android的java代码是黄色

所以初步怀疑是因为AM心跳汇报出现问题或则RM因为繁忙hang住，AM因为某些机制导致等待10min不汇报心跳，所以我们还是先了解，AM是如何向RM汇报心跳的。

在MRAppMaster中，ContainerAllocatorRouter负责向RM申请资源（发送心跳）

RMContainerAllocator其最终父类是RMCommunicator，它实现了RMHeartbeatHandler接口

1.  public interface RMHeartbeatHandler {http://www.kmrlyy.com/fujianyan/33454.html
2.  long getLastHeartbeatTime(); // 获取上一次心跳的时间
3.  void runOnNextHeartbeat(Runnable callback); // 回调注册到callback队列的callback函数
4.  }http://www.nvzi91.cn/niaodaoyan/29938.html
5.

复制代码

每一次心跳回来，都会执行一次注册在heartbeatCallbacks中的回调函数：

1.  allocatorThread = new Thread(new Runnable() {

2.  @Override
3.  public void run() {

4.  while (!stopped.get() && !Thread.currentThread().isInterrupted()) {

5.  ......http://www.kmrlyy.com/fujianyan/33455.html
6.  heartbeat();
7.  lastHeartbeatTime = context.getClock().getTime();// 记录上一次心跳时间
8.  executeHeartbeatCallbacks(); // 执行回调函数
9.  ....http://www.nvzi91.cn/yindaoyan/29939.html
10.  });http://www.nvzi91.cn/zigongjiliu/29942.html

复制代码

RMCommunicator类中：

1.  private void executeHeartbeatCallbacks() {

2.  Runnable callback = null;
3.  while ((callback = heartbeatCallbacks.poll()) != null) {

4.  callback.run();
5.  }
6.  }http://www.nvzi91.cn/yindaoyan/29940.html
7.  http://www.kmrlyy.com/penqiangyan/33457.html

复制代码

首先会向RM注册，把自己的host和port告诉RM，然后在启动一条线程（startAllocatorThread）定期的调用RMContainerAllocator中实现的heartbeat方法（向RM申请资源，定期汇报信息，告诉RM自己还活着）。

AM初始化同时也会初始化RMCommunicator：

1.  protected void serviceStart() throws Exception {

2.  scheduler= createSchedulerProxy(); // 获取RM的代理
3.  register(); // 注册
4.  startAllocatorThread(); // 心跳线程
5.  ....
6.  }http://www.nvzi91.cn/luanchaonanzhong/29941.html
7.  http://www.kmrlyy.com/niaodaoyan/33458.html

复制代码

AM的ContainerAllocatorRouter事件处理流程如下图：

注册流程：

调用RMCommunicator远程调用ApplicationMasterService的registerApplicationMaster方法，设置维护responseId，然后把它加入AMLivelinessMonitor中，并使用map记录时间，用来监控AM是否因为长时间没有心跳而超时，如果AM长时间没有心跳信息更新，RM就会通知NodeManager把AM移除。

心跳线程：

在发送心跳的过程中，即也是获取资源的过程

1.  @Overridehttp://m.nvzi91.cn/penqiangyan/29351.html
2.  protected synchronized void heartbeat() throws Exception {

3.  scheduleStats.updateAndLogIfChanged("Before Scheduling: ");
4.  List<Container> allocatedContainers = getResources();// 重要的方法
5.  if (allocatedContainers.size() > 0) {

6.  scheduledRequests.assign(allocatedContainers);
7.  }
8.  ......
9.  }
10.

复制代码

获取资源的过程：

1.  private List<Container> getResources() throws Exception {

2.  ...
3.  response = makeRemoteRequest(); // 和RM进行交互
4.  ...
5.  // 优先处理RM发送过来的命令
6.  if (response.getAMCommand() != null) {

7.  switch(response.getAMCommand()) {

8.  case AM_RESYNC:
9.  case AM_SHUTDOWN:
10.  eventHandler.handle(new JobEvent(this.getJob().getID(),
11.  JobEventType.JOB_AM_REBOOT));
12.  throw new YarnRuntimeException("Resource Manager doesn't recognize AttemptId: " +
13.  this.getContext().getApplicationID());
14.  default:
15.  ....
16.  }http://m.nvzi91.cn/zigongai/29352.html
17.  // 等等一系列处理
18.  }
19.  }

复制代码

构建请求：

1.  protected AllocateResponse makeRemoteRequest() throws IOException {

2.  AllocateRequest allocateRequest =
3.  AllocateRequest.newInstance(lastResponseID,
4.  super.getApplicationProgress(), new ArrayList<ResourceRequest>(ask),
5.  new ArrayList<ContainerId>(release), blacklistRequest);
6.  AllocateResponse allocateResponse;
7.  allocateResponse = scheduler.allocate(allocateRequest); // RPC调用ApplicationMasterService的allocate方法
8.  .....
9.  }http://m.nvzi91.cn/jiankang/29353.html

复制代码

每一次心跳的调用都会刷新AMLivelinessMonitor的时间，代表AM还活着

而且我们通过代码可以看出，资源请求被封装为一个ask，即一个ResourceRequest的ArrayList的资源列表 例如：

1.  priority:20 host:host9 capability:<memory:2048, vCores:1>
2.  priority:20 host:host2 capability:<memory:2048, vCores:1>
3.  priority:20 host:host10 capability:<memory:2048, vCores:1>
4.  priority:20 host:/rack/rack3203 capability:<memory:2048, vCores:1>
5.  priority:20 host:/rack/rack3202 capability:<memory:2048, vCores:1>
6.  priority:20 host:* capability:<memory:2048, vCores:1>
7.  www.kmrlyy.com

复制代码

然而，ask是如何被构造的呢？

RMContainerAllocator中的addMap,addReduce,assign方法中对ask的数据内容进行了修改

1.  addContainerReq --> addResourceRequest --> addResourceRequestToAsk;
2.  www.nvzi91.cn

复制代码

通过在代码自己添加日志可以看出，资源会被分为local，rack，和any级别去申请资源

最终变为一个ask list发送到RM上：

1.  ask Capability:<memory:2048, vCores:1> ResourceName:* NumContainers:384 Priority:20 RelaxLocality:true
2.  ask Capability:<memory:2048, vCores:1> ResourceName:/rack/rack3201 NumContainers:227 Priority:20 RelaxLocality:true
3.  ask Capability:<memory:2048, vCores:1> ResourceName:/rack/rack3202 NumContainers:231 Priority:20 RelaxLocality:true
4.  ask Capability:<memory:2048, vCores:1> ResourceName:/rack/rack3203 NumContainers:152 Priority:20 RelaxLocality:true
5.  ask Capability:<memory:2048, vCores:1> ResourceName:/rack/rack3204 NumContainers:158 Priority:20 RelaxLocality:true
6.  ask Capability:<memory:2048, vCores:1> ResourceName:host1 NumContainers:46 Priority:20 RelaxLocality:true
7.  ask Capability:<memory:2048, vCores:1> ResourceName:host5 NumContainers:52 Priority:20 RelaxLocality:true
8.  ask Capability:<memory:2048, vCores:1> ResourceName:host6 NumContainers:38 Priority:20 RelaxLocality:true
9.  m.nvzi91.cn

复制代码

类似日志为：

1.  getResources() for application_1438330253091_0004: ask=29 release= 0 newContainers=0 finishedContainers=0 resourcelimit=<memory:0, vCores:0> knownNMs=24