zhijiangW edited a comment on issue #11534: [FLINK-16537][network] Implement 
ResultPartition state recovery for unaligned checkpoint
URL: https://github.com/apache/flink/pull/11534#issuecomment-604887982
 
 
   Thanks for review @AHeise !
   
   > I'd like to have a test case where restoration and normal reading are 
mixed if that is possible.
   
   What do you mean by normal reading? Could you explain a bit then i can judge 
whether it is possible.
   
   > I'm also missing the glue code that uses the restore code in StreamTask.
   
   I also do not get the point here. The `StreamTask` only triggers the process 
of initializing output states besides previous operator states. And only the 
output states are loaded into the subpartition queue, it would be requested 
finally by downstream side via `PartitionRequest`. And I also added the unit 
tests to verify this process.
   
   I guess you might refer to the process of requesting this output restored 
data by downstream side. It should be compatible with existing 
`PartitionRequest`, no need changes on upstream side. But it might need some 
changes on downstream side to decide when to request this restored output, and 
it is actually related to the input side recovery, to avoid the upstream's 
output overtaking the downstream's input states. I would consider it in the 
input recovery PR.

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


With regards,
Apache Git Services

Reply via email to