spark通信原理
https://github.com/apache/spark/tree/master/core/src/main/scala/org/apache/spark/network
https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/network/BlockTransferService.scala
https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/network/BlockDataManager.scala
package org.apache.spark.network import scala.reflect.ClassTag import org.apache.spark.network.buffer.ManagedBuffer
import org.apache.spark.storage.{BlockId, StorageLevel} private[spark]
trait BlockDataManager { /**
* Interface to get local block data. Throws an exception if the block cannot be found or
* cannot be read successfully.
*/
def getBlockData(blockId: BlockId): ManagedBuffer /**
* Put the block locally, using the given storage level.
*
* Returns true if the block was stored and false if the put operation failed or the block
* already existed.
*/
def putBlockData(
blockId: BlockId,
data: ManagedBuffer,
level: StorageLevel,
classTag: ClassTag[_]): Boolean /**
* Release locks acquired by [[putBlockData()]] and [[getBlockData()]].
*/
def releaseLock(blockId: BlockId, taskAttemptId: Option[Long]): Unit
}
package org.apache.spark.network import java.io.Closeable
import java.nio.ByteBuffer import scala.concurrent.{Future, Promise}
import scala.concurrent.duration.Duration
import scala.reflect.ClassTag import org.apache.spark.internal.Logging
import org.apache.spark.network.buffer.{FileSegmentManagedBuffer, ManagedBuffer, NioManagedBuffer}
import org.apache.spark.network.shuffle.{BlockFetchingListener, ShuffleClient, TempFileManager}
import org.apache.spark.storage.{BlockId, StorageLevel}
import org.apache.spark.util.ThreadUtils private[spark]
abstract class BlockTransferService extends ShuffleClient with Closeable with Logging { /**
* Initialize the transfer service by giving it the BlockDataManager that can be used to fetch
* local blocks or put local blocks.
*/
def init(blockDataManager: BlockDataManager): Unit /**
* Tear down the transfer service.
*/
def close(): Unit /**
* Port number the service is listening on, available only after [[init]] is invoked.
*/
def port: Int /**
* Host name the service is listening on, available only after [[init]] is invoked.
*/
def hostName: String /**
* Fetch a sequence of blocks from a remote node asynchronously,
* available only after [[init]] is invoked.
*
* Note that this API takes a sequence so the implementation can batch requests, and does not
* return a future so the underlying implementation can invoke onBlockFetchSuccess as soon as
* the data of a block is fetched, rather than waiting for all blocks to be fetched.
*/
override def fetchBlocks(
host: String,
port: Int,
execId: String,
blockIds: Array[String],
listener: BlockFetchingListener,
tempFileManager: TempFileManager): Unit /**
* Upload a single block to a remote node, available only after [[init]] is invoked.
*/
def uploadBlock(
hostname: String,
port: Int,
execId: String,
blockId: BlockId,
blockData: ManagedBuffer,
level: StorageLevel,
classTag: ClassTag[_]): Future[Unit] /**
* A special case of [[fetchBlocks]], as it fetches only one block and is blocking.
*
* It is also only available after [[init]] is invoked.
*/
def fetchBlockSync(
host: String,
port: Int,
execId: String,
blockId: String,
tempFileManager: TempFileManager): ManagedBuffer = {
// A monitor for the thread to wait on.
val result = Promise[ManagedBuffer]()
fetchBlocks(host, port, execId, Array(blockId),
new BlockFetchingListener {
override def onBlockFetchFailure(blockId: String, exception: Throwable): Unit = {
result.failure(exception)
}
override def onBlockFetchSuccess(blockId: String, data: ManagedBuffer): Unit = {
data match {
case f: FileSegmentManagedBuffer =>
result.success(f)
case _ =>
val ret = ByteBuffer.allocate(data.size.toInt)
ret.put(data.nioByteBuffer())
ret.flip()
result.success(new NioManagedBuffer(ret))
}
}
}, tempFileManager)
ThreadUtils.awaitResult(result.future, Duration.Inf)
} /**
* Upload a single block to a remote node, available only after [[init]] is invoked.
*
* This method is similar to [[uploadBlock]], except this one blocks the thread
* until the upload finishes.
*/
def uploadBlockSync(
hostname: String,
port: Int,
execId: String,
blockId: BlockId,
blockData: ManagedBuffer,
level: StorageLevel,
classTag: ClassTag[_]): Unit = {
val future = uploadBlock(hostname, port, execId, blockId, blockData, level, classTag)
ThreadUtils.awaitResult(future, Duration.Inf)
}
}
spark通信原理的更多相关文章
- Spark Shuffle 堆外内存溢出问题与解决(Shuffle通信原理)
Spark Shuffle 堆外内存溢出问题与解决(Shuffle通信原理) http://xiguada.org/spark-shuffle-direct-buffer-oom/ 问题描述 Spar ...
