[
https://issues.apache.org/jira/browse/KAFKA-14045?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
gdd sop updated KAFKA-14045:
----------------------------
Description:
1. Shut down all brokers
2. Execute
{code:java}
top {code}
command and found out the kafka coordinator CPU usage is high
[https://img-blog.csdnimg.cn/dbfcba626ff84058b908ef7ad7a0e801.png#pic_center]
[https://img-blog.csdnimg.cn/4fdf53e2aa7b4aa1bca1c5cb6d3cafd5.png#pic_center]
Here is the flame graph collectd by arthas.
[https://img-blog.csdnimg.cn/31f65768c1264cbea1d73305345647c5.png#pic_center]
The issues will recover after restart the brokers.
*Maybe it should wait before the pollNoWakeup methods*
[https://github.com/apache/kafka/blob/2.8/clients/src/main/java/org/apache/kafka/clients/consumer/internals/AbstractCoordinator.java#L1372]
was:
1. Shut down all brokers
2. Execute
{code:java}
top {code}
command and found out the kafka coordinator CPU usage is high
[https://img-blog.csdnimg.cn/dbfcba626ff84058b908ef7ad7a0e801.png#pic_center]
https://img-blog.csdnimg.cn/4fdf53e2aa7b4aa1bca1c5cb6d3cafd5.png#pic_center
Here is the flame graph collectd by arthas.
https://img-blog.csdnimg.cn/31f65768c1264cbea1d73305345647c5.png#pic_center
The issues will recover after restart the brokers.
{*}{{*}}Maybe it should wait before the pollNoWakeup methods
[https://github.com/apache/kafka/blob/2.8/clients/src/main/java/org/apache/kafka/clients/consumer/internals/AbstractCoordinator.java#L1372]
> Heartbeat threads cause high CPU usage after broker shut down
> --------------------------------------------------------------
>
> Key: KAFKA-14045
> URL: https://issues.apache.org/jira/browse/KAFKA-14045
> Project: Kafka
> Issue Type: Bug
> Reporter: gdd sop
> Priority: Major
>
> 1. Shut down all brokers
> 2. Execute
> {code:java}
> top {code}
> command and found out the kafka coordinator CPU usage is high
> [https://img-blog.csdnimg.cn/dbfcba626ff84058b908ef7ad7a0e801.png#pic_center]
> [https://img-blog.csdnimg.cn/4fdf53e2aa7b4aa1bca1c5cb6d3cafd5.png#pic_center]
>
> Here is the flame graph collectd by arthas.
> [https://img-blog.csdnimg.cn/31f65768c1264cbea1d73305345647c5.png#pic_center]
> The issues will recover after restart the brokers.
>
> *Maybe it should wait before the pollNoWakeup methods*
> [https://github.com/apache/kafka/blob/2.8/clients/src/main/java/org/apache/kafka/clients/consumer/internals/AbstractCoordinator.java#L1372]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
