[jira] [Commented] (FLINK-7347) "removeAll" is extremely inefficient in MessageAcknowledgingSourceBase.notifyCheckpointComplete

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

    [ https://issues.apache.org/jira/browse/FLINK-7347?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16112347#comment-16112347 ] 

ASF GitHub Bot commented on FLINK-7347:
---------------------------------------

GitHub user ymost opened a pull request:

    https://github.com/apache/flink/pull/4468

    [FLINK-7347] [streaming] Keep ids for current checkpoint in a set instead of a list

    JIRA issue: FLINK-7347
    
    ## What is the purpose of the change
    
    This pull request changes the data structure used to store acknowledge-pending message ids from an `ArrayList` to a `HashSet`. This is done to eliminate an extremely inefficient call to the `removeAll` method in `MessageAcknowledgingSourceBase.notifyCheckpointComplete`.
    The implementation of `removeAll` is such that if the set is smaller than the collection to remove, then the set is iterated and every item is checked for containment in the collection. The `contains` action on an `ArrayList` is very inefficient, and it is performed for every item the set.
    In our pipeline we had about 10 million events processed, and the checkpoint was stuck on the `removeAll` call for hours.
    
    
    ## Brief change log
    
    Keep ids for current checkpoint in a set instead of a list
    
    
    ## Verifying this change
    
    This change is already covered by existing tests, such as `RMQSourceTest`.
    
    ## Does this pull request potentially affect one of the following parts:
    
      - Dependencies (does it add or upgrade a dependency): **no**
      - The public API, i.e., is any changed class annotated with `@Public(Evolving)`: **no**
      - The serializers: **no**
      - The runtime per-record code paths (performance sensitive): **yes** - performance should be improved
      - Anything that affects deployment or recovery: JobManager (and its components), Checkpointing, Yarn/Mesos, ZooKeeper: **yes** - this should resolve the problem where checkpoints would get stuck on the call to `removeAll`
    
    ## Documentation
    
      - Does this pull request introduce a new feature? **no**
      - If yes, how is the feature documented? **not applicable**
    


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

    $ git pull https://github.com/ymost/flink FLINK-7347-checkpoint-ids-set

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

    https://github.com/apache/flink/pull/4468.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 #4468
    
----

----


> "removeAll" is extremely inefficient in MessageAcknowledgingSourceBase.notifyCheckpointComplete
> -----------------------------------------------------------------------------------------------
>
>                 Key: FLINK-7347
>                 URL: https://issues.apache.org/jira/browse/FLINK-7347
>             Project: Flink
>          Issue Type: Improvement
>          Components: DataStream API
>    Affects Versions: 1.3.1
>            Reporter: Yonatan Most
>
> Observe this line in {{MessageAcknowledgingSourceBase.notifyCheckpointComplete}}:
> {code}
> idsProcessedButNotAcknowledged.removeAll(checkpoint.f1);
> {code}
> The implementation of {{removeAll}} is such that if the set is smaller than the collection to remove, then the set is iterated and every item is checked for containment in the collection. The type of {{checkpoint.f1}} here is {{ArrayList}}, so the {{contains}} action is very inefficient, and it is performed for every item in {{idsProcessedButNotAcknowledged}}.
> In our pipeline we had about 10 million events processed, and the checkpoint was stuck on the {{removeAll}} call for hours.
> A simple solution is to make {{idsForCurrentCheckpoint}} a {{HashSet}} instead of an {{ArrayList}}. The fact that it's a list is not really used anywhere.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)