rkhachatryan commented on a change in pull request #13928:
URL: https://github.com/apache/flink/pull/13928#discussion_r517430882
##########
File path:
flink-runtime/src/main/java/org/apache/flink/runtime/io/network/partition/PipelinedApproximateSubpartition.java
##########
@@ -92,6 +94,11 @@ private void releaseView() {
}
}
+ @Override
+ public void finishReadRecoveredState(boolean
notifyAndBlockOnCompletion) throws IOException {
+ // The Approximate Local Recovery can not work with unaligned
checkpoint for now, so no need to recover channel state
Review comment:
Another option would be to pass `false` to the existing method from
`StreamTask`, as it's done for iterative jobs:
```
private void readRecoveredChannelState() throws IOException,
InterruptedException {
reader.readOutputData(getEnvironment().getAllWriters(),
!configuration.isGraphContainingLoops());
```
In the end, I think we'll need a check that we are either not using
approximate recovery or not rescaling with UC.
It would probably be more consistent and easier to implement with the 2nd
approach.
WDYT?
The check itself can be added to JM after we compute virtual mappings (cc:
@AHeise).
----------------------------------------------------------------
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]
