Spark围绕着BlockManager构建了存储模块,包括RDD,Shuffle,Broadcast的存储都使用了BlockManager。而BlockManager在实现上是一个针对每个应用的Master/Executor结构,即Driver上BlockManager充当了Master角色,而各个Slave上(具体到应用范围,就是Executor)的BlockManager充当了Slave角色。
因此,BlockManager在Driver和应用的各个Executor之间各有一份,Driver上的BlockManager不具备实际存储的能力,它记录了各个Executor的BlockManager的状态(通过查看BlockManagerMaster和BlockManagerMasterActor的源代码,BlockManagerMaster和BlockManagerMasterActor并没有持有一个BlockManager对象,那么每个Executor BlockManager的状态存储在什么地方?通过查看BlockManager的类注释,发现BlockManager确实运行在Driver上)。Master BlockManager和ExecutorBlockManager之间的通信也是基于Akka,消息格式定义于BlockManagerMessages类中。
上面的描述并不准确,事实上在Driver端,同Executor一样,各有一个BlockManager。除此之外,Driver上还有一个BlockManager Master,它的实现类是BlockManagerManager,因此,对于BlockManager而言,Driver既是Master也是Slave
0.BlockManager类注释:
/**
* Manager running on every node (driver and executors) which provides interfaces for putting and
* retrieving blocks both locally and remotely into various stores (memory, disk, and off-heap).
*
* Note that #initialize() must be called before the BlockManager is usable.
*/
private[spark] class BlockManager(
1. Master/Slave的通信内容
Master BlockManager向Executor BlockManager可以发送的消息包括:
sealed trait ToBlockManagerSlave
// Remove a block from the slaves that have it. This can only be used to remove
// blocks that the master knows about.
case class RemoveBlock(blockId: BlockId) extends ToBlockManagerSlave
// Remove all blocks belonging to a specific RDD.
case class RemoveRdd(rddId: Int) extends ToBlockManagerSlave
// Remove all blocks belonging to a specific shuffle.
case class RemoveShuffle(shuffleId: Int) extends ToBlockManagerSlave
// Remove all blocks belonging to a specific broadcast.
case class RemoveBroadcast(broadcastId: Long, removeFromDriver: Boolean = true)
extends ToBlockManagerSlave
Executor BlockManager向Master BlockManager可以发送的消息包括:
sealed trait ToBlockManagerMaster
case class RegisterBlockManager(
blockManagerId: BlockManagerId,
maxMemSize: Long,
sender: ActorRef)
extends ToBlockManagerMaster
//获取某个Block在哪些Executor的BlockManager上
case class GetLocations(blockId: BlockId) extends ToBlockManagerMaster
//获取一组Block在哪些Executor的BlockManager上
case class GetLocationsMultipleBlockIds(blockIds: Array[BlockId]) extends ToBlockManagerMaster
case class GetPeers(blockManagerId: BlockManagerId) extends ToBlockManagerMaster
case class GetActorSystemHostPortForExecutor(executorId: String) extends ToBlockManagerMaster
//删除Executor
case class RemoveExecutor(execId: String) extends ToBlockManagerMaster
case object StopBlockManagerMaster extends ToBlockManagerMaster
case object GetMemoryStatus extends ToBlockManagerMaster
case object GetStorageStatus extends ToBlockManagerMaster
case class GetBlockStatus(blockId: BlockId, askSlaves: Boolean = true)
extends ToBlockManagerMaster
case class GetMatchingBlockIds(filter: BlockId => Boolean, askSlaves: Boolean = true)
extends ToBlockManagerMaster
case class BlockManagerHeartbeat(blockManagerId: BlockManagerId) extends ToBlockManagerMaster
case object ExpireDeadHosts extends ToBlockManagerMaster
//更新Block信息
case class UpdateBlockInfo(
var blockManagerId: BlockManagerId,
var blockId: BlockId,
var storageLevel: StorageLevel,
var memSize: Long,
var diskSize: Long,
var tachyonSize: Long)
extends ToBlockManagerMaster
with Externalizable {
def this() = this(null, null, null, 0, 0, 0) // For deserialization only
override def writeExternal(out: ObjectOutput): Unit = Utils.tryOrIOException {
blockManagerId.writeExternal(out)
out.writeUTF(blockId.name)
storageLevel.writeExternal(out)
out.writeLong(memSize)
out.writeLong(diskSize)
out.writeLong(tachyonSize)
}
override def readExternal(in: ObjectInput): Unit = Utils.tryOrIOException {
blockManagerId = BlockManagerId(in)
blockId = BlockId(in.readUTF())
storageLevel = StorageLevel(in)
memSize = in.readLong()
diskSize = in.readLong()
tachyonSize = in.readLong()
}
}
2. BlockManagerMasterActor说明
/**
* BlockManagerMasterActor is an actor on the master node to track statuses of
* all slaves' block managers.
*/
包含的数据结构:
// Mapping from block manager id to the block manager's information.
///BlockManagerId与BlockManagerInfo之间的对应,每个Executor对应一个BlockManagerId
private val blockManagerInfo = new mutable.HashMap[BlockManagerId, BlockManagerInfo]
// Mapping from executor ID to block manager ID.
//Executor ID与BlockManagerID之间的对应关系
private val blockManagerIdByExecutor = new mutable.HashMap[String, BlockManagerId]
// Mapping from block id to the set of block managers that have the block.
//BlockId与包含这个Block的Location(由BlockManagerId表示)
private val blockLocations = new JHashMap[BlockId, mutable.HashSet[BlockManagerId]]
BlockManagerId是一个由host,port和executorId表示的数据结构,从这里也可以看出来BlockManager是Executor范围的数据结构
3. BlockManagerSlaveActor说明
/**
* An actor to take commands from the master to execute options. For example,
* this is used to remove blocks from the slave's BlockManager.
*/
BlockManagerSlaveActor包含的数据结构都体现在构造函数中了,如下所示,BlockManagerSlaveActor包含了本Executor对应的BlockManager以及该Executor的MapOutputTracker用于记录Map Shuffle输出
private[storage]
class BlockManagerSlaveActor(
blockManager: BlockManager,
mapOutputTracker: MapOutputTracker)
extends Actor with ActorLogReceive with Logging
弄清楚BlockManager的通信机制,发现要比分析BlockManager的读写数据(依赖于DiskStore和MemoryStore实现,而DiskStore又依赖于DiskBlockManager实现)复杂一些,主要是头脑中没有清晰的picuture:关于BlockManager,Driver有什么,Executor上有什么,它们之间如何通信,这个继续分析吧。
本文深入解析了Spark中BlockManager的设计原理及其在Driver和Executor间的通信机制。介绍了BlockManager的架构,包括Master/Slave结构,以及BlockManagerMasterActor和BlockManagerSlaveActor的作用。详细解释了它们之间的消息交互内容。

被折叠的 条评论
为什么被折叠?



