[ 
https://issues.apache.org/jira/browse/FLINK-8516?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16347001#comment-16347001
 ] 

ASF GitHub Bot commented on FLINK-8516:
---------------------------------------

Github user tzulitai commented on a diff in the pull request:

    https://github.com/apache/flink/pull/5393#discussion_r165083343
  
    --- Diff: 
flink-connectors/flink-connector-kinesis/src/main/java/org/apache/flink/streaming/connectors/kinesis/internals/KinesisDataFetcher.java
 ---
    @@ -584,17 +594,34 @@ private static ShardMetricsReporter 
registerShardMetrics(MetricGroup metricGroup
        //  Miscellaneous utility functions
        // 
------------------------------------------------------------------------
     
    +   /**
    +    * Function to map a Kinesis shard to a Flink subtask index.
    +    */
    +   public interface ShardToSubtaskIndexFn {
    +           /**
    +            * Function to map a Kinesis shard to a Flink subtask index.
    +            *
    +            * @param shard the shard to determine
    +            * @param totalNumberOfSubtasks total number of subtasks
    +            * @return index or hash code
    +            */
    +           // TODO: extra parameter can be eliminated by creating hash 
function after runtime context is present
    +           int getSubTaskIndex(StreamShardHandle shard, int 
totalNumberOfSubtasks);
    --- End diff --
    
    I actually prefer passing the `totalNumberOfSubtasks` value independently, 
instead of passing in `runtimeContext`. IMO, it provides more context on the 
nature of the assignment.
    
    Moreover, IMO, having a factory as the API is much more complicating for 
the user.


> FlinkKinesisConsumer does not balance shards over subtasks
> ----------------------------------------------------------
>
>                 Key: FLINK-8516
>                 URL: https://issues.apache.org/jira/browse/FLINK-8516
>             Project: Flink
>          Issue Type: Bug
>          Components: Kinesis Connector
>    Affects Versions: 1.4.0, 1.3.2, 1.5.0
>            Reporter: Thomas Weise
>            Assignee: Thomas Weise
>            Priority: Major
>
> The hash code of the shard is used to distribute discovered shards over 
> subtasks round robin. This works as long as shard identifiers are sequential. 
> After shards are rebalanced in Kinesis, that may no longer be the case and 
> the distribution become skewed.



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

Reply via email to