RussellSpitzer commented on a change in pull request #3983: URL: https://github.com/apache/iceberg/pull/3983#discussion_r833787697
########## File path: spark/v3.2/spark/src/main/java/org/apache/iceberg/spark/actions/Spark3ZOrderUDF.java ########## @@ -0,0 +1,257 @@ +/* + * 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.iceberg.spark.actions; + +import java.io.IOException; +import java.io.ObjectInputStream; +import java.io.Serializable; +import java.nio.ByteBuffer; +import java.nio.charset.CharsetEncoder; +import java.nio.charset.StandardCharsets; +import org.apache.iceberg.util.ZOrderByteUtils; +import org.apache.spark.sql.Column; +import org.apache.spark.sql.expressions.UserDefinedFunction; +import org.apache.spark.sql.functions; +import org.apache.spark.sql.types.BinaryType; +import org.apache.spark.sql.types.BooleanType; +import org.apache.spark.sql.types.ByteType; +import org.apache.spark.sql.types.DataType; +import org.apache.spark.sql.types.DataTypes; +import org.apache.spark.sql.types.DateType; +import org.apache.spark.sql.types.DoubleType; +import org.apache.spark.sql.types.FloatType; +import org.apache.spark.sql.types.IntegerType; +import org.apache.spark.sql.types.LongType; +import org.apache.spark.sql.types.ShortType; +import org.apache.spark.sql.types.StringType; +import org.apache.spark.sql.types.TimestampType; +import scala.collection.Seq; + +class Spark3ZOrderUDF implements Serializable { + private static final byte[] PRIMITIVE_EMPTY = new byte[ZOrderByteUtils.PRIMITIVE_BUFFER_SIZE]; + + /** + * Every Spark task runs iteratively on a rows in a single thread so ThreadLocal should protect from + * concurrent access to any of these structures. + */ + private transient ThreadLocal<ByteBuffer> outputBuffer; + private transient ThreadLocal<byte[][]> inputHolder; + private transient ThreadLocal<ByteBuffer>[] inputBuffers; + private transient ThreadLocal<CharsetEncoder> encoder; + + private final int numCols; + + private int inputCol = 0; + private int totalOutputBytes = 0; + private final int varTypeSize; + private final int maxOutputSize; + + Spark3ZOrderUDF(int numCols, int varTypeSize, int maxOutputSize) { + this.numCols = numCols; + this.varTypeSize = varTypeSize; + this.maxOutputSize = maxOutputSize; + } + + private void readObject(ObjectInputStream in) throws IOException, ClassNotFoundException { + in.defaultReadObject(); + inputBuffers = new ThreadLocal[numCols]; + inputHolder = ThreadLocal.withInitial(() -> new byte[numCols][]); + outputBuffer = ThreadLocal.withInitial(() -> ByteBuffer.allocate(totalOutputBytes)); + encoder = ThreadLocal.withInitial(() -> StandardCharsets.UTF_8.newEncoder()); + } + + private ByteBuffer inputBuffer(int position, int size) { + if (inputBuffers[position] == null) { + // May over allocate on concurrent calls + inputBuffers[position] = ThreadLocal.withInitial(() -> ByteBuffer.allocate(size)); Review comment: We discussed this a bit offline, for those interested our lifecycle basically looks like ``` Driver --- UDF --> Executor -> Core - Task (Thread) \_____> Core - Task (Thread) ``` So every task is run in a single thread on the executor, but the tasks are referring to a UDF which is basically a singleton on the Executor JVM. Each task in isolation runs as if it was single threaded but any UDF members need to be protected against MultiThreaded access. An alternative implementation we could do is to explicitly handle the multithreading with RDD operations. For example something like ``` rdd.mapPartitions( it -> { inputBuffer = allocate outputBuffer = allocate it.map( row -> ZOrderRow(row, inputBuffer, outputBuffer) } ) ``` -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
