phoenix-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Vincent Poon (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (PHOENIX-5131) Make spilling to disk for order/group by configurable
Date Fri, 08 Mar 2019 20:29:00 GMT

    [ https://issues.apache.org/jira/browse/PHOENIX-5131?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16788273#comment-16788273
] 

Vincent Poon commented on PHOENIX-5131:
---------------------------------------

[~abhishek.chouhan] for SizeBoundQueue, do we need to decrement currentSize when polling?
 I guess in your usage it might not matter as you only offer once?  but might still be good
to have.

Is there any difference between `context.getConnection().getQueryServices().getProps()` and
`services.getProps()` ?  Also, can we cache these properties in a field at factory creation,
or are you expecting them to be dynamically changed?

Also, I notice in some places thresholdByes is an int, and in others a long.  We should probably
keep it long everywhere


> Make spilling to disk for order/group by configurable
> -----------------------------------------------------
>
>                 Key: PHOENIX-5131
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-5131
>             Project: Phoenix
>          Issue Type: Improvement
>            Reporter: Abhishek Singh Chouhan
>            Assignee: Abhishek Singh Chouhan
>            Priority: Major
>             Fix For: 4.15.0, 5.1.0
>
>         Attachments: PHOENIX-5131-master-v2.patch, PHOENIX-5131-master-v2.patch, PHOENIX-5131-master.patch,
PHOENIX-5131-master.patch
>
>
> We've observed that large queries, doing order/group by leading to issues on the regionserver
(crashes/long gc pauses/file handler exhaustion etc.). We should make spilling to disk configurable
and in case its disabled, fail the query once it hits the spilling limit on any of the region
servers. Also make spooling threshold server-side property only to prevent clients from controlling
memory allocation on the rs side.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message