[
https://issues.apache.org/jira/browse/KAFKA-6011?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16192105#comment-16192105
]
Ewen Cheslack-Postava commented on KAFKA-6011:
----------------------------------------------
[~rsivaram] Fair point, I see that we have app-info for both client-id= and id=
which seems a bit silly. We'd need to deprecate to to remove one of them. For
connect we're currently registering by worker ID which is not just a client ID,
so this would actually be a different bean. But really it just seems like
overkill to even record this for every client. I guess with multiple
classloaders you could end up with different versions in the same process, I
guess maybe that's the motivation for not having a single global metric?
> AppInfoParser should only use metrics API and should not register JMX mbeans
> directly
> -------------------------------------------------------------------------------------
>
> Key: KAFKA-6011
> URL: https://issues.apache.org/jira/browse/KAFKA-6011
> Project: Kafka
> Issue Type: Bug
> Components: metrics
> Reporter: Ewen Cheslack-Postava
> Priority: Minor
>
> AppInfoParser collects info about the app ID, version, and commit ID and logs
> them + exposes corresponding metrics. For some reason we ended up with the
> app ID metric being registered directly to JMX while the version and commit
> ID use the metrics API. This means the app ID would not be accessible to
> custom metrics reporter.
> This isn't a huge loss as this is probably a rarely used metric, but we
> should really only be using the metrics API. Only using the metrics API would
> also reduce and centralize the places we need to do name mangling to handle
> characters that might not be valid for metrics.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
