[
https://issues.apache.org/jira/browse/PHOENIX-4718?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16459211#comment-16459211
]
James Taylor edited comment on PHOENIX-4718 at 4/30/18 10:46 PM:
-----------------------------------------------------------------
Please review, [~tdsilva]. Instead of always tracking memory used by an
Aggregator, we only do it if we're using an Aggregator in which trackSize()
returns true. [~mujtabachohan] - can you try with this patch and see if perf is
better again? Let me attach an 0.98 version of this patch, though.
was (Author: jamestaylor):
Please review, [~tdsilva]. Instead of always tracking memory used by an
Aggregator, we only do it if we're using an Aggregator in which trackSize()
returns true. [~mujtabachohan] - can you try with this patch and see if perf is
better again?
> Decrease overhead of tracking aggregate heap size
> -------------------------------------------------
>
> Key: PHOENIX-4718
> URL: https://issues.apache.org/jira/browse/PHOENIX-4718
> Project: Phoenix
> Issue Type: Bug
> Reporter: James Taylor
> Assignee: James Taylor
> Priority: Major
> Attachments: PHOENIX-4718.patch
>
>
> Since PHOENIX-4148, we track the heap size while aggregation is occurring.
> This decreased performance of aggregation by ~20%. We really only need to
> track this for the DistinctValueWithCountServerAggregator (used by DISTINCT
> COUNT, DISTINCT, PERCENTILE functions, and STDDEV functions). By
> conditionally tracking, we should be able to bring perf back to what it was
> before.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
