[
https://issues.apache.org/jira/browse/SPARK-46946?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Thang Long Vu updated SPARK-46946:
----------------------------------
Description:
This PR extends `DynamicPruningSubquery` to support broadcasting of multiple
filtering keys (instead of one as before). The majority of the PR is to simply
generalise singularity to plurality.
*{*}Note:{*} We actually do not use the multiple filtering keys
`DynamicPruningSubquery` in this PR, we are doing this to make supporting DPP
Null Safe Equality or multiple Equality predicates easier in the future.
In Null Safe Equality JOIN, the JOIN condition `a <=> b` is transformed to
`Coalesce(key1, Literal(key1.dataType)) = Coalesce(key2,
Literal(key2.dataType)) AND IsNull(key1) = IsNull(key2)`. In order to have the
highest pruning efficiency, we broadcast the 2 keys `Coalesce(key,
Literal(key.dataType))` and `IsNull(key)` and use them to prune the other side
at the same time.
Before, the `DynamicPruningSubquery` only has one broadcasting key and we only
supports DPP for one `EqualTo` JOIN predicate, now we are extending the
subquery to multiple broadcasting keys. Please note that DPP has not been
supported for multiple JOIN predicates.
Put it in another way, at the moment, we don't insert a DPP Filter for multiple
JOIN predicates at the same time, only potentially insert a DPP Filter for a
given Equality JOIN predicate.
was:
This PR extends `DynamicPruningSubquery` to support broadcasting of multiple
filtering keys (instead of one as before). The majority of the PR is to simply
generalise singularity to plurality.
**Note:** We actually do not use the multiple filtering keys
`DynamicPruningSubquery` in this PR, we are doing this to make supporting DPP
Null Safe Equality or multiple Equality predicates easier in the future.
In Null Safe Equality JOIN, the JOIN condition `a <=> b` is transformed to
`Coalesce(key1, Literal(key1.dataType)) = Coalesce(key2,
Literal(key2.dataType)) AND IsNull(key1) = IsNull(key2)`. In order to have the
highest pruning efficiency, we broadcast the 2 keys `Coalesce(key,
Literal(key.dataType))` and `IsNull(key)` and use them to prune the other side
at the same time.
Before, the `DynamicPruningSubquery` only has one broadcasting key and we only
supports DPP for one `EqualTo` JOIN predicate, now we are extending the
subquery to multiple broadcasting keys. Please note that DPP has not been
supported for multiple JOIN predicates.
Put it in another way, at the moment, we don't insert a DPP Filter for multiple
JOIN predicates at the same time, only potentially insert a DPP Filter for a
given Equality JOIN predicate.
> Supporting broadcast of multiple filtering keys in DynamicPruning
> -----------------------------------------------------------------
>
> Key: SPARK-46946
> URL: https://issues.apache.org/jira/browse/SPARK-46946
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 4.0.0, 3.5.1
> Reporter: Thang Long Vu
> Priority: Major
>
> This PR extends `DynamicPruningSubquery` to support broadcasting of multiple
> filtering keys (instead of one as before). The majority of the PR is to
> simply generalise singularity to plurality.
> *{*}Note:{*} We actually do not use the multiple filtering keys
> `DynamicPruningSubquery` in this PR, we are doing this to make supporting DPP
> Null Safe Equality or multiple Equality predicates easier in the future.
> In Null Safe Equality JOIN, the JOIN condition `a <=> b` is transformed to
> `Coalesce(key1, Literal(key1.dataType)) = Coalesce(key2,
> Literal(key2.dataType)) AND IsNull(key1) = IsNull(key2)`. In order to have
> the highest pruning efficiency, we broadcast the 2 keys `Coalesce(key,
> Literal(key.dataType))` and `IsNull(key)` and use them to prune the other
> side at the same time.
> Before, the `DynamicPruningSubquery` only has one broadcasting key and we
> only supports DPP for one `EqualTo` JOIN predicate, now we are extending the
> subquery to multiple broadcasting keys. Please note that DPP has not been
> supported for multiple JOIN predicates.
> Put it in another way, at the moment, we don't insert a DPP Filter for
> multiple JOIN predicates at the same time, only potentially insert a DPP
> Filter for a given Equality JOIN predicate.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
