gaoyunhaii commented on pull request #18428:
URL: https://github.com/apache/flink/pull/18428#issuecomment-1027668111


   Hi Fabian~ 
   
   > I am not really familiar with this part of the code base but so far I did 
not see any failing tests regarding the TableOperatorWrapperGenerator can you 
elaborate a bit more why the current solution may cause problems.
   
   Sorry I'm also not expert in table / sql , but from the current 
implementation of TableOperatorWrapperGenerator, it tries to merge multiple 
transformations into one `MultipleInputTransformation`, and the parallelism of 
this one is set with the maximum of all the merged transformations. Previously 
the transformation might returns env.defaultParallelism in this step, but now 
it might return -1, suppose we have multiple parallelisms [200, env.default, 
env.default] and env.default = 300, previously the result would be 300 while 
now the result becomes 200.
   
   > Afaict the motivation to update the commit summary is exactly for 
unaligned checkpointing. In case the checkpoint barrier overtakes the records 
we only emit an incomplete summary and later need to emit a new summary with 
the updated amount of seen records.
   > In general, I am also open to restricting it in the first version and 
figuring out the unaligned checkpoint use cases later. WDYT?
   
   I think with unaligned checkpointing we should still be able to keep one 
summary for each checkpoint: each operator could only emit the summary after 
checkpoint complete notified & received all the summary from each input 
subtask? If this is feasible we could force the current implementation of 
pre-commit topology to follow this pattern, but not provide the same promise 
(only one summary to the same checkpont) to the post-commit topology, then we 
would not have compatibility no matter which pattern we want in the future~


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to