[MAPREDUCE-2378] Reduce fails when running on 1 small file.

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Cannot Reproduce
• Affects Version/s: 0.21.0
• Labels: 1 failed file log4j reduce single small tiny

Environment

java version "1.6.0_07"
Diablo Java(TM) SE Runtime Environment (build 1.6.0_07-b02)
Diablo Java HotSpot(TM) 64-Bit Server VM (build 10.0-b23, mixed mode)

• Assignee: Unassigned
• Reporter: Simon Dircks
• Created: 11/Mar/11 23:18
• Updated: 17/Jul/14 15:22
• Resolved: 17/Jul/14 15:22

Description

If i run the wordcount example on 1 small (less than 2MB) file i get the following error:

log4j:ERROR Failed to flush writer,
java.io.InterruptedIOException
at java.io.FileOutputStream.writeBytes(Native Method)
at java.io.FileOutputStream.write(FileOutputStream.java:260)
at sun.nio.cs.StreamEncoder.writeBytes(StreamEncoder.java:202)
at sun.nio.cs.StreamEncoder.implFlushBuffer(StreamEncoder.java:272)
at sun.nio.cs.StreamEncoder.implFlush(StreamEncoder.java:276)
at sun.nio.cs.StreamEncoder.flush(StreamEncoder.java:122)
at java.io.OutputStreamWriter.flush(OutputStreamWriter.java:212)
at org.apache.log4j.helpers.QuietWriter.flush(QuietWriter.java:58)
at org.apache.log4j.WriterAppender.subAppend(WriterAppender.java:316)
at org.apache.log4j.WriterAppender.append(WriterAppender.java:160)
at org.apache.hadoop.mapred.TaskLogAppender.append(TaskLogAppender.java:58)
at org.apache.log4j.AppenderSkeleton.doAppend(AppenderSkeleton.java:251)
at org.apache.log4j.helpers.AppenderAttachableImpl.appendLoopOnAppenders(AppenderAttachableImpl.java:66)
at org.apache.log4j.Category.callAppenders(Category.java:206)
at org.apache.log4j.Category.forcedLog(Category.java:391)
at org.apache.log4j.Category.log(Category.java:856)
at org.apache.commons.logging.impl.Log4JLogger.info(Log4JLogger.java:199)
at org.apache.hadoop.mapreduce.task.reduce.ShuffleScheduler.freeHost(ShuffleScheduler.java:345)
at org.apache.hadoop.mapreduce.task.reduce.Fetcher.run(Fetcher.java:152)

If i run the wordcount test with 2 files, it works fine.

I have actually repeated this with my own code. I am working on something that requires me to map/reduce a small file and I had to work around the problem by splitting the file into 2 1MB pieces for my job to run.

All our jobs that run on 1 single larger file (over 1GB) work flawlessly. I am not exactly sure the threshold, From the testing i have done it seems to be any file smaller than the default HDFS block size (64MB) Sometimes it seems random in the 5-64MB range. But its 100% for the 5MB and smaller files.