Github user cloud-fan commented on a diff in the pull request:
https://github.com/apache/spark/pull/16228#discussion_r91829681
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/estimation/JoinEstimation.scala
---
@@ -0,0 +1,175 @@
+/*
+ * 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.logical.estimation
+
+import scala.collection.mutable
+
+import org.apache.spark.sql.catalyst.expressions.{AttributeReference,
Expression}
+import org.apache.spark.sql.catalyst.planning.ExtractEquiJoinKeys
+import org.apache.spark.sql.catalyst.plans.logical.{ColumnStat, Join,
Statistics}
+import org.apache.spark.sql.types.DataType
+
+
+object JoinEstimation {
+ import EstimationUtils._
+
+ // scalastyle:off
+ /**
+ * Estimate output size and number of rows after a join operator, and
propogate updated column
+ * statsitics.
+ * The number of rows of A inner join B on A.k1 = B.k1 is estimated by
this basic formula:
+ * T(A IJ B) = T(A) * T(B) / max(V(A.k1), V(B.k1)), where V is the
number of distinct values of
+ * that column. The underlying assumption for this formula is: each
value of the smaller domain
+ * is included in the larger domain.
+ * Generally, inner join with multiple join keys can also be estimated
based on the above
+ * formula:
+ * T(A IJ B) = T(A) * T(B) / (max(V(A.k1), V(B.k1)) * max(V(A.k2),
V(B.k2)) * ... * max(V(A.kn), V(B.kn)))
+ * However, the denominator can become very large and excessively reduce
the result, so we use a
+ * conservative strategy to take only the largest max(V(A.ki), V(B.ki))
as the denominator.
+ *
+ * @return Return the updated statistics after join. Return `None` if
the join type is not
+ * supported, or we don't have enough statistics for estimation.
+ */
+ // scalastyle:on
+ def estimate(join: Join): Option[Statistics] = join match {
+ case ExtractEquiJoinKeys(joinType, leftKeys, rightKeys, condition,
left, right)
+ if supportsJoinType(joinType) && hasRowCountStat(left, right) =>
+
+ // 1. Compute the denominator
+ var ndvDenom: BigInt = -1
+ val keyPairs = extractJoinKeys(leftKeys, rightKeys)
+ val leftStats = left.statistics
+ val rightStats = right.statistics
+ val intersectedStats = new mutable.HashMap[String, ColumnStat]()
+ var i = 0
+ while(i < keyPairs.length && ndvDenom != 0) {
+ val (leftKey, rightKey) = keyPairs(i)
+ // Do estimation if we have enough statistics
+ if (hasColumnStat((left, leftKey), (right, rightKey))) {
+ val leftKeyStats = leftStats.colStats(leftKey.name)
+ val rightKeyStats = rightStats.colStats(rightKey.name)
+
+ // Check if the two sides are disjoint
+ val lRange = Range(leftKeyStats.min, leftKeyStats.max,
leftKey.dataType)
+ val rRange = Range(rightKeyStats.min, rightKeyStats.max,
rightKey.dataType)
+ if (Range.isIntersected(lRange, rRange)) {
+ // Get the largest ndv among pairs of join keys
+ val maxNdv =
leftKeyStats.distinctCount.max(rightKeyStats.distinctCount)
+ if (maxNdv > ndvDenom) ndvDenom = maxNdv
+
+ // Update intersected column stats
+ val minNdv =
leftKeyStats.distinctCount.min(rightKeyStats.distinctCount)
+ val (newMin1, newMax1, newMin2, newMax2) =
+ Range.intersect(lRange, rRange, leftKey.dataType,
rightKey.dataType)
+ intersectedStats.put(leftKey.name,
intersectedColumnStat(leftKeyStats, minNdv,
+ newMin1, newMax1))
+ intersectedStats.put(rightKey.name,
intersectedColumnStat(rightKeyStats, minNdv,
+ newMin2, newMax2))
+ } else {
+ // Set ndvDenom to zero to indicate that this join should have
no output
+ ndvDenom = 0
+ }
+ }
+ i += 1
+ }
+
+ // 2. Estimate the number of output rows
+ val leftRows = leftStats.rowCount.get
+ val rightRows = rightStats.rowCount.get
+ val outputRows: BigInt = if (ndvDenom == 0) {
+ // One of the join key pairs is disjoint, thus the two sides of
join is disjoint.
