diff options
author | Sameer Agarwal <sameer@databricks.com> | 2016-02-02 22:22:50 -0800 |
---|---|---|
committer | Michael Armbrust <michael@databricks.com> | 2016-02-02 22:22:50 -0800 |
commit | 138c300f97d29cb0d04a70bea98a8a0c0548318a (patch) | |
tree | c9271588249560b21cccf5dc92a618360d8a66be /sql | |
parent | e86f8f63bfa3c15659b94e831b853b1bc9ddae32 (diff) | |
download | spark-138c300f97d29cb0d04a70bea98a8a0c0548318a.tar.gz spark-138c300f97d29cb0d04a70bea98a8a0c0548318a.tar.bz2 spark-138c300f97d29cb0d04a70bea98a8a0c0548318a.zip |
[SPARK-12957][SQL] Initial support for constraint propagation in SparkSQL
Based on the semantics of a query, we can derive a number of data constraints on output of each (logical or physical) operator. For instance, if a filter defines `‘a > 10`, we know that the output data of this filter satisfies 2 constraints:
1. `‘a > 10`
2. `isNotNull(‘a)`
This PR proposes a possible way of keeping track of these constraints and propagating them in the logical plan, which can then help us build more advanced optimizations (such as pruning redundant filters, optimizing joins, among others). We define constraints as a set of (implicitly conjunctive) expressions. For e.g., if a filter operator has constraints = `Set(‘a > 10, ‘b < 100)`, it’s implied that the outputs satisfy both individual constraints (i.e., `‘a > 10` AND `‘b < 100`).
Design Document: https://docs.google.com/a/databricks.com/document/d/1WQRgDurUBV9Y6CWOBS75PQIqJwT-6WftVa18xzm7nCo/edit?usp=sharing
Author: Sameer Agarwal <sameer@databricks.com>
Closes #10844 from sameeragarwal/constraints.
Diffstat (limited to 'sql')
4 files changed, 302 insertions, 7 deletions
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala index b43b7ee71e..05f5bdbfc0 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala @@ -17,7 +17,7 @@ package org.apache.spark.sql.catalyst.plans -import org.apache.spark.sql.catalyst.expressions.{Attribute, AttributeSet, Expression, VirtualColumn} +import org.apache.spark.sql.catalyst.expressions._ import org.apache.spark.sql.catalyst.trees.TreeNode import org.apache.spark.sql.types.{DataType, StructType} @@ -27,6 +27,56 @@ abstract class QueryPlan[PlanType <: TreeNode[PlanType]] extends TreeNode[PlanTy def output: Seq[Attribute] /** + * Extracts the relevant constraints from a given set of constraints based on the attributes that + * appear in the [[outputSet]]. + */ + protected def getRelevantConstraints(constraints: Set[Expression]): Set[Expression] = { + constraints + .union(constructIsNotNullConstraints(constraints)) + .filter(constraint => + constraint.references.nonEmpty && constraint.references.subsetOf(outputSet)) + } + + /** + * Infers a set of `isNotNull` constraints from a given set of equality/comparison expressions. + * For e.g., if an expression is of the form (`a > 5`), this returns a constraint of the form + * `isNotNull(a)` + */ + private def constructIsNotNullConstraints(constraints: Set[Expression]): Set[Expression] = { + // Currently we only propagate constraints if the condition consists of equality + // and ranges. For all other cases, we return an empty set of constraints + constraints.map { + case EqualTo(l, r) => + Set(IsNotNull(l), IsNotNull(r)) + case GreaterThan(l, r) => + Set(IsNotNull(l), IsNotNull(r)) + case GreaterThanOrEqual(l, r) => + Set(IsNotNull(l), IsNotNull(r)) + case LessThan(l, r) => + Set(IsNotNull(l), IsNotNull(r)) + case LessThanOrEqual(l, r) => + Set(IsNotNull(l), IsNotNull(r)) + case _ => + Set.empty[Expression] + }.foldLeft(Set.empty[Expression])(_ union _.toSet) + } + + /** + * A sequence of expressions that describes the data property of the output rows of this + * operator. For example, if the output of this operator is column `a`, an example `constraints` + * can be `Set(a > 10, a < 20)`. + */ + lazy val constraints: Set[Expression] = getRelevantConstraints(validConstraints) + + /** + * This method can be overridden by any child class of QueryPlan to specify a set of constraints + * based on the given operator's constraint propagation logic. These constraints are then + * canonicalized and filtered automatically to contain only those attributes that appear in the + * [[outputSet]] + */ + protected def validConstraints: Set[Expression] = Set.empty + + /** * Returns the set of attributes that are output by this node. */ def outputSet: AttributeSet = AttributeSet(output) @@ -59,6 +109,7 @@ abstract class QueryPlan[PlanType <: TreeNode[PlanType]] extends TreeNode[PlanTy * Runs [[transform]] with `rule` on all expressions present in this query operator. * Users should not expect a specific directionality. If a specific directionality is needed, * transformExpressionsDown or transformExpressionsUp should be used. + * * @param rule the rule to be applied to every expression in this operator. */ def transformExpressions(rule: PartialFunction[Expression, Expression]): this.type = { @@ -67,6 +118,7 @@ abstract class QueryPlan[PlanType <: TreeNode[PlanType]] extends TreeNode[PlanTy /** * Runs [[transformDown]] with `rule` on all expressions present in this query operator. + * * @param rule the rule to be applied to every expression in this operator. */ def transformExpressionsDown(rule: PartialFunction[Expression, Expression]): this.type = { @@ -99,6 +151,7 @@ abstract class QueryPlan[PlanType <: TreeNode[PlanType]] extends TreeNode[PlanTy /** * Runs [[transformUp]] with `rule` on all expressions present in this query operator. + * * @param rule the rule to be applied to every expression in this operator. * @return */ diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala index 6d859551f8..d8944a4241 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala @@ -305,6 +305,8 @@ abstract class UnaryNode extends LogicalPlan { def child: LogicalPlan override def children: Seq[LogicalPlan] = child :: Nil + + override protected def validConstraints: Set[Expression] = child.constraints } /** diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/basicOperators.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/basicOperators.scala index 16f4b355b1..8150ff8434 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/basicOperators.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/basicOperators.scala @@ -87,11 +87,27 @@ case class Generate( } } -case class Filter(condition: Expression, child: LogicalPlan) extends UnaryNode { +case class Filter(condition: Expression, child: LogicalPlan) + extends UnaryNode with PredicateHelper { override def output: Seq[Attribute] = child.output + + override protected def validConstraints: Set[Expression] = { + child.constraints.union(splitConjunctivePredicates(condition).toSet) + } } -abstract class SetOperation(left: LogicalPlan, right: LogicalPlan) extends BinaryNode +abstract class SetOperation(left: LogicalPlan, right: LogicalPlan) extends BinaryNode { + + protected def leftConstraints: Set[Expression] = left.constraints + + protected def rightConstraints: Set[Expression] = { + require(left.output.size == right.output.size) + val attributeRewrites = AttributeMap(right.output.zip(left.output)) + right.constraints.map(_ transform { + case a: Attribute => attributeRewrites(a) + }) + } +} private[sql] object SetOperation { def unapply(p: SetOperation): Option[(LogicalPlan, LogicalPlan)] = Some((p.left, p.right)) @@ -106,6 +122,10 @@ case class Intersect(left: LogicalPlan, right: LogicalPlan) extends SetOperation leftAttr.withNullability(leftAttr.nullable && rightAttr.nullable) } + override protected def validConstraints: Set[Expression] = { + leftConstraints.union(rightConstraints) + } + // Intersect are only resolved if they don't introduce ambiguous expression ids, // since the Optimizer will convert Intersect to Join. override lazy val resolved: Boolean = @@ -119,6 +139,8 @@ case class Except(left: LogicalPlan, right: LogicalPlan) extends SetOperation(le /** We don't use right.output because those rows get excluded from the set. */ override def output: Seq[Attribute] = left.output + override protected def validConstraints: Set[Expression] = leftConstraints + override lazy val resolved: Boolean = childrenResolved && left.output.length == right.output.length && @@ -157,13 +179,36 @@ case class Union(children: Seq[LogicalPlan]) extends LogicalPlan { val sizeInBytes = children.map(_.statistics.sizeInBytes).sum Statistics(sizeInBytes = sizeInBytes) } + + /** + * Maps the constraints containing a given (original) sequence of attributes to those with a + * given (reference) sequence of attributes. Given the nature of union, we expect that the + * mapping between the original and reference sequences are symmetric. + */ + private def rewriteConstraints( + reference: Seq[Attribute], + original: Seq[Attribute], + constraints: Set[Expression]): Set[Expression] = { + require(reference.size == original.size) + val attributeRewrites = AttributeMap(original.zip(reference)) + constraints.map(_ transform { + case a: Attribute => attributeRewrites(a) + }) + } + + override protected def validConstraints: Set[Expression] = { + children + .map(child => rewriteConstraints(children.head.output, child.output, child.constraints)) + .reduce(_ intersect _) + } } case class Join( - left: LogicalPlan, - right: LogicalPlan, - joinType: JoinType, - condition: Option[Expression]) extends BinaryNode { + left: LogicalPlan, + right: LogicalPlan, + joinType: JoinType, + condition: Option[Expression]) + extends BinaryNode with PredicateHelper { override def output: Seq[Attribute] = { joinType match { @@ -180,6 +225,28 @@ case class Join( } } + override protected def validConstraints: Set[Expression] = { + joinType match { + case Inner if condition.isDefined => + left.constraints + .union(right.constraints) + .union(splitConjunctivePredicates(condition.get).toSet) + case LeftSemi if condition.isDefined => + left.constraints + .union(splitConjunctivePredicates(condition.get).toSet) + case Inner => + left.constraints.union(right.constraints) + case LeftSemi => + left.constraints + case LeftOuter => + left.constraints + case RightOuter => + right.constraints + case FullOuter => + Set.empty[Expression] + } + } + def duplicateResolved: Boolean = left.outputSet.intersect(right.outputSet).isEmpty // Joins are only resolved if they don't introduce ambiguous expression ids. diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/plans/ConstraintPropagationSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/plans/ConstraintPropagationSuite.scala new file mode 100644 index 0000000000..b5cf91394d --- /dev/null +++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/plans/ConstraintPropagationSuite.scala @@ -0,0 +1,173 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.catalyst.plans + +import org.apache.spark.SparkFunSuite +import org.apache.spark.sql.catalyst.analysis._ +import org.apache.spark.sql.catalyst.dsl.expressions._ +import org.apache.spark.sql.catalyst.dsl.plans._ +import org.apache.spark.sql.catalyst.expressions._ +import org.apache.spark.sql.catalyst.plans.logical._ + +class ConstraintPropagationSuite extends SparkFunSuite { + + private def resolveColumn(tr: LocalRelation, columnName: String): Expression = + tr.analyze.resolveQuoted(columnName, caseInsensitiveResolution).get + + private def verifyConstraints(found: Set[Expression], expected: Set[Expression]): Unit = { + val missing = expected.filterNot(i => found.map(_.semanticEquals(i)).reduce(_ || _)) + val extra = found.filterNot(i => expected.map(_.semanticEquals(i)).reduce(_ || _)) + if (missing.nonEmpty || extra.nonEmpty) { + fail( + s""" + |== FAIL: Constraints do not match === + |Found: ${found.mkString(",")} + |Expected: ${expected.mkString(",")} + |== Result == + |Missing: ${if (missing.isEmpty) "N/A" else missing.mkString(",")} + |Found but not expected: ${if (extra.isEmpty) "N/A" else extra.mkString(",")} + """.stripMargin) + } + } + + test("propagating constraints in filters") { + val tr = LocalRelation('a.int, 'b.string, 'c.int) + + assert(tr.analyze.constraints.isEmpty) + + assert(tr.where('a.attr > 10).select('c.attr, 'b.attr).analyze.constraints.isEmpty) + + verifyConstraints(tr + .where('a.attr > 10) + .analyze.constraints, + Set(resolveColumn(tr, "a") > 10, + IsNotNull(resolveColumn(tr, "a")))) + + verifyConstraints(tr + .where('a.attr > 