[
https://issues.apache.org/jira/browse/SPARK-5511?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15557667#comment-15557667
]
Hyukjin Kwon commented on SPARK-5511:
-------------------------------------
1. I agree it needs a change on Spark.
2. We already supported this before via user-defiened filter but it was removed
due to the performance about filtering record-by-record. Then, there is a try
to add this back with the combinations of OR operators. See SPARK-17091
> [SQL] Possible optimisations for predicate pushdowns from Spark SQL to Parquet
> ------------------------------------------------------------------------------
>
> Key: SPARK-5511
> URL: https://issues.apache.org/jira/browse/SPARK-5511
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 1.2.0
> Reporter: Mick Davies
> Priority: Minor
>
> The following changes could make predicate pushdown more effective under
> certain conditions, which are not uncommon.
> 1. Parquet predicate evaluation does not use dictionary compression
> information, furthermore it circumvents dictionary decoding optimisations
> (https://issues.apache.org/jira/browse/PARQUET-36). This means predicates are
> re-evaluated repeatedly for the same Strings, and also Binary->String
> conversions are repeated. This is a change purely on the Parquet side.
> 2. Support IN clauses in predicate pushdown. This requires changes to Parquet
> and then subsequently in Spark SQL.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
