spark core 2.0 SortShuffleManager
2016-12-28 17:54
253 查看
在基于排序的shuffle,输入的记录都按目标分区ID进行排序,然后写入一个map 输出文件。为了读取reduder 对应的 map输出,Reducers 要获取这个文件一个连续区域。如果map的输出太大,以至于内存装不下,输出的内容排序后输出到磁盘,然后磁盘上的文件合并成一个最后的输出文件。
为了生成map输出文件,基于sort的shuffle 有两个不同的写入 方式。
第一,序列化排序,当以下三种条件都满足时使用:
1. Shuffle 依赖没有聚集或者不需要对输出排序。
2. Shuffle序列化支持对序列化值的重定位。(当前KryoSerializer和Spark SQL的定制的序列化支持)。
3. Shuffle生成的输出分区小于16777216个。
第二: 非序列化排序,用来处理其它所有的情况。
------------------------------
序列化排序模型
-----------------------------
在序列化排序模型,当输入的记录被传到shuffle 写入器时会被立即序列化,并且在排序过程中以序列化的格式在缓冲器中。 这种写入方式实现了以下几种优化:
1. 排序操作在序列化之后的二进制数据上,而不是java对象,这会减少memory占用和GC开销。这种优化需要记录序列化器有一些特性,用来把序列化的数据重新排序,而不用返序列化之后再排序。参照SPARK-4550, 这是此优化第一次提出并且实现。
2. 它使用一个特殊的节约缓存的排序器[ShuffleExternalSorter],这个排序器对压缩记录的指针和分区id进行排序,通过对每个记录仅仅使用8字节的空间并且在合并时不需要对记录进行反序列化。
3. 当溢出压缩格式支持对压缩数据进行合并时,溢出合并简单的把序列化和压缩的分区直接连在一起来输出一个最终的输出分区。 这种方法可以有效利用数据拷贝方式,像NIO的'transferTo',可以避免分配反序列化的空间或者拷贝缓冲区。
序列化排序模型的选择策略,首先是选择根据 dependency选择一个ShuffleHandler,然后根据Shufflehandler选择一个Writer ,代码如下所示:
/**
* Register a shuffle with the manager and obtain a handle for it to pass to tasks.
*/
override def registerShuffle[K, V, C](
shuffleId: Int,
numMaps: Int,
dependency: ShuffleDependency[K, V, C]): ShuffleHandle = {
if (SortShuffleWriter.shouldBypassMergeSort(SparkEnv.get.conf, dependency)) {
// If there are fewer than spark.shuffle.sort.bypassMergeThreshold partitions and we don't
// need map-side aggregation, then write numPartitions files directly and just concatenate
// them at the end. This avoids doing serialization and deserialization twice to merge
// together the spilled files, which would happen with the normal code path. The downside is
// having multiple files open at a time and thus more memory allocated to buffers.
new BypassMergeSortShuffleHandle[K, V](
shuffleId, numMaps, dependency.asInstanceOf[ShuffleDependency[K, V, V]])
} else if (SortShuffleManager.canUseSerializedShuffle(dependency)) {
// Otherwise, try to buffer map outputs in a serialized form, since this is more efficient:
new SerializedShuffleHandle[K, V](
shuffleId, numMaps, dependency.asInstanceOf[ShuffleDependency[K, V, V]])
} else {
// Otherwise, buffer map outputs in a deserialized form:
new BaseShuffleHandle(shuffleId, numMaps, dependency)
}
}
为了生成map输出文件,基于sort的shuffle 有两个不同的写入 方式。
第一,序列化排序,当以下三种条件都满足时使用:
1. Shuffle 依赖没有聚集或者不需要对输出排序。
2. Shuffle序列化支持对序列化值的重定位。(当前KryoSerializer和Spark SQL的定制的序列化支持)。
3. Shuffle生成的输出分区小于16777216个。
第二: 非序列化排序,用来处理其它所有的情况。
------------------------------
序列化排序模型
-----------------------------
在序列化排序模型,当输入的记录被传到shuffle 写入器时会被立即序列化,并且在排序过程中以序列化的格式在缓冲器中。 这种写入方式实现了以下几种优化:
1. 排序操作在序列化之后的二进制数据上,而不是java对象,这会减少memory占用和GC开销。这种优化需要记录序列化器有一些特性,用来把序列化的数据重新排序,而不用返序列化之后再排序。参照SPARK-4550, 这是此优化第一次提出并且实现。
2. 它使用一个特殊的节约缓存的排序器[ShuffleExternalSorter],这个排序器对压缩记录的指针和分区id进行排序,通过对每个记录仅仅使用8字节的空间并且在合并时不需要对记录进行反序列化。
3. 当溢出压缩格式支持对压缩数据进行合并时,溢出合并简单的把序列化和压缩的分区直接连在一起来输出一个最终的输出分区。 这种方法可以有效利用数据拷贝方式,像NIO的'transferTo',可以避免分配反序列化的空间或者拷贝缓冲区。
序列化排序模型的选择策略,首先是选择根据 dependency选择一个ShuffleHandler,然后根据Shufflehandler选择一个Writer ,代码如下所示:
/**
* Register a shuffle with the manager and obtain a handle for it to pass to tasks.
