[
https://issues.apache.org/jira/browse/FLINK-10022?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16579634#comment-16579634
]
ASF GitHub Bot commented on FLINK-10022:
----------------------------------------
zentol commented on a change in pull request #6551:
[FLINK-10022][network][metrics] add metrics for input/output buffers
URL: https://github.com/apache/flink/pull/6551#discussion_r209906457
##########
File path:
flink-runtime/src/main/java/org/apache/flink/runtime/executiongraph/IOMetrics.java
##########
@@ -43,38 +43,74 @@
protected double numBytesInRemotePerSecond;
protected double numBytesOutPerSecond;
- public IOMetrics(Meter recordsIn, Meter recordsOut, Meter bytesLocalIn,
Meter bytesRemoteIn, Meter bytesOut) {
+ protected long numBuffersInLocal;
+ protected long numBuffersInRemote;
+ protected long numBuffersOut;
+
+ protected double numBuffersInLocalPerSecond;
+ protected double numBuffersInRemotePerSecond;
+ protected double numBuffersOutPerSecond;
+
+ public IOMetrics(
Review comment:
so this here is really the big question mark for me. So far we only kept
metrics in here that are displayed in the UI, but with this PR we'd break that.
This is mostly done to have accurate displays at the end of the job, which in
particular in the case of batch jobs is important.
The question is, do we truly need 100% accurate buffer metrics here? For
_debugging_ purposes there are always other avenues one can take to get
accurate readings. (delay shutdown of job until metrics are updated)
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
> Add metrics for input/output buffers
> ------------------------------------
>
> Key: FLINK-10022
> URL: https://issues.apache.org/jira/browse/FLINK-10022
> Project: Flink
> Issue Type: Improvement
> Components: Metrics, Network
> Affects Versions: 1.5.2, 1.6.0, 1.7.0
> Reporter: Nico Kruber
> Assignee: Nico Kruber
> Priority: Major
> Labels: pull-request-available
>
> Currently, we provide metrics for the input and output queue lengths, records
> and bytes. For diagnosing network issues, it would also be nice to know the
> number of buffers because we would then know their fill rate.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
