[GitHub] spark pull request #19872: [SPARK-22274][PYTHON][SQL] User-defined aggregati...

[ueshin <gi...@git.apache.org>]

Github user ueshin commented on a diff in the pull request:

    https://github.com/apache/spark/pull/19872#discussion_r160617597
  
    --- Diff: sql/core/src/main/scala/org/apache/spark/sql/execution/python/AggregateInPandasExec.scala ---
    @@ -0,0 +1,152 @@
    +/*
    + * 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 java.io.File
    +
    +import scala.collection.mutable.ArrayBuffer
    +
    +import org.apache.spark.{SparkEnv, TaskContext}
    +import org.apache.spark.api.python.{ChainedPythonFunctions, PythonEvalType}
    +import org.apache.spark.rdd.RDD
    +import org.apache.spark.sql.catalyst.InternalRow
    +import org.apache.spark.sql.catalyst.expressions._
    +import org.apache.spark.sql.catalyst.plans.physical.{AllTuples, ClusteredDistribution, Distribution, Partitioning}
    +import org.apache.spark.sql.execution.{GroupedIterator, SparkPlan, UnaryExecNode}
    +import org.apache.spark.sql.types.{DataType, StructField, StructType}
    +import org.apache.spark.util.Utils
    +
    +/**
    + * Physical node for aggregation with group aggregate Pandas UDF.
    + *
    + * This plan works by sending the necessary (projected) input grouped data as Arrow record batches
    + * to the python worker, the python worker invokes the UDF and sends the results to the executor,
    + * finally the executor evaluates any post-aggregation expressions and join the result with the
    + * grouped key.
    + */
    +case class AggregateInPandasExec(
    +    groupingExpressions: Seq[NamedExpression],
    +    udfExpressions: Seq[PythonUDF],
    +    resultExpressions: Seq[NamedExpression],
    +    child: SparkPlan)
    +  extends UnaryExecNode {
    +
    +  override val output: Seq[Attribute] = resultExpressions.map(_.toAttribute)
    +
    +  override def outputPartitioning: Partitioning = child.outputPartitioning
    +
    +  override def producedAttributes: AttributeSet = AttributeSet(output)
    +
    +  override def requiredChildDistribution: Seq[Distribution] = {
    +    if (groupingExpressions.isEmpty) {
    +      AllTuples :: Nil
    +    } else {
    +      ClusteredDistribution(groupingExpressions) :: Nil
    +    }
    +  }
    +
    +  private def collectFunctions(udf: PythonUDF): (ChainedPythonFunctions, Seq[Expression]) = {
    +    udf.children match {
    +      case Seq(u: PythonUDF) =>
    +        val (chained, children) = collectFunctions(u)
    +        (ChainedPythonFunctions(chained.funcs ++ Seq(udf.func)), children)
    +      case children =>
    +        // There should not be any other UDFs, or the children can't be evaluated directly.
    +        assert(children.forall(_.find(_.isInstanceOf[PythonUDF]).isEmpty))
    +        (ChainedPythonFunctions(Seq(udf.func)), udf.children)
    +    }
    +  }
    +
    +  override def requiredChildOrdering: Seq[Seq[SortOrder]] =
    +    Seq(groupingExpressions.map(SortOrder(_, Ascending)))
    +
    +  override protected def doExecute(): RDD[InternalRow] = {
    +    val inputRDD = child.execute()
    +
    +    val bufferSize = inputRDD.conf.getInt("spark.buffer.size", 65536)
    +    val reuseWorker = inputRDD.conf.getBoolean("spark.python.worker.reuse", defaultValue = true)
    +    val sessionLocalTimeZone = conf.sessionLocalTimeZone
    +    val pandasRespectSessionTimeZone = conf.pandasRespectSessionTimeZone
    +
    +    val (pyFuncs, inputs) = udfExpressions.map(collectFunctions).unzip
    +
    +    val allInputs = new ArrayBuffer[Expression]
    +    val dataTypes = new ArrayBuffer[DataType]
    +    val argOffsets = inputs.map { input =>
    +      input.map { e =>
    +        if (allInputs.exists(_.semanticEquals(e))) {
    +          allInputs.indexWhere(_.semanticEquals(e))
    +        } else {
    +          allInputs += e
    +          dataTypes += e.dataType
    +          allInputs.length - 1
    +        }
    +      }.toArray
    +    }.toArray
    +
    +    val schema = StructType(dataTypes.zipWithIndex.map { case (dt, i) =>
    +      StructField(s"_$i", dt)
    +    })
    +
    +    val input = groupingExpressions.map(_.toAttribute) ++ udfExpressions.map(_.resultAttribute)
    --- End diff --
    
    nit: maybe this name `input` is confusing.


---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org