A mapreduce job that depends on a contrib jar such as hbase must take
the extra step of either building the contrib jar into its job jar or
copying the contrib jar to ${HADOOP_HOME}/lib across the cluster. While
a relatively minor inconvenience, what do folks think of just adding
contrib classes/jars to the CLASSPATH built by bin/hadoop?
Thanks,
St.Ack
P.S. Below is a suggested (untested) patch
Index: bin/hadoop
===================================================================
--- bin/hadoop (revision 556809)
+++ bin/hadoop (working copy)
@@ -95,7 +95,11 @@
if [ -d "$HADOOP_HOME/build/test/classes" ]; then
CLASSPATH=${CLASSPATH}:$HADOOP_HOME/build/test/classes
fi
+if [ -d "$HADOOP_HOME/build/contrib/*/classes" ]; then
+ CLASSPATH=${CLASSPATH}:$HADOOP_HOME/build/contrib/*/classes
+fi
+
# so that filenames w/ spaces are handled correctly in loops below
IFS=
@@ -106,6 +110,9 @@
for f in $HADOOP_HOME/hadoop-*-core.jar; do
CLASSPATH=${CLASSPATH}:$f;
done
+for f in $HADOOP_HOME/contrib/*.jar; do
+ CLASSPATH=${CLASSPATH}:$f;
+done
# add libs to CLASSPATH
for f in $HADOOP_HOME/lib/*.jar; do
