wsry commented on code in PR #19653:
URL: https://github.com/apache/flink/pull/19653#discussion_r878647986
##########
flink-runtime/src/main/java/org/apache/flink/runtime/shuffle/ShuffleMaster.java:
##########
@@ -84,6 +86,33 @@ CompletableFuture<T> registerPartitionWithProducer(
PartitionDescriptor partitionDescriptor,
ProducerDescriptor producerDescriptor);
+ /**
+ * Returns all the shuffle descriptors for the partitions in the
intermediate data set with the
+ * given id.
+ *
+ * @param intermediateDataSetID The id of the intermediate data set.
+ * @return all the shuffle descriptors for the partitions in the
intermediate data set. Null if
+ * not exist.
+ */
+ default Collection<T> getClusterPartitionShuffleDescriptors(
Review Comment:
I mean ClusterPartitionManager and ResourceManagerPartitionTracker, sorry to
cause misunderstanding. These classes already have some ability to manage
cluster partitions, for example, if a task executor dead, it will remove the
lost cluster partitions. But it currently does not support the interface you
need which returns a list of ShuffleDescriptors. I think it's better to unify
the cluster partition management. BTW, I think it is better to also take care
of the case when a cluster partition gets lost in the remote shuffle cluster
like what we do currently for the task executor dead.
--
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]