+ 0
+ } else if (ndvDenom < 0) {
+ // There isn't join keys or column stats for any of the join key
pairs, we estimate like
+ // cartesian product.
+ leftRows * rightRows
+ } else {
+ ceil(BigDecimal(leftRows * rightRows) / BigDecimal(ndvDenom))
+ }
+
+ // 3. Update statistics based on the output of join
+ val originalStats = leftStats.colStats ++ rightStats.colStats
+ val outputWithStat = join.output.filter(attr =>
originalStats.contains(attr.name))
+
+ val outputColStats = new mutable.HashMap[String, ColumnStat]()
+ if (ndvDenom == 0) {
+ // empty output
+ outputWithStat.foreach(a => outputColStats.put(a.name,
emptyColumnStat(a.dataType)))
+ } else if (ndvDenom < 0) {
+ // cartesian product, column stats will not change
+ outputWithStat.foreach(a => outputColStats.put(a.name,
originalStats(a.name)))
+ } else {
+ val leftRatio = BigDecimal(outputRows) / BigDecimal(leftRows)
+ val rightRatio = BigDecimal(outputRows) / BigDecimal(rightRows)
+ outputWithStat.foreach { a =>
+ // check if this attribute is a join key
+ if (intersectedStats.contains(a.name)) {
+ outputColStats.put(a.name, intersectedStats(a.name))
+ } else {
+ val oldColStat = originalStats(a.name)
+ val oldNdv = oldColStat.distinctCount
+ // We only change (scale down) the number of distinct values
if the number of rows
+ // decreases after join, because join won't produce new values
even if the number of
+ // rows increases.
+ val newNdv = if (left.outputSet.contains(a) && leftRatio < 1) {
+ ceil(BigDecimal(oldNdv) * leftRatio)
+ } else if (right.outputSet.contains(a) && rightRatio < 1) {
+ ceil(BigDecimal(oldNdv) * rightRatio)
+ } else {
+ oldNdv
+ }
+ outputColStats.put(a.name, oldColStat.copy(distinctCount =
newNdv, nullCount = 0))
+ }
+ }
+ }
+
+ Some(Statistics(
+ sizeInBytes = outputRows * getRowSize(join.output,
outputColStats.toMap),
+ rowCount = Some(outputRows),
+ colStats = outputColStats.toMap,
+ isBroadcastable = false))
+
+ case _ => None
+ }
+
+ def emptyColumnStat(dataType: DataType): ColumnStat = {
+ ColumnStat(distinctCount = 0, min = None, max = None, nullCount = 0,
+ avgLen = dataType.defaultSize, maxLen = dataType.defaultSize)
+ }
+
+ def intersectedColumnStat(
+ origin: ColumnStat,
+ newDistinctCount: BigInt,
+ newMin: Option[Any],
+ newMax: Option[Any]): ColumnStat = {
+ origin.copy(distinctCount = newDistinctCount, min = newMin, max =
newMax, nullCount = 0)
+ }
+
+ def extractJoinKeys(
+ leftKeys: Seq[Expression],
+ rightKeys: Seq[Expression]): Seq[(AttributeReference,
AttributeReference)] = {
+ leftKeys.zip(rightKeys).flatMap {
+ case (ExtractAttr(left), ExtractAttr(right)) => Some((left, right))
+ // Currently we don't deal with equal joins like key1 = key2 + 5.
--- End diff --
I don't think we should add a new expression just for the current
implementation limitations. BTW, only handle `Cast` may also lose a lot of
estimation opportunities, we should support expression before we release this
feature.
---
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]
