Compared with original hadoop-1.2.0, this version can

1. Output detailed logs. The detailed logs contain:

   • concrete processing steps
   • Volume of input and output data after user code finishes
   • Current dataflow counters when user code starts

   For example,

    Mapper 
    22:57:44 INFO [map() starts]
    23:00:13 INFO [map() ends] <Input: records = 6128490, bytes = 362377587> <Output: records = 6128490, bytes = 503316368> 
    23:00:37 INFO [Start combine() in spill 1][partition 1] <currentInputRecords = 0, currentOutputRecords = 0>
    23:00:51 INFO [Start combine() in spill 1][partition 2] <currentInputRecords = 8582, currentOutputRecords = 68>
    
    Reducer 
    17:33:57 INFO [Shuffling] Segment (id = 9, bytes = 104523972) into RAM from mapper-9
    17:33:57 INFO [Shuffling] Segment (id = 7, bytes = 104585760) into RAM from mapper-7
    17:34:00 INFO [combine() in shuffle starts] SegmentIds(5, 4, 9, 0, 1) <currentCombineInputRecords = 0, currentCombineOutputRecords = 0> 
    17:34:05 INFO [combine() in shuffle ends] <Input: records = 10917050, bytes = 731980261> <Output: records = 513020, bytes = 34387617> 
    17:34:46 INFO [reduce() starts] <Input: Bytes(in-memory segments) = 0, Bytes(on-disk segments) = 1254859834>
   

2. Display real-time memory usage in task's JVM.

   Users only need to specify a configuration 'child.monitor.jstat.seconds = 2'

   For example, the real-time memory usage of a reducer is as follows.

   Eden space New generation Old generation GC counts

3. Generate 'heap dump@record(i)' if one or some following configurations are specified.

   Heap dump configurations.

   Configuration	Description	The modified code
   heapdump.path	where the dump will be generated (by default, /tmp)
   heapdump.map.file.bytes.read	external bytes read	Task.java
   heapdump.map.input.records	map input records	Mapper.java & MapRunner.java
   heapdump.map.combine.input.records	combine input records in mapper	ReduceContext.java & Task.java
   heapdump.reduce.combine.input.records	combine input records in reducer	ReduceContext.java & Task.java
   heapdump.reduce.input.group	reduce input groups	Reducer.java & ReduceTask.java
   heapdump.reduce.input.records	reduce input records	ReduceTask.java & ReduceContext.java
   heapdump.reduce.file.bytes.read	external bytes read	Task.java
   task.reporter.update.ms	3000	Task.java

   For example, if users specify conf.set("heapdump.map.input.records", "9425[5]")

   Mapper will dump the heap when it encounter the i-th specified record (1885th, 3769th, 5653rd, 7537th, 9425th)

    2014-04-27 21:03:38,797 INFO org.apache.hadoop.mapreduce.Mapper: [map() begins]
    2014-04-27 21:03:54,368 INFO org.apache.hadoop.mapred.Utils: [HeapDump] generate /tmp/mapInRecords-1885-out-0-pid-26746.hprof
    2014-04-27 21:04:10,118 INFO org.apache.hadoop.mapred.Utils: [HeapDump] generate /tmp/mapInRecords-3769-out-0-pid-26746.hprof
    2014-04-27 21:04:33,437 INFO org.apache.hadoop.mapred.Utils: [HeapDump] generate /tmp/mapInRecords-5653-out-0-pid-26746.hprof
    2014-04-27 21:04:57,338 INFO org.apache.hadoop.mapred.Utils: [HeapDump] generate /tmp/mapInRecords-7537-out-0-pid-26746.hprof
    2014-04-27 21:05:08,634 INFO org.apache.hadoop.mapred.Utils: [HeapDump] generate /tmp/mapInRecords-9425-out-0-pid-26746.hprof