forked from apache/spark
/
ExtractPythonUDFs.scala
168 lines (150 loc) · 6.2 KB
/
ExtractPythonUDFs.scala
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
/*
* 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.execution.python
import scala.collection.mutable
import scala.collection.mutable.ArrayBuffer
import org.apache.spark.sql.catalyst.expressions._
import org.apache.spark.sql.catalyst.expressions.aggregate.AggregateExpression
import org.apache.spark.sql.catalyst.plans.logical.{Aggregate, LogicalPlan, Project}
import org.apache.spark.sql.catalyst.rules.Rule
import org.apache.spark.sql.execution
import org.apache.spark.sql.execution.SparkPlan
/**
* Extracts all the Python UDFs in logical aggregate, which depends on aggregate expression or
* grouping key, evaluate them after aggregate.
*/
private[spark] object ExtractPythonUDFFromAggregate extends Rule[LogicalPlan] {
/**
* Returns whether the expression could only be evaluated within aggregate.
*/
private def belongAggregate(e: Expression, agg: Aggregate): Boolean = {
e.isInstanceOf[AggregateExpression] ||
agg.groupingExpressions.exists(_.semanticEquals(e))
}
private def hasPythonUdfOverAggregate(expr: Expression, agg: Aggregate): Boolean = {
expr.find {
e => e.isInstanceOf[PythonUDF] && e.find(belongAggregate(_, agg)).isDefined
}.isDefined
}
private def extract(agg: Aggregate): LogicalPlan = {
val projList = new ArrayBuffer[NamedExpression]()
val aggExpr = new ArrayBuffer[NamedExpression]()
agg.aggregateExpressions.foreach { expr =>
if (hasPythonUdfOverAggregate(expr, agg)) {
// Python UDF can only be evaluated after aggregate
val newE = expr transformDown {
case e: Expression if belongAggregate(e, agg) =>
val alias = e match {
case a: NamedExpression => a
case o => Alias(e, "agg")()
}
aggExpr += alias
alias.toAttribute
}
projList += newE.asInstanceOf[NamedExpression]
} else {
aggExpr += expr
projList += expr.toAttribute
}
}
// There is no Python UDF over aggregate expression
Project(projList, agg.copy(aggregateExpressions = aggExpr))
}
def apply(plan: LogicalPlan): LogicalPlan = plan transformUp {
case agg: Aggregate if agg.aggregateExpressions.exists(hasPythonUdfOverAggregate(_, agg)) =>
extract(agg)
}
}
/**
* Extracts PythonUDFs from operators, rewriting the query plan so that the UDF can be evaluated
* alone in a batch.
*
* Only extracts the PythonUDFs that could be evaluated in Python (the single child is PythonUDFs
* or all the children could be evaluated in JVM).
*
* This has the limitation that the input to the Python UDF is not allowed include attributes from
* multiple child operators.
*/
private[spark] object ExtractPythonUDFs extends Rule[SparkPlan] {
private def hasPythonUDF(e: Expression): Boolean = {
e.find(_.isInstanceOf[PythonUDF]).isDefined
}
private def canEvaluateInPython(e: PythonUDF): Boolean = {
e.children match {
// single PythonUDF child could be chained and evaluated in Python
case Seq(u: PythonUDF) => canEvaluateInPython(u)
// Python UDF can't be evaluated directly in JVM
case children => !children.exists(hasPythonUDF)
}
}
private def collectEvaluatableUDF(expr: Expression): Seq[PythonUDF] = expr match {
case udf: PythonUDF if canEvaluateInPython(udf) => Seq(udf)
case e => e.children.flatMap(collectEvaluatableUDF)
}
def apply(plan: SparkPlan): SparkPlan = plan transformUp {
case plan: SparkPlan => extract(plan)
}
/**
* Extract all the PythonUDFs from the current operator and evaluate them before the operator.
*/
private def extract(plan: SparkPlan): SparkPlan = {
val udfs = plan.expressions.flatMap(collectEvaluatableUDF)
// ignore the PythonUDF that come from second/third aggregate, which is not used
.filter(udf => udf.references.subsetOf(plan.inputSet))
if (udfs.isEmpty) {
// If there aren't any, we are done.
plan
} else {
val attributeMap = mutable.HashMap[PythonUDF, Expression]()
// Rewrite the child that has the input required for the UDF
val newChildren = plan.children.map { child =>
// Pick the UDF we are going to evaluate
val validUdfs = udfs.filter { case udf =>
// Check to make sure that the UDF can be evaluated with only the input of this child.
udf.references.subsetOf(child.outputSet)
}
if (validUdfs.nonEmpty) {
val resultAttrs = udfs.zipWithIndex.map { case (u, i) =>
AttributeReference(s"pythonUDF$i", u.dataType)()
}
val evaluation = BatchEvalPythonExec(validUdfs, child.output ++ resultAttrs, child)
attributeMap ++= validUdfs.zip(resultAttrs)
evaluation
} else {
child
}
}
// Other cases are disallowed as they are ambiguous or would require a cartesian
// product.
udfs.filterNot(attributeMap.contains).foreach { udf =>
sys.error(s"Invalid PythonUDF $udf, requires attributes from more than one child.")
}
val rewritten = plan.transformExpressions {
case p: PythonUDF if attributeMap.contains(p) =>
attributeMap(p)
}.withNewChildren(newChildren)
// extract remaining python UDFs recursively
val newPlan = extract(rewritten)
if (newPlan.output != plan.output) {
// Trim away the new UDF value if it was only used for filtering or something.
execution.ProjectExec(plan.output, newPlan)
} else {
newPlan
}
}
}
}