10) + .select('c.attr, 'a.attr) + .where('c.attr < 100) + .analyze.constraints, + Set(resolveColumn(tr, "a") > 10, + resolveColumn(tr, "c") < 100, + IsNotNull(resolveColumn(tr, "a")), + IsNotNull(resolveColumn(tr, "c")))) + } + + test("propagating constraints in union") { + val tr1 = LocalRelation('a.int, 'b.int, 'c.int) + val tr2 = LocalRelation('d.int, 'e.int, 'f.int) + val tr3 = LocalRelation('g.int, 'h.int, 'i.int) + + assert(tr1 + .where('a.attr > 10) + .unionAll(tr2.where('e.attr > 10) + .unionAll(tr3.where('i.attr > 10))) + .analyze.constraints.isEmpty) + + verifyConstraints(tr1 + .where('a.attr > 10) + .unionAll(tr2.where('d.attr > 10) + .unionAll(tr3.where('g.attr > 10))) + .analyze.constraints, + Set(resolveColumn(tr1, "a") > 10, + IsNotNull(resolveColumn(tr1, "a")))) + } + + test("propagating constraints in intersect") { + val tr1 = LocalRelation('a.int, 'b.int, 'c.int) + val tr2 = LocalRelation('a.int, 'b.int, 'c.int) + + verifyConstraints(tr1 + .where('a.attr > 10) + .intersect(tr2.where('b.attr < 100)) + .analyze.constraints, + Set(resolveColumn(tr1, "a") > 10, + resolveColumn(tr1, "b") < 100, + IsNotNull(resolveColumn(tr1, "a")), + IsNotNull(resolveColumn(tr1, "b")))) + } + + test("propagating constraints in except") { + val tr1 = LocalRelation('a.int, 'b.int, 'c.int) + val tr2 = LocalRelation('a.int, 'b.int, 'c.int) + verifyConstraints(tr1 + .where('a.attr > 10) + .except(tr2.where('b.attr < 100)) + .analyze.constraints, + Set(resolveColumn(tr1, "a") > 10, + IsNotNull(resolveColumn(tr1, "a")))) + } + + test("propagating constraints in inner join") { + val tr1 = LocalRelation('a.int, 'b.int, 'c.int).subquery('tr1) + val tr2 = LocalRelation('a.int, 'd.int, 'e.int).subquery('tr2) + verifyConstraints(tr1 + .where('a.attr > 10) + .join(tr2.where('d.attr < 100), Inner, Some("tr1.a".attr === "tr2.a".attr)) + .analyze.constraints, + Set(tr1.resolveQuoted("a", caseInsensitiveResolution).get > 10, + tr2.resolveQuoted("d", caseInsensitiveResolution).get < 100, + tr1.resolveQuoted("a", caseInsensitiveResolution).get === + tr2.resolveQuoted("a", caseInsensitiveResolution).get, + IsNotNull(tr2.resolveQuoted("a", caseInsensitiveResolution).get), + IsNotNull(tr1.resolveQuoted("a", caseInsensitiveResolution).get), + IsNotNull(tr2.resolveQuoted("d", caseInsensitiveResolution).get))) + } + + test("propagating constraints in left-semi join") { + val tr1 = LocalRelation('a.int, 'b.int, 'c.int).subquery('tr1) + val tr2 = LocalRelation('a.int, 'd.int, 'e.int).subquery('tr2) + verifyConstraints(tr1 + .where('a.attr > 10) + .join(tr2.where('d.attr < 100), LeftSemi, Some("tr1.a".attr === "tr2.a".attr)) + .analyze.constraints, + Set(tr1.resolveQuoted("a", caseInsensitiveResolution).get > 10, + IsNotNull(tr1.resolveQuoted("a", caseInsensitiveResolution).get))) + } + + test("propagating constraints in left-outer join") { + val tr1 = LocalRelation('a.int, 'b.int, 'c.int).subquery('tr1) + val tr2 = LocalRelation('a.int, 'd.int, 'e.int).subquery('tr2) + verifyConstraints(tr1 + .where('a.attr > 10) + .join(tr2.where('d.attr < 100), LeftOuter, Some("tr1.a".attr === "tr2.a".attr)) + .analyze.constraints, + Set(tr1.resolveQuoted("a", caseInsensitiveResolution).get > 10, + IsNotNull(tr1.resolveQuoted("a", caseInsensitiveResolution).get))) + } + + test("propagating constraints in right-outer join") { + val tr1 = LocalRelation('a.int, 'b.int, 'c.int).subquery('tr1) + val tr2 = LocalRelation('a.int, 'd.int, 'e.int).subquery('tr2) + verifyConstraints(tr1 + .where('a.attr > 10) + .join(tr2.where('d.attr < 100), RightOuter, Some("tr1.a".attr === "tr2.a".attr)) + .analyze.constraints, + Set(tr2.resolveQuoted("d", caseInsensitiveResolution).get < 100, + IsNotNull(tr2.resolveQuoted("d", caseInsensitiveResolution).get))) + } + + test("propagating constraints in full-outer join") { + val tr1 = LocalRelation('a.int, 'b.int, 'c.int).subquery('tr1) + val tr2 = LocalRelation('a.int, 'd.int, 'e.int).subquery('tr2) + assert(tr1.where('a.attr > 10) + .join(tr2.where('d.attr < 100), FullOuter, Some("tr1.a".attr === "tr2.a".attr)) + .analyze.constraints.isEmpty) + } +} |