[
https://issues.apache.org/jira/browse/SPARK-9258?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Reynold Xin updated SPARK-9258:
-------------------------------
Description:
We have 4 semi join operators. In this case, they are not very very necessary.
We can still use an equi-join operator to do the join, and just not include any
values from the other join.
We waste a little bit space due to building a hash map rather than a hash set,
but at the end of the day unless we are going to spend a lot of time optimizing
hash set, our Tungsten hash map will be a lot more efficient than the hash set
anyway. This way, semi-join automatically benefits from all the work we do in
Tungsten.
was:
We have too many join operators than our resources to optimize them. In this
case, BroadcastLeftSemiJoinHash isn't very necessary. We can still use an
equi-join operator to do the join, and just not include any values from the
other join.
We waste a little bit space due to building a hash map rather than a hash set,
but at the end of the day unless we are going to spend a lot of time optimizing
hash set, our Tungsten hash map will be a lot more efficient than the hash set
anyway ...
> Remove all semi join physical operator
> --------------------------------------
>
> Key: SPARK-9258
> URL: https://issues.apache.org/jira/browse/SPARK-9258
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Reporter: Reynold Xin
>
> We have 4 semi join operators. In this case, they are not very very
> necessary. We can still use an equi-join operator to do the join, and just
> not include any values from the other join.
> We waste a little bit space due to building a hash map rather than a hash
> set, but at the end of the day unless we are going to spend a lot of time
> optimizing hash set, our Tungsten hash map will be a lot more efficient than
> the hash set anyway. This way, semi-join automatically benefits from all the
> work we do in Tungsten.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