*/
override def registerShuffle[K, V, C](
shuffleId: Int,
numMaps: Int,
dependency: ShuffleDependency[K, V, C]): ShuffleHandle = {
if (SortShuffleWriter.shouldBypassMergeSort(SparkEnv.get.conf, dependency)) {
// If there are fewer than spark.shuffle.sort.bypassMergeThreshold partitions and we don't
// need map-side aggregation, then write numPartitions files directly and just concatenate
// them at the end. This avoids doing serialization and deserialization twice to merge
// together the spilled files, which would happen with the normal code path. The downside is
// having multiple files open at a time and thus more memory allocated to buffers.
new BypassMergeSortShuffleHandle[K, V](
shuffleId, numMaps, dependency.asInstanceOf[ShuffleDependency[K, V, V]])
} else if (SortShuffleManager.canUseSerializedShuffle(dependency)) {
// Otherwise, try to buffer map outputs in a serialized form, since this is more efficient:
new SerializedShuffleHandle[K, V](
shuffleId, numMaps, dependency.asInstanceOf[ShuffleDependency[K, V, V]])
} else {
// Otherwise, buffer map outputs in a deserialized form:
new BaseShuffleHandle(shuffleId, numMaps, dependency)
}
}
/** Get a writer for a given partition. Called on executors by map tasks. */ override def getWriter[K, V]( handle: ShuffleHandle, mapId: Int, context: TaskContext): ShuffleWriter[K, V] = { numMapsForShuffle.putIfAbsent( handle.shuffleId, handle.asInstanceOf[BaseShuffleHandle[_, _, _]].numMaps) val env = SparkEnv.get handle match { case unsafeShuffleHandle: SerializedShuffleHandle[K @unchecked, V @unchecked] => new UnsafeShuffleWriter( env.blockManager, shuffleBlockResolver.asInstanceOf[IndexShuffleBlockResolver], context.taskMemoryManager(), unsafeShuffleHandle, mapId, context, env.conf) case bypassMergeSortHandle: BypassMergeSortShuffleHandle[K @unchecked, V @unchecked] => new BypassMergeSortShuffleWriter( env.blockManager, shuffleBlockResolver.asInstanceOf[IndexShuffleBlockResolver], bypassMergeSortHandle, mapId, context, env.conf) case other: BaseShuffleHandle[K @unchecked, V @unchecked, _] => new SortShuffleWriter(shuffleBlockResolver, other, mapId, context) } }
* In sort-based shuffle, incoming records are sorted according to their target partition ids, then * written to a single map output file. Reducers fetch contiguous regions of this file in order to * read their portion of the map output. In cases where the map output data is too large to fit in * memory, sorted subsets of the output can are spilled to disk and those on-disk files are merged * to produce the final output file. * * Sort-based shuffle has two different write paths for producing its map output files: * * - Serialized sorting: used when all three of the following conditions hold: * 1. The shuffle dependency specifies no aggregation or output ordering. * 2. The shuffle serializer supports relocation of serialized values (this is currently * supported by KryoSerializer and Spark SQL's custom serializers). * 3. The shuffle produces fewer than 16777216 output partitions. * - Deserialized sorting: used to handle all other cases. * * ----------------------- * Serialized sorting mode * ----------------------- * * In the serialized sorting mode, incoming records are serialized as soon as they are passed to the * shuffle writer and are buffered in a serialized form during sorting. This write path implements * several optimizations: * * - Its sort operates on serialized binary data rather than Java objects, which reduces memory * consumption and GC overheads. This optimization requires the record serializer to have certain * properties to allow serialized records to be re-ordered without requiring deserialization. * See SPARK-4550, where this optimization was first proposed and implemented, for more details. * * - It uses a specialized cache-efficient sorter ([[ShuffleExternalSorter]]) that sorts * arrays of compressed record pointers and partition ids. By using only 8 bytes of space per * record in the sorting array, this fits more of the array into cache. * * - The spill merging procedure operates on blocks of serialized records that belong to the same * partition and does not need to deserialize records during the merge. * * - When the spill compression codec supports concatenation of compressed data, the spill merge * simply concatenates the serialized and compressed spill partitions to produce the final output * partition. This allows efficient data copying methods, like NIO's `transferTo`, to be used * and avoids the need to