yarn local dirs的管理实现

在yarn中对 yarn.nodemanager.local-dirs的状态更新操作，定义在 LocalDirsHandlerService（org.apache.hadoop.yarn.server.nodemanager.LocalDirsHandlerService）相关类中，在nm启动时，会启动一个LocalDirsHandlerService服务，循环检测yarn.nodemanager.local-dirs和yarn.nodemanager.log-dirs目录的可用性，本质上其实是用java.util.Timer 和java.util.TimerTask 实现的一个服务线程。

LocalDirsHandlerService的内部类 MonitoringTimerTask扩展了TimerTask类

通过MonitoringTimerTas的构造函数对进行初始化，比如获取设置的yarn.nodemanager.log-dirs和yarn.nodemanager.local-dirs 设置有效的local路径

这个线程常用的参数：

YarnConfiguration.NM_DISK_HEALTH_CHECK_INTERVAL_MS
//yarn.nodemanager.disk-health-checker.interval-ms 默认是2分钟
YarnConfiguration.NM_DISK_HEALTH_CHECK_ENABLE
//yarn.nodemanager.disk-health-checker.enable 默认是开启
YarnConfiguration.NM_MIN_HEALTHY_DISKS_FRACTION
//yarn.nodemanager.disk-health-checker.min-healthy-disks 默认是0.25，即最少应该是1/4的设置路径是正常的
在cdh4.6.0中，MonitoringTimerTask的构造函数如下：

public MonitoringTimerTask( Configuration conf) throws YarnException {
  localDirs = new DirectoryCollection(
      validatePaths(conf.getTrimmedStrings(YarnConfiguration.NM_LOCAL_DIRS)));
  logDirs = new DirectoryCollection(
      validatePaths(conf.getTrimmedStrings(YarnConfiguration.NM_LOG_DIRS)));
  localDirsAllocator = new LocalDirAllocator(
      YarnConfiguration.NM_LOCAL_DIRS);
  logDirsAllocator = new LocalDirAllocator( YarnConfiguration.NM_LOG_DIRS);
}

而在cdh5.2.0中，构造函数多了两个配置项

YarnConfiguration.NM_MAX_PER_DISK_UTILIZATION_PERCENTAGE
//percentage of disk that can be used before the dir is taken out of the good dirs list
//yarn.nodemanager.disk-health-checker.max-disk-utilization-per-disk-percentage 默认是100（这个值需要改成小于100，比如80，否则容易出现磁盘满地问题）
YarnConfiguration.NM_MIN_PER_DISK_FREE_SPACE_MB
//minimum space, in MB, that must be available on the disk for the dir to be marked as good
//yarn.nodemanager.disk-health-checker.min-free-space-per-disk-mb 默认是0MB
在检查localdirs的初始可用性会考虑到这两个设置（validatePaths方法）

public MonitoringTimerTask( Configuration conf) throws YarnRuntimeException {
  float maxUsableSpacePercentagePerDisk =
      conf.getFloat(
        YarnConfiguration.NM_MAX_PER_DISK_UTILIZATION_PERCENTAGE,
        YarnConfiguration.DEFAULT_NM_MAX_PER_DISK_UTILIZATION_PERCENTAGE);
  long minFreeSpacePerDiskMB =
      conf.getLong( YarnConfiguration.NM_MIN_PER_DISK_FREE_SPACE_MB,
        YarnConfiguration.DEFAULT_NM_MIN_PER_DISK_FREE_SPACE_MB);
  localDirs =
      new DirectoryCollection(
        validatePaths(conf
          .getTrimmedStrings( YarnConfiguration.NM_LOCAL_DIRS)),
        maxUsableSpacePercentagePerDisk, minFreeSpacePerDiskMB);
  logDirs =
      new DirectoryCollection(
        validatePaths(conf.getTrimmedStrings( YarnConfiguration.NM_LOG_DIRS)),
        maxUsableSpacePercentagePerDisk, minFreeSpacePerDiskMB);
  localDirsAllocator = new LocalDirAllocator(
      YarnConfiguration.NM_LOCAL_DIRS);
  logDirsAllocator = new LocalDirAllocator( YarnConfiguration.NM_LOG_DIRS);
}

local dirs的判断线程会每隔一段时间对目录的可用性进行测试，调用的方法是

checkDirs—->updateDirsAfterFailure—>areDisksHealthy
可用判断主要是判断错误的目录占配置目录的比例，当yarn.nodemanager.local-dirs或者yarn.nodemanager.log-dirs异常目录占了一定百分比后，磁盘检测就会失败，nm就会抛出异常：

具体的判断逻辑在areDisksHealthy方法中:

public boolean areDisksHealthy() {
if (! isDiskHealthCheckerEnabled) { //判断是否开启了磁盘状态检测的功能
return true;
}
int goodDirs = getLocalDirs().size();
int failedDirs = localDirs.getFailedDirs().size();
int totalConfiguredDirs = goodDirs + failedDirs;
if (goodDirs/( float)totalConfiguredDirs < minNeededHealthyDisksFactor ) { //异常的yarn.nodemanager.local-dirs比例判断
return false; // Not enough healthy local- dirs
}
goodDirs = getLogDirs().size();
failedDirs = logDirs.getFailedDirs().size();
totalConfiguredDirs = goodDirs + failedDirs;
if (goodDirs/( float)totalConfiguredDirs < minNeededHealthyDisksFactor ) { //异常的yarn.nodemanager.log-dirs比例判断
return false; // Not enough healthy log- dirs
}
return true;
}