[jira] [Commented] (FLINK-5479) Per-partition watermarks in FlinkKafkaConsumer should consider idle partitions

["ASF GitHub Bot (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-5479?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16385863#comment-16385863 ] 

ASF GitHub Bot commented on FLINK-5479:
---------------------------------------

GitHub user tzulitai opened a pull request:

    https://github.com/apache/flink/pull/5634

    [FLINK-5479] [kafka] Idleness detection for periodic per-partition watermarks in FlinkKafkaConsumer

    ## What is the purpose of the change
    
    This commit adds the capability to detect idle partitions in the `FlinkKafkaConsumer` when using periodic per-partition watermark generation. Users set the partition idle timeout using `setPartitionIdleTimeout(long)`. The value of the timeout determines how long a an idle partition may block watermark advancement downstream.
    
    
    ## Brief change log
    
    - Adds a `setPartitionIdleTimeout(long)` configuration method
    - Modifies `KafkaTopicPartitionStateWithPeriodicWatermarks` to keep track of necessary information to determine partition idleness.
    - Adds idleness detection logic to `AsbtractFetcher.PeriodicWatermarkEmitter`.
    
    ## Does this pull request potentially affect one of the following parts:
    
      - Dependencies (does it add or upgrade a dependency): (yes / **no**)
      - The public API, i.e., is any changed class annotated with `@Public(Evolving)`: (yes / **no**)
      - The serializers: (yes / **no** / don't know)
      - The runtime per-record code paths (performance sensitive): (yes / **no** / don't know)
      - Anything that affects deployment or recovery: JobManager (and its components), Checkpointing, Yarn/Mesos, ZooKeeper: (yes / **no** / don't know)
      - The S3 file system connector: (yes / **no** / don't know)
    
    ## Documentation
    
      - Does this pull request introduce a new feature? (**yes** / no)
      - If yes, how is the feature documented? (not applicable / docs / **JavaDocs** / not documented)


You can merge this pull request into a Git repository by running:

    $ git pull https://github.com/tzulitai/flink FLINK-5479

Alternatively you can review and apply these changes as the patch at:

    https://github.com/apache/flink/pull/5634.patch

To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:

    This closes #5634
    
----
commit d7b95ca1c8bb85f77dd6b83becf8fc1d5cccb810
Author: Tzu-Li (Gordon) Tai <tz...@...>
Date:   2018-03-05T09:35:02Z

    [FLINK-5479] [kafka] Idleness detection for periodic per-partition watermarks
    
    This commit adds the capability to detect idle partitions in the
    FlinkKafkaConsumer when using periodic per-partition watermark
    generation. Users set the partition idle timeout using
    `setPartitionIdleTimeout(long)`. The value of the timeout determines how
    long a an idle partition may block watermark advancement downstream.

----


> Per-partition watermarks in FlinkKafkaConsumer should consider idle partitions
> ------------------------------------------------------------------------------
>
>                 Key: FLINK-5479
>                 URL: https://issues.apache.org/jira/browse/FLINK-5479
>             Project: Flink
>          Issue Type: Improvement
>          Components: Kafka Connector
>            Reporter: Tzu-Li (Gordon) Tai
>            Priority: Blocker
>             Fix For: 1.5.0
>
>
> Reported in ML: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/Kafka-topic-partition-skewness-causes-watermark-not-being-emitted-td11008.html
> Similar to what's happening to idle sources blocking watermark progression in downstream operators (see FLINK-5017), the per-partition watermark mechanism in {{FlinkKafkaConsumer}} is also being blocked of progressing watermarks when a partition is idle. The watermark of idle partitions is always {{Long.MIN_VALUE}}, therefore the overall min watermark across all partitions of a consumer subtask will never proceed.
> It's normally not a common case to have Kafka partitions not producing any data, but it'll probably be good to handle this as well. I think we should have a localized solution similar to FLINK-5017 for the per-partition watermarks in {{AbstractFetcher}}.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)