Github user marmbrus commented on a diff in the pull request:
https://github.com/apache/spark/pull/5210#discussion_r27253297
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/parquet/newParquet.scala ---
@@ -435,11 +435,18 @@ private[sql] case class ParquetRelation2(
// Push down filters when possible. Notice that not all filters can be
converted to Parquet
// filter predicate. Here we try to convert each individual predicate
and only collect those
// convertible ones.
- predicates
- .flatMap(ParquetFilters.createFilter)
- .reduceOption(FilterApi.and)
- .filter(_ => sqlContext.conf.parquetFilterPushDown)
- .foreach(ParquetInputFormat.setFilterPredicate(jobConf, _))
+ if (sqlContext.conf.parquetFilterPushDown) {
+ predicates
+ // Don't push down predicates which reference partition columns
+ .filter { pred =>
+ val partitionColNames = partitionColumns.map(_.name).toSet
+ val referencedColNames = pred.references.map(_.name).toSet
+ referencedColNames.intersect(partitionColNames).isEmpty
+ }
+ .flatMap(ParquetFilters.createFilter)
--- End diff --
I think you guys are optimizing for the wrong thing here.
`reduceOption(<and>)` is an incredibly common pattern when you are munging
predicates. It occurs 15+ times in the code base.
In my experience, `reduce` is always a red flag and it means you aren't
considering what to do in the case that the list is empty. There have been
several bugs in catalyst where this was the problem.
You are correct, that you _can_ add an explicit if check. However, now as
I scan the code, I have to connect these two disparate lines and reason about
them independently. When a developer uses `reduceOption` I know that the
compiler is going to yell at them if they aren't handling all cases correctly,
so I can focus my review on the other logic.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
