[
https://issues.apache.org/jira/browse/HIVE-15511?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Rajesh Balamohan updated HIVE-15511:
------------------------------------
Attachment: HIVE-15511.1.patch
Sample query: select distinct CAST(MONTHS_BETWEEN(l_shipdate, l_commitdate) /
12 AS INT) from lineitem_date limit 10;
l_shipdate, l_commitdate are date columns in the table. This was run on a
medium scale cluster with 1 TB tpch dataset (HIVE-15501 is included which
enables vectorization for months_between udf).
||Iteration||with rounding off (default))||Without round off||
|Run 1|92.60 s|61.22 s|
|Run 2|87.39 s|59.45 s|
|Run 3|87.90 s|59.22 s|
> Provide an option in months_between UDF to disable rounding-off
> ---------------------------------------------------------------
>
> Key: HIVE-15511
> URL: https://issues.apache.org/jira/browse/HIVE-15511
> Project: Hive
> Issue Type: Improvement
> Reporter: Rajesh Balamohan
> Priority: Minor
> Attachments: HIVE-15511.1.patch
>
>
> Rounding off was added in {{GenericUDFMonthsBetween}} so that it can be
> compatible with systems like oracle. However, there are places where rounding
> off is not needed.
> E.g "CAST(MONTHS_BETWEEN(l_shipdate, l_commitdate) / 12 AS INT)" may not need
> rounding off via BigDecimal which is compute intensive.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
