[ 
https://issues.apache.org/jira/browse/FLINK-5048?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15662985#comment-15662985
 ] 

ASF GitHub Bot commented on FLINK-5048:
---------------------------------------

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

    https://github.com/apache/flink/pull/2789#discussion_r87738353
  
    --- Diff: 
flink-streaming-connectors/flink-connector-kafka-0.9/src/main/java/org/apache/flink/streaming/connectors/kafka/internal/Kafka09Fetcher.java
 ---
    @@ -143,133 +123,26 @@ public Kafka09Fetcher(
     
        @Override
        public void runFetchLoop() throws Exception {
    --- End diff --
    
    We will be throwing all exceptions, even if it's a 
`Handover.ClosedException`, correct?
    
    I wonder if it makes sense to suppress `Handover.ClosedException`s to not 
reach the main task thread, and only restore the interruption state that 
follows `cancel()`? So basically, we catch `InterruptedException` on the whole 
`runFetchLoop()` scope.
    
    This was what the exception passing behaviour was like before. Before, when 
`cancel()` was called on the fetcher, we won't be throwing any other 
exceptions, only restoring the interruption state to the main task thread.


> Kafka Consumer (0.9/0.10) threading model leads problematic cancellation 
> behavior
> ---------------------------------------------------------------------------------
>
>                 Key: FLINK-5048
>                 URL: https://issues.apache.org/jira/browse/FLINK-5048
>             Project: Flink
>          Issue Type: Bug
>          Components: Kafka Connector
>    Affects Versions: 1.1.3
>            Reporter: Stephan Ewen
>            Assignee: Stephan Ewen
>             Fix For: 1.2.0
>
>
> The {{FLinkKafkaConsumer}} (0.9 / 0.10) spawns a separate thread that 
> operates the KafkaConsumer. That thread is shielded from interrupts, because 
> the Kafka Consumer has not been handling thread interrupts well.
> Since that thread is also the thread that emits records, it may block in the 
> network stack (backpressure) or in chained operators. The later case leads to 
> situations where cancellations get very slow unless that thread would be 
> interrupted (which it cannot be).
> I propose to change the thread model as follows:
>   - A spawned consumer thread pull from the KafkaConsumer and pushes its 
> pulled batch of records into a blocking queue (size one)
>   - The main thread of the task will pull the record batches from the 
> blocking queue and emit the records.
> This allows actually for some additional I/O overlay while limiting the 
> additional memory consumption - only two batches are ever held, one being 
> fetched and one being emitted.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to