[
https://issues.apache.org/jira/browse/IMPALA-7219?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16527956#comment-16527956
]
Misha Dmitriev commented on IMPALA-7219:
----------------------------------------
[~tlipcon] correct, this is from the catalog server with ~8M tables. I agree
that the amount of waste of this kind is proportional to the number of tables,
so in most cases the savings would be (much) smaller. On the other hand, no one
can guarantee that some customer will not try to run with 15M tables tomorrow.
> 7.5% of Catalog Server heap wasted by empty HashMaps and ArrayLists
> -------------------------------------------------------------------
>
> Key: IMPALA-7219
> URL: https://issues.apache.org/jira/browse/IMPALA-7219
> Project: IMPALA
> Issue Type: Improvement
> Reporter: Misha Dmitriev
> Priority: Major
> Attachments: bad-colls-screenshot.png
>
>
> I recently analyzed a ~16GB heap dump of Catalog Server with jxray
> ([www.jxray.com).|http://www.jxray.com)./] It turns out that 7.5% of the heap
> is wasted by empty HashMap and ArrayList instances.
> !bad-colls-screenshot.png|width=464,height=531!
> These objects come from just four data fields in two classes:
> {{IncompleteTable}} and {{StructType}}. It should be easy to get rid of the
> overhead of these objects by initializing them lazily.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscr...@impala.apache.org
For additional commands, e-mail: issues-all-h...@impala.apache.org
