[
https://issues.apache.org/jira/browse/FLINK-6337?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15986690#comment-15986690
]
ASF GitHub Bot commented on FLINK-6337:
---------------------------------------
Github user zhijiangW commented on a diff in the pull request:
https://github.com/apache/flink/pull/3785#discussion_r113706078
--- Diff:
flink-tests/src/test/java/org/apache/flink/test/misc/SuccessAfterNetworkBuffersFailureITCase.java
---
@@ -68,7 +68,6 @@ public void testSuccessfulProgramAfterFailure() {
try {
runKMeans(cluster.getLeaderRPCPort());
- fail("This program execution should have
failed.");
--- End diff --
This case will fail if not remove it.
I guess the buffer amount is not enough for this case before, because some
are shared by buffer provider in handler. After remove the buffer provider from
handler, the buffer is enough now. So the `runKmeans` can execute correctly,
not enter catch part or fail process.
I am not familiar with this case and not sure whether my understanding is
right.
> Remove the buffer provider from PartitionRequestServerHandler
> -------------------------------------------------------------
>
> Key: FLINK-6337
> URL: https://issues.apache.org/jira/browse/FLINK-6337
> Project: Flink
> Issue Type: Improvement
> Components: Network
> Reporter: zhijiang
> Assignee: zhijiang
> Priority: Minor
>
> Currently, {{PartitionRequestServerHandler}} will create a
> {{LocalBufferPool}} when the channel is registered. The {{LocalBufferPool}}
> is only used to get segment size for creating read view in
> {{SpillableSubpartition}}, and the buffers in the pool will not be used all
> the time, so it will waste the buffer resource of global pool.
> We would like to remove the {{LocalBufferPool}} from the
> {{PartitionRequestServerHandler}}, and the {{LocalBufferPool}} in
> {{ResultPartition}} can also provide the segment size for creating sub
> partition view.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
