hadoop运行时偶尔出现的问题 java.lang.Throwable: Child Error

[liuscmail]

log.txt:java.lang.Throwable: Child Error
log.txt:Caused by: java.io.IOException: Task process exit with nonzero status of 126.
log.txt:12/10/20 04:35:43 WARN mapred.JobClient: Error reading task outputhttp://hadoop-slave2:50060/tasklog?plaintext=true&attemptid=attempt_201209071933_64238_r_000014_0&filter=stdout
log.txt:12/10/20 04:35:43 WARN mapred.JobClient: Error reading task outputhttp://hadoop-slave2:50060/tasklog?plaintext=true&attemptid=attempt_201209071933_64238_r_000014_0&filter=stderr


在我们的hadoop集群中，偶尔在运行日志中发现这样的错误（并非每天都有，不固定的错误，第二次执行都不一定出现），小弟实在不能发现这个问题到底是什么原因引发的，还请各位大牛帮个忙，给解释一下撒

只知道126其实是hadoop child进程在执行shell脚本时候的退出码
bash的相关资料上给出的126说明，
126 Command invoked cannot execute ; Permission problem or command is not an executable

[码上富贵]

http://blog.sina.com.cn/s/blog_b88e09dd01014x2g.html

[java8964]

Negative. The above link is for return code 1. Error code 126 normally means the resource allocation error from system level. I don't know why when people ask for help here, it always is WITHOUT any information about the cluster. You should provide at least following information, so people can help you with more accurate answers: 1) Your hadoop version 2) JDK version 3) Your OS version Without these, I am just guessing now for you: 1) Read this, there is already a Jira ticket for this error on hadoop 1.0.x baseline, but fixed long time ago. https://issues.apache.org/jira/browse/MAPREDUCE-4857 2) Another possible reason: Your "mapred.child.ulimit" is set too low 3) Later version of hadoop (1.1.x) will also see "126 return code" due to another bug, fixed in 1.1.2. See this Jira ticket: https://issues.apache.org/jira/browse/MAPREDUCE-2374

[coolbamboo2008]

学习了，一楼不错