wsry commented on a change in pull request #18173:
URL: https://github.com/apache/flink/pull/18173#discussion_r776549954
##########
File path:
flink-runtime/src/main/java/org/apache/flink/runtime/io/network/buffer/NetworkBufferPool.java
##########
@@ -158,7 +159,8 @@ public MemorySegment requestMemorySegment() {
public List<MemorySegment> requestMemorySegmentsBlocking(int
numberOfSegmentsToRequest)
throws IOException {
- return internalRequestMemorySegments(numberOfSegmentsToRequest);
+ return internalRequestMemorySegments(
+ numberOfSegmentsToRequest,
this::internalRecycleMemorySegments);
Review comment:
> Is it because all of the buffers in the LocalBufferPool are floating
buffers? And that's the basis of the implementation of the whole
exclusive/floating buffer concept?
I guess you are right, if we use only one LocalBufferPool, the buffers will
be shared by all remote channels, but the credit based flow control needs
exclusive buffers (if we set the exclusive buffer to 0 and makes all buffers
floating, some network micro benchmarks incurs at most about 20% regression).
Another possible way is to create a local buffer pool per remote channel?
In a word, we can rename the method name, add some comments or try to unify
the code to always use local buffer pool (need run micro benchmark to ensure
that there is no regression). All of these are acceptable to me. WDYT, any
suggestion?
--
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: issues-unsubscr...@flink.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
