[MAPREDUCE-576] writing to status reporter before consuming standard input causes task failure.

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 0.20.1
• Component/s: contrib/streaming

Environment

amazon ec2 instance created with the given scripts (fedora, small)

• Assignee: Todd Lipcon
• Reporter: Karl Lehenbauer
• Created: 20/Oct/08 23:33
• Updated: 05/Dec/09 00:08
• Resolved: 05/Dec/09 00:08

Description

A Hadoop Streaming task which writes a status reporter line before consuming input causes the task to fail. Writing after consuming input does not fail.

I caused this failure using a Python reducer and writing a "reporter:status:foo\n" line to stderr. Didn't try writing anything else.

The reducer script which fails:

#!/usr/bin/env python

import sys

if _name_ == "_main_":
sys.stderr.write('reporter:status:foo\n')
sys.stderr.flush()
for line in sys.stdin:
print line

The reducer script which succeeds:

#!/usr/bin/env python

import sys

if _name_ == "_main_":
for line in sys.stdin:
sys.stderr.write('reporter:status:foo\n')
sys.stderr.flush()
print line

The hadoop invocation which I used:

hadoop jar /usr/local/hadoop-0.18.1/contrib/streaming/hadoop-0.18.1-streaming.jar -mapper cat -reducer ./reducer_foo.py -input vectors -output clusters_1 -jobconf mapred.map.tasks=512 -jobconf mapred.reduce.tasks=512 -file ./reducer_foo.py

This is on a 64 node hadoop-ec2 cluster.

One of the errors listed on the failures page (they all appear to be the same):

java.io.IOException: subprocess exited successfully
R/W/S=1/0/0 in:0=1/41 [rec/s] out:0=0/41 [rec/s]
minRecWrittenToEnableSkip_=9223372036854775807 LOGNAME=null
HOST=null
USER=root
HADOOP_USER=null
last Hadoop input: |null|
last tool output: |null|
Date: Mon Oct 20 19:13:38 EDT 2008
MROutput/MRErrThread failed:java.lang.NullPointerException
at org.apache.hadoop.streaming.PipeMapRed$MRErrorThread.setStatus(PipeMapRed.java:497)
at org.apache.hadoop.streaming.PipeMapRed$MRErrorThread.run(PipeMapRed.java:429)

at org.apache.hadoop.streaming.PipeReducer.reduce(PipeReducer.java:103)
at org.apache.hadoop.mapred.ReduceTask.run(ReduceTask.java:318)
at org.apache.hadoop.mapred.TaskTracker$Child.main(TaskTracker.java:2207)

The stderr log for a failed task:

Exception in thread "Timer thread for monitoring mapred" java.lang.NullPointerException
at org.apache.hadoop.metrics.ganglia.GangliaContext.xdr_string(GangliaContext.java:195)
at org.apache.hadoop.metrics.ganglia.GangliaContext.emitMetric(GangliaContext.java:138)
at org.apache.hadoop.metrics.ganglia.GangliaContext.emitRecord(GangliaContext.java:123)
at org.apache.hadoop.metrics.spi.AbstractMetricsContext.emitRecords(AbstractMetricsContext.java:304)
at org.apache.hadoop.metrics.spi.AbstractMetricsContext.timerEvent(AbstractMetricsContext.java:290)
at org.apache.hadoop.metrics.spi.AbstractMetricsContext.access$000(AbstractMetricsContext.java:50)
at org.apache.hadoop.metrics.spi.AbstractMetricsContext$1.run(AbstractMetricsContext.java:249)
at java.util.TimerThread.mainLoop(Timer.java:512)
at java.util.TimerThread.run(Timer.java:462)

Issue Links

• duplicates: HADOOP-5623 Streaming: process provided status messages are overwritten every 10 seoncds