[ 
https://issues.apache.org/jira/browse/BEAM-11952?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Yichi Zhang updated BEAM-11952:
-------------------------------
    Description: 
For example:

Using session(5), first merge window fn invocation of [1, 5) and [2, 6) will 
send back the merge result [1, 6): \{[1, 5), [2,6)}, this result will be 
processed by runner and the runner could have cleaned up [1, 5) and [2, 6).

Then the second invocation of [10, 15) and [11, 16) comes, instead of just 
sending back the result for this invocation [10, 16) : \{[10, 15), [11, 16)} 
the SDK harness will also send back previous result due to the merged widnows 
not being cleaned up on each invocation.

This will be a memory leak and will also cause runner to deal with invalid 
result from the merge window fn request.

  was:
For example:

Using session(5), first merge window fn invocation of [1, 5) and [2, 6) will 
send back the merge result [1, 6): \{[1, 5), [2,6)}, this result will be 
processed by runner and the runner could have cleaned up [1, 5) and [2, 6).

Then the second invocation of [10, 15) and [11, 16) comes, instead of just 
sending back the result for this invocation [10, 16) : \{[10, 15), [11, 16)} 
the SDK harness will also send back previous result due to the merged widnows 
not being cleaned up on each invocation.

This will be a memory leak and also will also cause runner to deal with invalid 
result from the merge window fn request.


> WindowMergingFnRunner unnecessarily keeps previous merged results
> -----------------------------------------------------------------
>
>                 Key: BEAM-11952
>                 URL: https://issues.apache.org/jira/browse/BEAM-11952
>             Project: Beam
>          Issue Type: Bug
>          Components: sdk-java-harness
>            Reporter: Yichi Zhang
>            Priority: P2
>          Time Spent: 50m
>  Remaining Estimate: 0h
>
> For example:
> Using session(5), first merge window fn invocation of [1, 5) and [2, 6) will 
> send back the merge result [1, 6): \{[1, 5), [2,6)}, this result will be 
> processed by runner and the runner could have cleaned up [1, 5) and [2, 6).
> Then the second invocation of [10, 15) and [11, 16) comes, instead of just 
> sending back the result for this invocation [10, 16) : \{[10, 15), [11, 16)} 
> the SDK harness will also send back previous result due to the merged widnows 
> not being cleaned up on each invocation.
> This will be a memory leak and will also cause runner to deal with invalid 
> result from the merge window fn request.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to