I'm using HBase 0.94.2.
Looking at TableMapReduceUtil.addDependencyJars(Job job) I see that
org.apache.zookeeper.ZooKeeper.class and com.google.protobuf.Message.class
are hard-coded as jars to send to cluster.
I can understand why all the others (map output key/value class, etc.) are
sent to cluster but if you are using HBase you are supposed to have
ZooKeeper deployed in your cluster, right? and protobuf is a part of that
installation, isn't it ?
Are these two jars necessary ?
In general, If I have all the classes I need deployed in the cluster
(including map output key/value class, etc.), can I skip
"addDependencyJars" ? I'm using an extended HFileOutputFormat I wrote...