[HADOOP-424] mapreduce jobs fail when no split is returned via inputFormat.getSplits

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.5.0
• Fix Version/s: 0.6.0
• Assignee: Unassigned
• Reporter: Frédéric Bertin
• Created: 04/Aug/06 15:05
• Updated: 08/Jul/09 16:51
• Resolved: 06/Sep/06 18:46

Description

I'm using a MapReduce job to process some data logged and timestamped into files.
When the job runs, it does not process the whole data, but filters only the data that has been logged since the last job run.

However, when no new data has been logged, the job fails because the getSplits method of InputFormat returns no split. Thus the number of map tasks is 0. This is not intercepted, and the job fails at reduce step because it seems it does not find any data to process:

java.io.FileNotFoundException: /local/home/hadoop/var/mapred/local/task_0030_r_000000_3/all.2 at org.apache.hadoop.fs.LocalFileSystem.openRaw(LocalFileSystem.java:121) at org.apache.hadoop.fs.FSDataInputStream$Checker.(FSDataInputStream.java:47) at org.apache.hadoop.fs.FSDataInputStream.(FSDataInputStream.java:221) at org.apache.hadoop.fs.FileSystem.open(FileSystem.java:150) at org.apache.hadoop.io.SequenceFile$Reader.(SequenceFile.java:259) at org.apache.hadoop.io.SequenceFile$Reader.(SequenceFile.java:253) at org.apache.hadoop.mapred.ReduceTask.run(ReduceTask.java:241) at org.apache.hadoop.mapred.TaskTracker$Child.main(TaskTracker.java:1013)

What should be Hadoop's behaviour in such a case?

IMHO, the job should be considered as successful. Indeed, this is not a job failure, but just a lack of input data. WDYT?