allocate decompression or copying buffers during the merge. * * For more details on these optimizations, see SPARK-7081. */ private[spark] class SortShuffleManager(conf: SparkConf) extends ShuffleManager with Logging { if (!conf.getBoolean("spark.shuffle.spill", true)) { logWarning( "spark.shuffle.spill was set to false, but this configuration is ignored as of Spark 1.6+." + " Shuffle will continue to spill to disk when necessary.") } /** * A mapping from shuffle ids to the number of mappers producing output for those shuffles. */ private[this] val numMapsForShuffle = new ConcurrentHashMap[Int, Int]() override val shuffleBlockResolver = new IndexShuffleBlockResolver(conf) /** * Get a reader for a range of reduce partitions (startPartition to endPartition-1, inclusive). * Called on executors by reduce tasks. */ override def getReader[K, C]( handle: ShuffleHandle, startPartition: Int, endPartition: Int, context: TaskContext): ShuffleReader[K, C] = { new BlockStoreShuffleReader( handle.asInstanceOf[BaseShuffleHandle[K, _, C]], startPartition, endPartition, context) } /** Remove a shuffle's metadata from the ShuffleManager. */ override def unregisterShuffle(shuffleId: Int): Boolean = { Option(numMapsForShuffle.remove(shuffleId)).foreach { numMaps => (0 until numMaps).foreach { mapId => shuffleBlockResolver.removeDataByMap(shuffleId, mapId) } } true } /** Shut down this ShuffleManager. */ override def stop(): Unit = { shuffleBlockResolver.stop() } } private[spark] object SortShuffleManager extends Logging { /** * The maximum number of shuffle output partitions that SortShuffleManager supports when * buffering map outputs in a serialized form. This is an extreme defensive programming measure, * since it's extremely unlikely that a single shuffle produces over 16 million output partitions. * */ val MAX_SHUFFLE_OUTPUT_PARTITIONS_FOR_SERIALIZED_MODE = PackedRecordPointer.MAXIMUM_PARTITION_ID + 1 /** * Helper method for determining whether a shuffle should use an optimized serialized shuffle * path or whether it should fall back to the original path that operates on deserialized objects. */ def canUseSerializedShuffle(dependency: ShuffleDependency[_, _, _]): Boolean = { val shufId = dependency.shuffleId val numPartitions = dependency.partitioner.numPartitions if (!dependency.serializer.supportsRelocationOfSerializedObjects) { log.debug(s"Can't use serialized shuffle for shuffle $shufId because the serializer, " + s"${dependency.serializer.getClass.getName}, does not support object relocation") false } else if (dependency.aggregator.isDefined) { log.debug( s"Can't use serialized shuffle for shuffle $shufId because an aggregator is defined") false } else if (numPartitions > MAX_SHUFFLE_OUTPUT_PARTITIONS_FOR_SERIALIZED_MODE) { log.debug(s"Can't use serialized shuffle for shuffle $shufId because it has more than " + s"$MAX_SHUFFLE_OUTPUT_PARTITIONS_FOR_SERIALIZED_MODE partitions") false } else { log.debug(s"Can use serialized shuffle for shuffle $shufId") true } } } /** * Subclass of [[BaseShuffleHandle]], used to identify when we've chosen to use the * serialized shuffle. */ private[spark] class SerializedShuffleHandle[K, V]( shuffleId: Int, numMaps: Int, dependency: ShuffleDependency[K, V, V]) extends BaseShuffleHandle(shuffleId, numMaps, dependency) { } /** * Subclass of [[BaseShuffleHandle]], used to identify when we've chosen to use the * bypass merge sort shuffle path. */ private[spark] class BypassMergeSortShuffleHandle[K, V]( shuffleId: Int, numMaps: Int, dependency: ShuffleDependency[K, V, V]) extends BaseShuffleHandle(shuffleId, numMaps, dependency) { }
相关文章推荐
- spark core 2.0 BypassMergeSortShuffleWriter
- spark core 2.0 MemoryLocation
- spark core 2.0 TaskSchedulerImpl 源代码解析
- spark core 2.0 MemoryBlock Soruce Code Analysis
- shuffle性能调优之HashShuffleManager和SortShuffleManager
- spark学习-37-Spark的SortShuffleManager
- spark core 2.0 MetricsConfig
- spark core 2.0 UnifiedMemoryManager
- spark core 2.0 MemoryManager
- spark core 2.0 LongArray
- ASP.NET 2.0: Playing a bit with GridView "Sort Grouping"
- spark core 2.0 CoarseGrainedSchedulerBackend SchedulerBackend ExecutorAllocationClient 源代码解析
- spark core 2.0 Executor
- spark core 2.0 StorageMemoryPool
- spark core 2.0 PartitionCoalescer, PartitionGroup, DefaultPartitionCoalescer
- spark core 2.0 Partition and HadoopPartition
- spark core 2.0 ChunkedByteBufferOutputStream
- spark core 2.0 DiskBlockObjectWriter
- spark core 2.0 RedirectableOutputStream
- spark core 2.0 BlockManager dropFromMemory