- [Spark内核] 第32课:Spark Worker原理和源码剖析解密:Worker工作流程图、Worker启动Driver源码解密、Worker启动Executor源码解密等
本課主題 Spark Worker 原理 Worker 启动 Driver 源码鉴赏 Worker 启动 Executor 源码鉴赏 Worker 与 Master 的交互关系 [引言部份:你希望读者 ...
- Spark核心技术原理透视一(Spark运行原理)
在大数据领域,只有深挖数据科学领域,走在学术前沿,才能在底层算法和模型方面走在前面,从而占据领先地位. Spark的这种学术基因,使得它从一开始就在大数据领域建立了一定优势.无论是性能,还是方案的统一 ...
- spark核心原理
spark运行结构图如下: spark基本概念 应用程序(application):用户编写的spark应用程序,包含驱动程序(Driver)和分布在集群中多个节点上运行的Executor代码,在执行 ...
- Spark集群基础概念 与 spark架构原理
一.Spark集群基础概念 将DAG划分为多个stage阶段,遵循以下原则: 1.将尽可能多的窄依赖关系的RDD划为同一个stage阶段. 2.当遇到shuffle操作,就意味着上一个stage阶段结 ...
- Spark Worker原理和源码剖析解密:Worker工作流程图、Worker启动Driver源码解密、Worker启动Executor源码解密等
本课主题 Spark Worker 原理 Worker 启动 Driver 源码鉴赏 Worker 启动 Executor 源码鉴赏 Worker 与 Master 的交互关系 Spark Worke ...
- 基于web的IM软件通信原理分析
关于IM(InstantMessaging)即时通信类软件(如微信,QQ),大多数都是桌面应用程序或者native应用较为流行,而网上关于原生IM或桌面IM软件类的通信原理介绍也较多,此处不再赘述.而 ...
- Socket 通信原理(Android客户端和服务器以TCP&&UDP方式互通)
转载地址:http://blog.csdn.net/mad1989/article/details/9147661 ZERO.前言 有关通信原理内容是在网上或百科整理得到,代码部分为本人所写,如果不当 ...
- SSL 通信原理及Tomcat SSL 配置
SSL 通信原理及Tomcat SSL 双向配置 目录1 参考资料 .................................................................. ...
随机推荐
- [BZOJ3339] Rmq Problem(线段树)
传送门 这个题的方法好像很多啊 1.莫队暴力 2.线段树 + 离线处理 先预处理出sg[i]表示前i个数的sg值,next[i]表示i的下一位置在哪里,如果后面再没有i,那么next[i] = n + ...
- bzoj1225 [HNOI2001] 求正整数
1225: [HNOI2001] 求正整数 Time Limit: 10 Sec Memory Limit: 162 MBSubmit: 762 Solved: 313[Submit][Statu ...
- 【BZOJ3224】普通平衡树(splay)
题意: 您需要写一种数据结构(可参考题目标题),来维护一些数,其中需要提供以下操作:1. 插入x数2. 删除x数(若有多个相同的数,因只删除一个)3. 查询x数的排名(若有多个相同的数,因输出最小的排 ...
- NSArray,NSMutableArray的一些常用方法
不可变数组 ——NSArray 常用的初始化一个数组: NSArray *array1 = [[NSArray alloc] init]; NSArray *array2 = ...
- 在 VirtualBox 5.0 系列中让虚拟机支持 USB 3.0 必须开启 APIC
VirtualBox 5.0 系列正式支持 USB 3.0,能够在宿主机支持 USB 3.0 的情况下,让虚拟机也选择具备 USB 3.0 的功能.但是经过多方试验,发现必须在 VirtualBox ...
- SGU 106 在区间范围内的线性方程解个数
题意:求解方程ax+by+c=0,在区间x1->x2和y1->y2的解的个数. 看似简单,真心a的不容易啊! 开始跪于第8组数据,原因是没用long long !后来改了,跪于12组,超时 ...
- 使用nginx时,让web取得原始请求地址
问题描述 当使用nginx配置proxy_pass参数以后,后端web取得的Request.Uri是proxy_pass中配置的地址,而不是client访问的原始地址 举例说明: 假设nginx配置文 ...
- codeforces 1041 e 构造
Codeforces 1041 E 构造题. 给出一种操作,对于一棵树,去掉它的一条边.那么这颗树被分成两个部分,两个部分的分别的最大值就是这次操作的答案. 现在给出一棵树所有操作的结果,问能不能构造 ...
- luogu P2912 [USACO08OCT]牧场散步Pasture Walking
题目描述 The N cows (2 <= N <= 1,000) conveniently numbered 1..N are grazing among the N pastures ...
- MACBOOK 忘记密码
如果没有系统安装盘,可以采用下面的几种方法:首先,启动机器,启动时按住Apple和S键,以单用户模式(single user mode)进入系统.输入:mount -uv / 然后回车.接下来可以采用 ...