[ 
https://issues.apache.org/jira/browse/SPARK-44808?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17754280#comment-17754280
 ] 

Wei Liu commented on SPARK-44808:
---------------------------------

This seems to be against the design principle of spark connect – We pass 
everything to the server side. Client only keep ids to get status. That's the 
reason why the foreachBatch and Listener are ran on the server side. This can 
be closed

> refreshListener() API on StreamingQueryManager for spark connect
> ----------------------------------------------------------------
>
>                 Key: SPARK-44808
>                 URL: https://issues.apache.org/jira/browse/SPARK-44808
>             Project: Spark
>          Issue Type: New Feature
>          Components: Connect, Structured Streaming
>    Affects Versions: 4.0.0
>            Reporter: Wei Liu
>            Priority: Major
>
> I’m thinking of an improvement for connect python listener and foreachBatch. 
> Currently if you define a variable outside of the function, you can’t 
> actually see it on client side if it’s touched in the function because the 
> operation is on server side, e.g. 
>  
>  
> {code:java}
> x = 0
> class MyListener(StreamingQueryListener):
>     def onQueryStarted(e):
>         x = 100
>         self.y = 200
> spark.streams.addListener(MyListener())
> q = spark.readStream.....start()
>  
> # x is still 0, self.y is not defined
> {code}
>  
>  
> But for the self.y case, there could be an improvement. 
>  
> The server side is capable of pickle serialize the whole listener instance 
> again, and send back to the client.
>  
> So we could define a new interface on the streaming query manager, maybe 
> called refreshListener(listener), e.g. use it as 
> `spark.streams.refreshListener()`
>  
>  
> {code:java}
> def refreshListener(listener: StreamingQueryListener) -> 
> StreamingQueryListener
>  # send the listener id with this refresh request, server receives this 
> request and serializes the listener again, and send back to client, so the 
> returned new listener contains the updated value self.y
>  
> {code}
>  
> For `foreachBatch`, we could wrap the function to a new class
>  



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org

Reply via email to