[
https://issues.apache.org/jira/browse/FLINK-7416?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16217997#comment-16217997
]
ASF GitHub Bot commented on FLINK-7416:
---------------------------------------
Github user zhijiangW commented on a diff in the pull request:
https://github.com/apache/flink/pull/4533#discussion_r146741125
--- Diff:
flink-runtime/src/main/java/org/apache/flink/runtime/io/network/netty/CreditBasedClientHandler.java
---
@@ -322,22 +322,25 @@ private void
writeAndFlushNextMessageIfPossible(Channel channel) {
while (true) {
RemoteInputChannel inputChannel =
inputChannelsWithCredit.poll();
- // The input channel may be null because of the write
callbacks that are executed
- // after each write, and it is also no need to notify
credit for released channel.
- if (inputChannel == null || inputChannel.isReleased()) {
+ // The input channel may be null because of the write
callbacks
+ // that are executed after each write.
+ if (inputChannel == null) {
return;
}
- AddCredit msg = new AddCredit(
- inputChannel.getPartitionId(),
- inputChannel.getAndResetCredit(),
- inputChannel.getInputChannelId());
+ //It is no need to notify credit for the released
channel.
+ if (!inputChannel.isReleased()) {
+ AddCredit msg = new AddCredit(
+ inputChannel.getPartitionId(),
+ inputChannel.getAndResetCredit(),
+ inputChannel.getInputChannelId());
- // Write and flush and wait until this is done before
- // trying to continue with the next input channel.
- channel.writeAndFlush(msg).addListener(writeListener);
+ // Write and flush and wait until this is done
before
+ // trying to continue with the next input
channel.
+
channel.writeAndFlush(msg).addListener(writeListener);
- return;
+ return;
--- End diff --
Yes, the current mechanism is the same with the upstream writing
`BufferResponse`.
We loop the `inputChannelsWithCredit` to find the first un-released
`InputChannel`, then call `writeAndFlush` to break this loop. The remaining
`InputChannel` s in the queue will be continued after `writeAndFlush` is done
to callback `writeAndFlushNextMessageIfPossible`.
I think it is mainly because we have to verify the channel is still
writable before loop the next `InputChannel`. So we have to wait the current
`writeAndFlush` done then check the channel writable to continue the next
`InputChannel`.
The current comment on this code
`channel.writeAndFlush(msg).addListener(writeListener)` would be helpful for
this explanation.
> Implement Netty receiver outgoing pipeline for credit-based
> -----------------------------------------------------------
>
> Key: FLINK-7416
> URL: https://issues.apache.org/jira/browse/FLINK-7416
> Project: Flink
> Issue Type: Sub-task
> Components: Network
> Reporter: zhijiang
> Assignee: zhijiang
> Fix For: 1.4.0
>
>
> This is a part of work for credit-based network flow control.
> The related works are :
> * We define a new message called {{AddCredit}} to notify the incremental
> credit during data shuffle.
> * Whenever an {{InputChannel}}’s unannounced credit goes up from zero, the
> channel is enqueued in the pipeline.
> * Whenever the channel becomes writable, it takes the next {{InputChannel}}
> and sends its unannounced credit. The credit is reset to zero after each sent.
> * That way, messages are sent as often as the network has capacity and
> contain as much credit as available for the channel at that point in time.
> Otherwise, it would only add latency to the announcements and not increase
> throughput.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
