[
https://issues.apache.org/jira/browse/FLINK-7416?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16268586#comment-16268586
]
ASF GitHub Bot commented on FLINK-7416:
---------------------------------------
Github user NicoK commented on a diff in the pull request:
https://github.com/apache/flink/pull/4533#discussion_r153446896
--- Diff:
flink-runtime/src/main/java/org/apache/flink/runtime/io/network/netty/CreditBasedClientHandler.java
---
@@ -102,13 +102,21 @@ void cancelRequestFor(InputChannelID inputChannelId) {
}
}
+ /**
+ * The credit is announced based on sender's backlog from buffer
response, so the channel is
+ * already active then. If the channel is closed because of error, we
can skip this announcement.
+ *
+ * @param inputChannel The input channel with unannounced credits.
+ */
void notifyCreditAvailable(final RemoteInputChannel inputChannel) {
- ctx.executor().execute(new Runnable() {
- @Override
- public void run() {
-
ctx.pipeline().fireUserEventTriggered(inputChannel);
- }
- });
+ if (ctx != null) {
--- End diff --
I actually preferred the previous implementation because it would throw in
case of invalid use, i.e. if the channel had not been active before. Or is
there a use case for `ctx == null` that is valid and where we really should
skip credit notification? But recall since we are only ever calling this when
the credit moves from 0 to 1 and therefore, if we ever skip one notification,
we will not be notified again!
> 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: 1.5.0
>
>
> 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)
