[jira] [Commented] (KAFKA-15250) ConsumerNetworkThread is running tight loop

2024-05-22 Thread Philip Nee (Jira) 


[ 
https://issues.apache.org/jira/browse/KAFKA-15250?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17848703#comment-17848703
 ] 

Philip Nee commented on KAFKA-15250:


After the inflight check fix. I think most request managers are backing up 
correctly.  Closing this issue.

> ConsumerNetworkThread is running tight loop
> ---
>
> Key: KAFKA-15250
> URL: https://issues.apache.org/jira/browse/KAFKA-15250
> Project: Kafka
>  Issue Type: Bug
>  Components: clients, consumer
>Reporter: Philip Nee
>Assignee: Philip Nee
>Priority: Critical
>  Labels: consumer-threading-refactor, events
> Fix For: 3.8.0
>
>
> The DefaultBackgroundThread is running tight loops and wasting CPU cycles.  I 
> think we need to reexamine the timeout pass to networkclientDelegate.poll.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

[jira] [Commented] (KAFKA-15250) ConsumerNetworkThread is running tight loop

2024-04-16 Thread Philip Nee (Jira) 


[ 
https://issues.apache.org/jira/browse/KAFKA-15250?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17837836#comment-17837836
 ] 

Philip Nee commented on KAFKA-15250:


Added a testing branch under my fork: network-thread-mbean-metrics

 

I ran the testAsyncCommit in the integration test (because it actually polls).  
These are the stats:

backoff time avg:0.29459381324082107
(backoff time max,92.0)
poll time avg:9.2161365261735107E18
(poll time max,9.223372036854776E18)

 

The loop doesn't really backoff.  

> ConsumerNetworkThread is running tight loop
> ---
>
> Key: KAFKA-15250
> URL: https://issues.apache.org/jira/browse/KAFKA-15250
> Project: Kafka
>  Issue Type: Bug
>  Components: clients, consumer
>Reporter: Philip Nee
>Assignee: Kirk True
>Priority: Critical
>  Labels: consumer-threading-refactor, events
> Fix For: 3.8.0
>
>
> The DefaultBackgroundThread is running tight loops and wasting CPU cycles.  I 
> think we need to reexamine the timeout pass to networkclientDelegate.poll.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)