[
https://issues.apache.org/jira/browse/KAFKA-1800?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14233813#comment-14233813
]
Guozhang Wang commented on KAFKA-1800:
--------------------------------------
Copying [~jjkoshy]'s comment on the RB here:
{code}
It would be useful to clarify the comment on why this needed to be moved
further up as you explained offline - i.e., since buffer exhaustion (for
example) can happen before the sender gets a chance to register the metrics.
Also, we should probably discuss on the jira the additional caveat of failed
metadata fetches. i.e., since that happens in the network-client the true
record error rate would be higher than what's counted by sendermetrics.
The options that we have are:
* Expose Sender's maybeRegisterTopicMetrics and use that in NetworkClient
maybeUpdateMetadata if there are no known partitions for a topic
* Keep it as you have it for now and just accept the above discrepancy - (or we
could address that in a separate jira as it is orthogonal).
{code}
> KafkaException was not recorded at the per-topic metrics
> --------------------------------------------------------
>
> Key: KAFKA-1800
> URL: https://issues.apache.org/jira/browse/KAFKA-1800
> Project: Kafka
> Issue Type: Bug
> Reporter: Guozhang Wang
> Assignee: Guozhang Wang
> Fix For: 0.9.0
>
> Attachments: KAFKA-1800.patch
>
>
> When KafkaException was thrown from producer.send() call, it is not recorded
> on the per-topic record-error-rate, but only the global error-rate.
> Since users are usually monitoring on the per-topic metrics, loosing all
> dropped message counts at this level that are caused by kafka producer thrown
> exceptions such as BufferExhaustedException could be very dangerous.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
