[
https://issues.apache.org/jira/browse/HIVE-18722?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16366166#comment-16366166
]
Gopal V edited comment on HIVE-18722 at 2/15/18 7:44 PM:
---------------------------------------------------------
SUM(hash()) out of a Group-by query is not expected to be consistent (since
JDK8), because it uses an iterator from a a HashMap
If you're seeing different results with {{set hive.map.aggr=false}}, then it
might be worth investigating.
Integer operations are commutative, if they are of the same type - except it
can be (long)(int)(a+b) & (long(a) + long(b)) within the partial result.
The GC canary in the group-by operator is the usual problem.
was (Author: gopalv):
SUM(hash()) out of a Group-by query is not expected to be consistent (since
JDK8), because it uses an iterator from a a HashMap
If you're seeing different results with {{set hive.map.aggr=false}}, then it
might be worth investigating.
> Vectorization: Adding SUM(HASH(..)) to full query seems to produce flakey
> results -- need to investiage
> -------------------------------------------------------------------------------------------------------
>
> Key: HIVE-18722
> URL: https://issues.apache.org/jira/browse/HIVE-18722
> Project: Hive
> Issue Type: Bug
> Components: Hive
> Affects Versions: 3.0.0
> Reporter: Matt McCline
> Assignee: Matt McCline
> Priority: Critical
>
> When added to HIVE-18622 changes, the query results vary from laptop results
> when run on Hive QA cluster. Need to investigate after HIVE-18622 commits.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
