Github user pwendell commented on a diff in the pull request: https://github.com/apache/spark/pull/4450#discussion_r29118405 --- Diff: core/src/main/scala/org/apache/spark/util/collection/PartitionedSerializedPairBuffer.scala --- @@ -0,0 +1,254 @@ +/* + * 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.util.collection + +import java.io.InputStream +import java.nio.IntBuffer +import java.util.Comparator + +import org.apache.spark.SparkEnv +import org.apache.spark.serializer.{JavaSerializerInstance, SerializerInstance} +import org.apache.spark.storage.BlockObjectWriter +import org.apache.spark.util.collection.PartitionedSerializedPairBuffer._ + +/** + * Append-only buffer of key-value pairs, each with a corresponding partition ID, that serializes + * its records upon insert and stores them as raw bytes. + * + * We use two data-structures to store the contents. The serialized records are stored in a + * ChainedBuffer that can expand gracefully as records are added. This buffer is accompanied by a + * metadata buffer that stores pointers into the data buffer as well as the partition ID of each + * record. Each entry in the metadata buffer takes up a fixed amount of space. + * + * Sorting the collection means swapping entries in the metadata buffer - the record buffer need not + * be modified at all. Storing the partition IDs in the metadata buffer means that comparisons can + * happen without following any pointers, which should minimize cache misses. + * + * Currently, only sorting by partition is supported. + * + * @param metaInitialRecords The initial number of entries in the metadata buffer. + * @param kvBlockSize The size of each byte buffer in the ChainedBuffer used to store the records. + * @param serializerInstance the serializer used for serializing inserted records. + */ +private[spark] class PartitionedSerializedPairBuffer[K, V]( + metaInitialRecords: Int, + kvBlockSize: Int, + serializerInstance: SerializerInstance = SparkEnv.get.serializer.newInstance) + extends WritablePartitionedPairCollection[K, V] { + + if (serializerInstance.isInstanceOf[JavaSerializerInstance]) { + throw new IllegalArgumentException("PartitionedSerializedPairBuffer does not support" + + " Java-serialized objects.") + } + + private var metaBuffer = IntBuffer.allocate(metaInitialRecords * NMETA) + + private val kvBuffer: ChainedBuffer = new ChainedBuffer(kvBlockSize) + private val kvOutputStream = new ChainedBufferOutputStream(kvBuffer) + private val kvSerializationStream = serializerInstance.serializeStream(kvOutputStream) + + def insert(partition: Int, key: K, value: V): Unit = { + if (metaBuffer.position == metaBuffer.capacity) { + growMetaBuffer() + } + + val keyStart = kvBuffer.size + if (keyStart < 0) { + throw new Exception(s"Can't grow buffer beyond ${1 << 31} bytes") + } + kvSerializationStream.writeObject[Any](key) + kvSerializationStream.flush() + val valueStart = kvBuffer.size + kvSerializationStream.writeObject[Any](value) + kvSerializationStream.flush() + val valueEnd = kvBuffer.size + + metaBuffer.put(keyStart) + metaBuffer.put(valueStart) + metaBuffer.put(valueEnd) + metaBuffer.put(partition) + } + + /** Double the size of the array because we've reached capacity */ + private def growMetaBuffer(): Unit = { + if (metaBuffer.capacity * 4 >= (1 << 30)) { --- End diff -- Any reason not to just check for `metaBuffer.capacity.toLong > Integer.MAX_VALUE` - seems easier to reason about.
--- 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 infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org