[
https://issues.apache.org/jira/browse/KAFKA-14112?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17733965#comment-17733965
]
Elkhan Eminov commented on KAFKA-14112:
---------------------------------------
[~viktorsomogyi] I remember implementing this back in CLDR and I copied the
ticket desc. over here, but it seems like the description is incorrect, it
states that the difference is LRO-LEO but in fact this doesn't make sense to
me, it should be LEO-LRO. can I get your opinion on this pls?
> Expose replication-offset-lag Mirror metric
> -------------------------------------------
>
> Key: KAFKA-14112
> URL: https://issues.apache.org/jira/browse/KAFKA-14112
> Project: Kafka
> Issue Type: Improvement
> Components: mirrormaker
> Reporter: Elkhan Eminov
> Assignee: Elkhan Eminov
> Priority: Minor
>
> The offset lag is the difference of the last replicated record's (LRO) source
> offset and the end offset of the source (LEO).
> The offset lag is a difference (LRO-LEO), but its constituents calculated at
> different points of time and place
> * LEO shall be calculated during source task's poll loop (ready to get it
> from the consumer)
> * LRO shall be kept in an in-memory "cache", that is updated during the
> task's producer callback
> LRO is initialized when task is started, from the offset store. The
> difference shall be calculated when the freshest LEO acquired
> in the poll loop. The calculated amount shall be defined as a MirrorMaker
> metric.
> This would describe to amount of "to be replicated" number of records for a
> certain topic-partition.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
