWrapper
9f72698 [FLINK-4108] [scala] Respect ResultTypeQueryable for InputFormats.
9591d50 [FLINK-4506] [DataSet] Fix documentation of CsvOutputFormat about
incorrect default of allowNullValues
c9433bf [FLINK-3706] Fix YARN test instability
2203f74 [FLINK-4778] [docs] Fix WordCount parameters in CLI examples.

-Max


On Wed, Oct 26, 2016 at 7:05 AM, Jean-Baptiste Onofré <jb@nanthrax.net>
wrote:
> +1
>
> Looking forward this release !
>
> Regards
> JB
>
> ⁣
>
> On Oct 25, 2016, 14:43, at 14:43, Robert Metzger <rmetzger@apache.org>
wrote:
>>+1 for a bugfix release soon.
>>
>>On Tue, Oct 25, 2016 at 10:53 AM, Stephan Ewen <sewen@apache.org>
>>wrote:
>>
>>> Thanks fort starting this Ufuk.
>>>
>>> I would like to add the following issues to 1.1.4:
>>>
>>> Build errors due to Storm dependencies *(fix pending)*
>>>     - [FLINK-4298] [storm compatibility] Add proper repository for
>>Closure
>>> dependencies.
>>>
>>> Stability on S3 considering eventual consistency *(fix pending)*
>>>     - [FLINK-4218] [checkpoints] Do not fail checkpoints when state
>>size
>>> cannot be determined
>>>
>>> Avoiding Zombie TaskManagers *(still needs to be done)*
>>>     - [FLINK-3347] [akka] TaskManager (or its ActorSystem) need to
>>restart
>>> in case they notice quarantine
>>>
>>> Adding a limit to the amount of data spilled during checkpoint
>>alignments
>>> *(fix
>>> is work in progress)*
>>>     - [FLINK-4904] [checkpoints] Add a limit for how much data may be
>>> spilled in checkpoint alignments
>>>
>>>
>>> I can push the first two fixes to the 1.1.4 branch in a bit, the
>>fourth one
>>> later today.
>>> The third one (akka) is still pending.
>>>
>>> Best,
>>> Stephan
>>>
>>>
>>>
>>> On Mon, Oct 24, 2016 at 3:32 PM, Ufuk Celebi <uce@apache.org> wrote:
>>>
>>> > Hey all,
>>> >
>>> > I would like to start the discussion for kicking off the next bug
>>fix
>>> > release, Flink 1.1.4. What do you think about aiming for a RC by
>>end
>>> > of this week?
>>> >
>>> > Users reported some instabilities/inconveniences that would be good
>>to
>>> fix.
>>> >
>>> > Personally, I would like to backport the following fixes:
>>> >
>>> > (1) https://issues.apache.org/jira/browse/FLINK-4619: Answer client
>>if
>>> > savepoint restore fails (Already merged for master, needs minimal
>>> > adjustment for 1.1)
>>> > (2) https://issues.apache.org/jira/browse/FLINK-4715: Safety net
>>for
>>> > stuck task cancellation (Already reviewed for master, waiting for
>>> > tests to finish of backport)
>>> > (3) https://issues.apache.org/jira/browse/FLINK-4510: Always create
>>> > CheckpointCoordinator (Already merged for master, needs minimal
>>> > adjustments for 1.1)
>>> >
>>> > Furthermore, I would like to address the following:
>>> >
>>> > (4) https://issues.apache.org/jira/browse/FLINK-4445: Add option to
>>> > ignore unmatched state when restoring from savepoint
>>> > (5) https://issues.apache.org/jira/browse/FLINK-4894: Don't block
>>on
>>> > buffer request after broadcast event
>>> >
>>> > Strictly speaking, the (4) is not a bug fix. But given that it
>>would
>>> > only add an optional flag to savepoint restoring and should have
>>been
>>> > addressed for 1.1.0 already, I would like to get it in.
>>> >
>>>

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