[
https://issues.apache.org/jira/browse/KAFKA-6826?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Xavier Léauté updated KAFKA-6826:
---------------------------------
Description:
This is a follow-up to KAFKA-6560, where we missed at least one case that
should be using single point queries instead of range-scans when forwarding
values during aggregation.
Since a single range scan can sometimes account for 75% of aggregation cpu
time, fixing this should provide some significant speedups (see attached
flamegraph)
!Screen Shot 2018-04-25 at 11.14.25 AM.png|width=797!
was:
This is a follow-up to KAFKA-6560, where we missed at least one case that
should be using single point queries instead of range-scans when forwarding
values during aggregation.
Since a single range scan can sometimes account for 75% of aggregation cpu
time, fixing this should provide some significant speedups (see attached
flamegraph)
!Screen Shot 2018-04-25 at 11.14.25 AM.png!
> Avoid range scans when forwarding values in window store aggregations
> ---------------------------------------------------------------------
>
> Key: KAFKA-6826
> URL: https://issues.apache.org/jira/browse/KAFKA-6826
> Project: Kafka
> Issue Type: Bug
> Components: streams
> Reporter: Xavier Léauté
> Assignee: Xavier Léauté
> Priority: Critical
> Attachments: Screen Shot 2018-04-25 at 11.14.25 AM.png
>
>
> This is a follow-up to KAFKA-6560, where we missed at least one case that
> should be using single point queries instead of range-scans when forwarding
> values during aggregation.
> Since a single range scan can sometimes account for 75% of aggregation cpu
> time, fixing this should provide some significant speedups (see attached
> flamegraph)
> !Screen Shot 2018-04-25 at 11.14.25 AM.png|width=797!
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
