nutch搏斗之一

nutch搏斗之一

问题描述： 
在用nutch1.0做generate 包括5亿url的crawldb时，它默认按照64M分块，分成777个map task，在运行的后期出现 
Could not find taskTracker/jobcache/job_200903231519_0017/attempt_200903231519_0017_r_000051_0/output/file.out in any of the configured local directories 
异常。 
解决办法： 
减小task数目，改成按照crawldb里面文件个数划分的策略： 

public static class InputFormat extends SequenceFileInputFormat<WritableComparable, Writable> {
   /** Don't split inputs, to keep things polite. */
   public InputSplit[] getSplits(JobConf job, int nSplits)
     throws IOException {
     FileStatus[] files = listStatus(job);
     FileSystem fs = FileSystem.get(job);
     InputSplit[] splits = new InputSplit[files.length];
     for (int i = 0; i < files.length; i++) {
       FileStatus cur = files[i];
       splits[i] = new FileSplit(cur.getPath(), 0,
           cur.getLen(), (String[])null);
     }
     return splits;
   }
 }

这次出现了新问题，有数个task因为十分钟无反应而导致整个任务failed 
解决办法： 
修改hadoop-site.xml 

<property>
  <name>mapred.task.timeout</name>
  <value>3600000</value>
  <description>The number of milliseconds before a task will be
  terminated if it neither reads an input, writes an output, nor
  updates its status string.
  </description>
</property>

总结： 
大与小，多与少，长与短，在不同的情况下是不断变化的，对于大数据量而言，更要跟具实际情况灵活变化，所谓运用之刀，存乎一心是也！