[
https://issues.apache.org/jira/browse/KAFKA-14868?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
hudeqi updated KAFKA-14868:
---------------------------
Description:
In some current classes, some metrics are always registered when loading, but
it often occurs that some or all metrics are forgotten to be removed when
closing. For example, in the ReplicaManager class, IsrExpandsPerSec,
IsrShrinksPerSec, and FailedIsrUpdatesPerSec are all removed by omission.
The problem of not closing the metric in the replicaManager has been solved in
the patch. As for how to improve the KafkaMetricsGroup to make the inspection
of such issues more general, there is no conclusive solution.
was:
In some current classes, some metrics are always registered when loading, but
it often occurs that some or all metrics are forgotten to be removed when
closing. For example, in the ReplicaManager class, IsrExpandsPerSec,
IsrShrinksPerSec, and FailedIsrUpdatesPerSec are all removed by omission.
Therefore, increase the counting capability in KafkaMetricsGroup. When using
the method of KafkaMetricsGroup to add or remove a metric, it will add 1 to the
corresponding counter, so that you can check whether there is a forgotten
metric to be removed by counting and comparing in the unit test of the
corresponding class.
> Remove some forgotten metrics when the replicaManager is closed
> ---------------------------------------------------------------
>
> Key: KAFKA-14868
> URL: https://issues.apache.org/jira/browse/KAFKA-14868
> Project: Kafka
> Issue Type: Improvement
> Reporter: hudeqi
> Assignee: hudeqi
> Priority: Major
>
> In some current classes, some metrics are always registered when loading, but
> it often occurs that some or all metrics are forgotten to be removed when
> closing. For example, in the ReplicaManager class, IsrExpandsPerSec,
> IsrShrinksPerSec, and FailedIsrUpdatesPerSec are all removed by omission.
> The problem of not closing the metric in the replicaManager has been solved
> in the patch. As for how to improve the KafkaMetricsGroup to make the
> inspection of such issues more general, there is no conclusive solution.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
