On Thu, Apr 8, 2010 at 9:46 PM, Ted Yu <yuzhih...@gmail.com> wrote: > Feel free to comment in this JIRA: > https://issues.apache.org/jira/browse/HADOOP-6695 >
Please read the documentation cited above. It discusses the three options including one which does not necessitate restart of MR cluster, the one Todd refers to, where you bundle all your dependent jars as one fat mapreduce jar and pass this as your -jar starting your job. The hbase.jar doesn't bundle all dependencies. We could I suppose make a hbase-all.jar that has all in it and that also could serve as a mapreduce job jar (currently its Main-Class is a driver which will direct you to a particular hbase MapReduce such as import/export/rowcounter switching off the passed command-line parameter). St.Ack
