[
https://issues.apache.org/jira/browse/FLINK-7416?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16274379#comment-16274379
]
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_r154338813
--- Diff:
flink-runtime/src/test/java/org/apache/flink/runtime/io/network/netty/PartitionRequestClientHandlerTest.java
---
@@ -244,54 +237,89 @@ public void testCancelBeforeActive() throws Exception
{
}
/**
- * Verifies that {@link RemoteInputChannel} is enqueued in the
pipeline, and
- * {@link AddCredit} message is sent to the producer.
+ * Verifies that {@link RemoteInputChannel} is enqueued in the pipeline
for notifying credits,
+ * and verifies the behaviour of credit notification by triggering
channel's writability changed.
*/
@Test
public void testNotifyCreditAvailable() throws Exception {
+ final SingleInputGate inputGate = createSingleInputGate();
+ final RemoteInputChannel inputChannel1 =
spy(createRemoteInputChannel(inputGate));
+ final RemoteInputChannel inputChannel2 =
spy(createRemoteInputChannel(inputGate));
final CreditBasedClientHandler handler = new
CreditBasedClientHandler();
- final EmbeddedChannel channel = new EmbeddedChannel(handler);
+ final EmbeddedChannel channel = spy(new
EmbeddedChannel(handler));
- final RemoteInputChannel inputChannel =
createRemoteInputChannel(mock(SingleInputGate.class));
+ // Increase the credits to enqueue the input channels
+ inputChannel1.increaseCredit(1);
+ inputChannel2.increaseCredit(1);
+ handler.notifyCreditAvailable(inputChannel1);
+ handler.notifyCreditAvailable(inputChannel2);
- // Enqueue the input channel
- handler.notifyCreditAvailable(inputChannel);
+ channel.runPendingTasks();
+
+ // The two input channels should notify credits via writable
channel
+ assertTrue(channel.isWritable());
+ assertEquals(channel.readOutbound().getClass(),
AddCredit.class);
+ verify(inputChannel1, times(1)).getAndResetCredit();
+ verify(inputChannel2, times(1)).getAndResetCredit();
+
+ final int highWaterMark =
channel.config().getWriteBufferHighWaterMark();
+ // Set the writer index to the high water mark to ensure that
all bytes are written
+ // to the wire although the buffer is "empty".
+
channel.write(Unpooled.buffer(highWaterMark).writerIndex(highWaterMark));
+
+ // Enqueue the input channel on the condition of un-writable
channel
+ inputChannel1.increaseCredit(1);
+ handler.notifyCreditAvailable(inputChannel1);
channel.runPendingTasks();
- // Read the enqueued msg
- Object msg1 = channel.readOutbound();
+ // The input channel will not notify credits via un-writable
channel
+ assertFalse(channel.isWritable());
+ verify(inputChannel1, times(1)).getAndResetCredit();
- // Should notify credit
- assertEquals(msg1.getClass(), AddCredit.class);
+ // Flush the buffer to make the channel writable again
+ channel.flush();
+
+ // The input channel should notify credits via triggering
channel's writability changed event
+ assertTrue(channel.isWritable());
+ assertEquals(channel.readOutbound().getClass(),
AddCredit.class);
--- End diff --
actually, here we receive the buffer that was blocking the channel first,
and then the `AddCredit` message - both should be checked appropriately (see
above)
> 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)
