I'd like to get contrib jars added to the general hadoop CLASSPATH so
that contrib classes are available to MR jobs. For example, a task that
wants to read from an hbase table will need to be able to load the
contrib class org.apache.hadoop.hbase.mapred.TableInputFormat. The
alternative, requiring that every job jar include its contrib
dependencies doesn't seem user friendly.
HADOOP-1648 is about messing w/ the bin/hadoop CLASSPATH construction to
shoe horn in contrib jars but it doesn't yet handle contrib lib
directories. hbase, at least, has its own lib subdirectory of a couple
of jars that it depends on that are not in HADOOP_HOME/lib.
HADOOP-1648 should add a general copy of contrib lib directories so they
too can be added to the hadoop CLASSPATH.
Where do folks think these extra dependencies should land when running
the ant 'package' target? Into contrib/lib along side all of the
contrib jars or contrib/CONTRIB_NAME/lib or into the main hadoop lib dir
in a subdir named contrib? (This latter would seem to make most sense to
me -- its contents could be added after those of lib).
Thanks,
St.Ack
