Hi,
I make my first pull request here(just a minor fix):
https://github.com/apache/incubator-spark/pull/274
Another:
Spark use assemble plug in to generate a fat jar during build,
We should specify SPARK_JAR in the command line, then Spark upload spark
jar and user jar to the HDFS,
but yarn has a configurable option:
<property>
<name>yarn.application.classpath</name>
</property>
so we can do another way:
Exclude hadoop-* jar and hadoop related jars during build, which can
decreash the fat jar size. then put spark-fat-jar under HADOOP_LIB_DIR,
then Spark don't need to update spark jar to the HDFS, also we don't need
to specify SPARK_JAR in the command line.
If there is no problem, I want to do this change.
