The task itself moves it when it receives a commitTask message. See the OutputCommitter class: http://hadoop.apache.org/docs/current/api/org/apache/hadoop/mapred/OutputCommitter.html#commitTask(org.apache.hadoop.mapred.TaskAttemptContext)
On Fri, May 10, 2013 at 8:49 AM, Ling Kun <lkun.e...@gmail.com> wrote: > Dear all, > > I am looking into the MR work flow, and want to know more details about > the reduce output data copy . > > Here is my question. > > For the DFSIO test or some other MR jobs. Each reduce task will run on a > TT, and generate files to some dirs named like this: " > XXX//_temporary/_attempt_201305101045_0005_r_000000_0/", there will also be > a result file named part-00000. > > After the reducer done the task. the reducer output data part-00000 should > be moved from the local disk to the HDFS. > > My question is: Is that the time that when reducer finish the task that > part-00000 will be copied to the HDFS? Who make this file copy happen? The > Reducer child? The TaskTracker which run the reduce task? Or the JobTracker? > > Thanks, > > yours, > Kun Ling > > -- > http://www.lingcc.com -- Harsh J
