Are you using the "job" file rather than the "jar" file? The former is the one with all transitive dependencies bundled. You do not need to set any classpaths.
On Thu, Dec 8, 2011 at 5:37 PM, Gary Snider <[email protected]> wrote: > I have a pojo that calls KMeansDriver. I'm on 0.6-SNAPSHOT Ubuntu Linux > > Once the mapper gets run in Hadoop, hadoop complains with: > Caused by: java.lang.NoClassDefFoundError: com/google/common/collect/Lists > at > > org.apache.mahout.clustering.kmeans.KMeansMapper.<init>(KMeansMapper.java:37) > ... 12 more > > I've tried different combinations of setting classpath but nothing helps. > com.google.common.collect.Lists is in guava-r09.jar > 1) setting $HADOOP_CLASSPATH to point to ALL possible jars before starting > hadoop > 2) setting $CLASSPATH before starting hadoop > 3) passing -classpath to my pojo > 4) setting $CLASSPATH before running my pojo > > The thing is, if it runs 'locally' from the local filesystem it works. > Once I point it to hadoop, I get the error. > Is this a hadoop classpath/classloader issue? >
