First of all, I want to claim that I used CDH5 beta, and managed project
using maven, and I googled and read a lot, e.g.
https://issues.apache.org/jira/browse/MAPREDUCE-1700
http://www.datasalt.com/2011/05/handling-dependencies-and-configuration-in-java-hadoop-projects-efficiently/
I believe the problem is quite common, when we write an MR job, we need
lots of dependencies,
which may not exist in or conflict with HADDOP_CLASSPATH.
There are several options, e.g.
1. add all libraries to my own JAR, and set HADOOP_USER_CLASSPATH_FIRST=true
This is what I do, which makes the jar very big, and still it doesn't
work.
e.g. I already packaged guava-16.0.jar in my package, but it still use
guava-11.0.2.jar in the HADDOP_CLASSPATH.
below is my build configuration.
<plugin>
<artifactId>maven-assembly-plugin</artifactId>
<configuration>
<archive>
<manifest>
<mainClass>xxx.xxx.xxx.Runner</mainClass>
</manifest>
</archive>
<descriptorRefs>
<descriptorRef>jar-with-dependencies</descriptorRef>
</descriptorRefs>
</configuration>
<executions>
<execution>
<id>make-assembly</id>
<phase>package</phase>
<goals>
<goal>single</goal>
</goals>
</execution>
</executions>
</plugin>
2. distinguish which library is not present in HADDOP_CLASSPATH, and put it
into DistributedCache
I think it's hard to distinguish, and still if it conflicts, which
dependency would be precedent?
*What's the best practice, especially using maven?*
