[
https://issues.apache.org/jira/browse/HIVE-30120?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Denys Kuzmenko updated HIVE-30120:
----------------------------------
Description:
hive-mr3 fork adds VectorHashKeyWrapperSingleString, TwoString, ThreeString,
SingleLongSingleString, SingleLongTwoString, TwoLongSingleString, ThreeLong and
GeneralLongString. VectorHashKeyWrapperFactory.allocate picks one of them by
key shape.
They were never contributed upstream, which only has the single- and two-long
versions.
- master 50.8 ns/row;
- HIVE-30107 15.7;
- HIVE-30118 9.6;
- HIVE-30119 2.7 (1TRC part1 shape, local);
- 1TRC part1 on the cluster 2.4× faster with HIVE-30107 alone.
was:
hive-mr3 fork adds VectorHashKeyWrapperSingleString, TwoString, ThreeString,
SingleLongSingleString, SingleLongTwoString, TwoLongSingleString, ThreeLong and
GeneralLongString. VectorHashKeyWrapperFactory.allocate picks one of them by
key shape.
They were never contributed upstream, which only has the single- and two-long
versions.
> Vectorization: speed up hash GROUP BY on string keys
> ----------------------------------------------------
>
> Key: HIVE-30120
> URL: https://issues.apache.org/jira/browse/HIVE-30120
> Project: Hive
> Issue Type: Improvement
> Reporter: Denys Kuzmenko
> Priority: Major
>
> hive-mr3 fork adds VectorHashKeyWrapperSingleString, TwoString, ThreeString,
> SingleLongSingleString, SingleLongTwoString, TwoLongSingleString, ThreeLong
> and GeneralLongString. VectorHashKeyWrapperFactory.allocate picks one of them
> by key shape.
> They were never contributed upstream, which only has the single- and two-long
> versions.
>
> - master 50.8 ns/row;
>
>
> - HIVE-30107 15.7;
>
>
> - HIVE-30118 9.6;
>
>
> - HIVE-30119 2.7 (1TRC part1 shape, local);
>
>
> - 1TRC part1 on the cluster 2.4× faster with HIVE-30107 alone.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
