Java ClusterMetrics.getMapSlotCapacity方法代码示例

import org.apache.hadoop.mapreduce.ClusterMetrics; //导入方法依赖的package包/类
/**
 * Choose a Machine in runtime according to the cluster status.
 */
private Machine chooseMachine(Configuration conf) throws IOException {
  final int parts = conf.getInt(N_PARTS, Integer.MAX_VALUE);
  try {
    for(;; Thread.sleep(2000)) {
      //get cluster status
      final ClusterMetrics status = cluster.getClusterStatus();
      final int m = 
        status.getMapSlotCapacity() - status.getOccupiedMapSlots();
      final int r = 
        status.getReduceSlotCapacity() - status.getOccupiedReduceSlots();
      if (m >= parts || r >= parts) {
        //favor ReduceSide machine
        final Machine value = r >= parts?
            ReduceSide.INSTANCE: MapSide.INSTANCE;
        Util.out.println("  " + this + " is " + value + " (m=" + m + ", r=" + r + ")");
        return value;
      }
    }
  } catch (InterruptedException e) {
    throw new IOException(e);
  }    
}
 

import org.apache.hadoop.mapreduce.ClusterMetrics; //导入方法依赖的package包/类
/**
 * Get status information about the Map-Reduce cluster.
 *  
 * @return the status information about the Map-Reduce cluster as an object
 *         of {@link ClusterStatus}.
 * @throws IOException
 */
public ClusterStatus getClusterStatus() throws IOException {
  try {
    ClusterMetrics metrics = cluster.getClusterStatus();
    return new ClusterStatus(metrics.getTaskTrackerCount(),
      metrics.getBlackListedTaskTrackerCount(), cluster.getTaskTrackerExpiryInterval(),
      metrics.getOccupiedMapSlots(),
      metrics.getOccupiedReduceSlots(), metrics.getMapSlotCapacity(),
      metrics.getReduceSlotCapacity(),
      cluster.getJobTrackerStatus(),
      metrics.getDecommissionedTaskTrackerCount());
  } catch (InterruptedException ie) {
    throw new IOException(ie);
  }
}
 

import org.apache.hadoop.mapreduce.ClusterMetrics; //导入方法依赖的package包/类
/**
 * We try to use some light-weight mechanism to determine cluster load.
 * @return Whether, from job client perspective, the cluster is overloaded.
 */
private boolean isOverloaded(long now) throws IOException {
  try {
    ClusterMetrics clusterMetrics = jobTracker.getClusterMetrics();
    
    // If there are more jobs than number of task trackers, we assume the
    // cluster is overloaded. This is to bound the memory usage of the
    // simulator job tracker, in situations where we have jobs with small
    // number of map tasks and large number of reduce tasks.
    if (runningJobs.size() >= clusterMetrics.getTaskTrackerCount()) {
      System.out.printf("%d Overloaded is %s: " +
              "#runningJobs >= taskTrackerCount (%d >= %d)\n",
              now, Boolean.TRUE.toString(),
              runningJobs.size(), clusterMetrics.getTaskTrackerCount());
      return true;    
    }

    float incompleteMapTasks = 0; // include pending & running map tasks.
    for (Map.Entry<JobID, JobSketchInfo> entry : runningJobs.entrySet()) {
      org.apache.hadoop.mapreduce.JobStatus jobStatus = jobTracker
          .getJobStatus(entry.getKey());
      incompleteMapTasks += (1 - Math.min(jobStatus.getMapProgress(), 1.0))
          * entry.getValue().numMaps;
    }

    boolean overloaded = incompleteMapTasks >
        OVERLAOD_MAPTASK_MAPSLOT_RATIO * clusterMetrics.getMapSlotCapacity();
    String relOp = (overloaded) ? ">" : "<=";
    System.out.printf("%d Overloaded is %s: "
        + "incompleteMapTasks %s %.1f*mapSlotCapacity (%.1f %s %.1f*%d)\n",
        now, Boolean.toString(overloaded), relOp, OVERLAOD_MAPTASK_MAPSLOT_RATIO,
        incompleteMapTasks, relOp, OVERLAOD_MAPTASK_MAPSLOT_RATIO, 
        clusterMetrics.getMapSlotCapacity());
    return overloaded;
  } catch (InterruptedException e) {
    throw new IOException("InterruptedException", e);
  }
}