[
https://issues.apache.org/jira/browse/PHOENIX-2800?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15213164#comment-15213164
]
ASF GitHub Bot commented on PHOENIX-2800:
-----------------------------------------
Github user JamesRTaylor commented on the pull request:
https://github.com/apache/phoenix/commit/48e589773cbf46a10a2c1bd5cf483f2390ae1160#commitcomment-16855409
When a group by is being performed, unless the group by key is made up of
the leading PK columns of a table or index, Phoenix holds in memory each
distinct group by key and the partial aggregation. This will spill to disk
based on phoenix.groupby.maxCacheSize, but performance will drop in this case.
There may be other in memory representations we can use to reduce the memory
footprint. I've filed PHOENIX-2800 for some potential future work. Let me know
if you'd be interested in contributing that.
> Use Apache Arrow for in memory representation for first level aggregation on
> RS
> -------------------------------------------------------------------------------
>
> Key: PHOENIX-2800
> URL: https://issues.apache.org/jira/browse/PHOENIX-2800
> Project: Phoenix
> Issue Type: Improvement
> Reporter: James Taylor
>
> To reduce memory consumption, we should use Apache Arrow
> (https://arrow.apache.org/) format when we're doing the first level
> aggregation on the region server.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
