jiangxin369 commented on code in PR #248:
URL: https://github.com/apache/flink-ml/pull/248#discussion_r1298092153
##########
flink-ml-iteration/flink-ml-iteration-common/src/main/java/org/apache/flink/iteration/operator/HeadOperator.java:
##########
@@ -422,16 +430,44 @@ public void close() throws Exception {
}
}
- private void registerFeedbackConsumer(Executor mailboxExecutor) {
+ private void registerFeedbackConsumer(Executor mailboxExecutor)
+ throws MemoryAllocationException {
+ StreamTask<?, ?> task = getContainingTask();
+
int indexOfThisSubtask = getRuntimeContext().getIndexOfThisSubtask();
int attemptNum = getRuntimeContext().getAttemptNumber();
FeedbackKey<StreamRecord<IterationRecord<?>>> feedbackKey =
OperatorUtils.createFeedbackKey(iterationId, feedbackIndex);
SubtaskFeedbackKey<StreamRecord<IterationRecord<?>>> key =
feedbackKey.withSubTaskIndex(indexOfThisSubtask, attemptNum);
- FeedbackChannelBroker broker = FeedbackChannelBroker.get();
- FeedbackChannel<StreamRecord<IterationRecord<?>>> channel =
broker.getChannel(key);
- OperatorUtils.registerFeedbackConsumer(channel, this, mailboxExecutor);
+ SpillableFeedbackChannelBroker broker =
SpillableFeedbackChannelBroker.get();
+ this.feedbackChannel = broker.getChannel(key);
Review Comment:
I agree with your 2nd point and I will pass the initialization logic as a
`Consumer`.
But the HeadOperator and TailOperator still need to coordinate on accessing
the shared channel. Because only the operator that called
`declareManagedMemoryUseCaseAtOperatorScope` can initialize the Channel, while
the other cannot since it has no managed memory to allocate.
--
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]
