diff options
author | Cheng Lian <lian@databricks.com> | 2015-09-18 18:42:20 -0700 |
---|---|---|
committer | Yin Huai <yhuai@databricks.com> | 2015-09-18 18:42:20 -0700 |
commit | 22be2ae147a111e88896f6fb42ed46bbf108a99b (patch) | |
tree | 37c48798e53746296a01e6b8c896129d22854f65 /sql/hive/src/main | |
parent | c8149ef2c57f5c47ab97ee8d8d58a216d4bc4156 (diff) | |
download | spark-22be2ae147a111e88896f6fb42ed46bbf108a99b.tar.gz spark-22be2ae147a111e88896f6fb42ed46bbf108a99b.tar.bz2 spark-22be2ae147a111e88896f6fb42ed46bbf108a99b.zip |
[SPARK-10623] [SQL] Fixes ORC predicate push-down
When pushing down a leaf predicate, ORC `SearchArgument` builder requires an extra "parent" predicate (any one among `AND`/`OR`/`NOT`) to wrap the leaf predicate. E.g., to push down `a < 1`, we must build `AND(a < 1)` instead. Fortunately, when actually constructing the `SearchArgument`, the builder will eliminate all those unnecessary wrappers.
This PR is based on #8783 authored by zhzhan. I also took the chance to simply `OrcFilters` a little bit to improve readability.
Author: Cheng Lian <lian@databricks.com>
Closes #8799 from liancheng/spark-10623/fix-orc-ppd.
Diffstat (limited to 'sql/hive/src/main')
-rw-r--r-- | sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcFilters.scala | 56 |
1 files changed, 22 insertions, 34 deletions
diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcFilters.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcFilters.scala index b3d9f7f71a..27193f54d3 100644 --- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcFilters.scala +++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcFilters.scala @@ -31,11 +31,13 @@ import org.apache.spark.sql.sources._ * and cannot be used anymore. */ private[orc] object OrcFilters extends Logging { - def createFilter(expr: Array[Filter]): Option[SearchArgument] = { - expr.reduceOption(And).flatMap { conjunction => - val builder = SearchArgumentFactory.newBuilder() - buildSearchArgument(conjunction, builder).map(_.build()) - } + def createFilter(filters: Array[Filter]): Option[SearchArgument] = { + for { + // Combines all filters with `And`s to produce a single conjunction predicate + conjunction <- filters.reduceOption(And) + // Then tries to build a single ORC `SearchArgument` for the conjunction predicate + builder <- buildSearchArgument(conjunction, SearchArgumentFactory.newBuilder()) + } yield builder.build() } private def buildSearchArgument(expression: Filter, builder: Builder): Option[Builder] = { @@ -102,46 +104,32 @@ private[orc] object OrcFilters extends Logging { negate <- buildSearchArgument(child, builder.startNot()) } yield negate.end() - case EqualTo(attribute, value) => - Option(value) - .filter(isSearchableLiteral) - .map(builder.equals(attribute, _)) + case EqualTo(attribute, value) if isSearchableLiteral(value) => + Some(builder.startAnd().equals(attribute, value).end()) - case EqualNullSafe(attribute, value) => - Option(value) - .filter(isSearchableLiteral) - .map(builder.nullSafeEquals(attribute, _)) + case EqualNullSafe(attribute, value) if isSearchableLiteral(value) => + Some(builder.startAnd().nullSafeEquals(attribute, value).end()) - case LessThan(attribute, value) => - Option(value) - .filter(isSearchableLiteral) - .map(builder.lessThan(attribute, _)) + case LessThan(attribute, value) if isSearchableLiteral(value) => + Some(builder.startAnd().lessThan(attribute, value).end()) - case LessThanOrEqual(attribute, value) => - Option(value) - .filter(isSearchableLiteral) - .map(builder.lessThanEquals(attribute, _)) + case LessThanOrEqual(attribute, value) if isSearchableLiteral(value) => + Some(builder.startAnd().lessThanEquals(attribute, value).end()) - case GreaterThan(attribute, value) => - Option(value) - .filter(isSearchableLiteral) - .map(builder.startNot().lessThanEquals(attribute, _).end()) + case GreaterThan(attribute, value) if isSearchableLiteral(value) => + Some(builder.startNot().lessThanEquals(attribute, value).end()) - case GreaterThanOrEqual(attribute, value) => - Option(value) - .filter(isSearchableLiteral) - .map(builder.startNot().lessThan(attribute, _).end()) + case GreaterThanOrEqual(attribute, value) if isSearchableLiteral(value) => + Some(builder.startNot().lessThan(attribute, value).end()) case IsNull(attribute) => - Some(builder.isNull(attribute)) + Some(builder.startAnd().isNull(attribute).end()) case IsNotNull(attribute) => Some(builder.startNot().isNull(attribute).end()) - case In(attribute, values) => - Option(values) - .filter(_.forall(isSearchableLiteral)) - .map(builder.in(attribute, _)) + case In(attribute, values) if values.forall(isSearchableLiteral) => + Some(builder.startAnd().in(attribute, values.map(_.asInstanceOf[AnyRef]): _*).end()) case _ => None } |