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

    https://github.com/apache/flink/pull/2386#discussion_r83207030
  
    --- Diff: 
flink-streaming-java/src/main/java/org/apache/flink/streaming/runtime/tasks/OperatorChain.java
 ---
    @@ -273,7 +276,7 @@ public void releaseOutputs() {
     
                // now create the operator and give it the output collector to 
write its output to
                OneInputStreamOperator<IN, OUT> chainedOperator = 
operatorConfig.getStreamOperator(userCodeClassloader);
    -           chainedOperator.setup(containingTask, operatorConfig, output);
    +           chainedOperator.setup(containingTask, operatorConfig, output, 
streamOutputs.size() == 0);
    --- End diff --
    
    The reason why we still differentiate between those is that we don't want 
to keep the latency statistics for all parallel source instances at the 
intermediate operators.
    So in the current implementation:
    - Intermediate operators maintain latency statistics for each logical 
source (so not much data)
    - Sinks maintain latency statistics for all parallel instances of each 
logical source (a bit more data)
    The sinks maintain the additional data to allow users debugging latency 
issues with individual parallel instances (for example when a machine running a 
source has issues).


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---

Reply via email to