Manuel, You can also just start the second HDFS cluster in parallel, and do an "hadoop fs -cp hdfs://original-nn/hbase hdfs://new-nn/hbase" (or a distcp) and then start HBase services on the new cluster (make sure zk quorum is separate or has a different hbase-rootdir though).
On Tue, Apr 17, 2012 at 2:22 PM, Manuel de Ferran <[email protected]> wrote: > Greetings, > > we have a 4 nodes cluster running HBase-0.90.3 over Hadoop-0.20-append. > We'd like to create another HBase cluster from this one with minimal HBase > downtime. We have plenty of disk space on each datanode. > > Here is what we have in mind: > - Add a new datanode aka. DN5 > - Raise HDFS replication factor to 5 to have a whole copy on each datanode > - Wait until replication done > - Disable all tables > - Stop DN5 > - Copy Namenode data (dfs/name/current ...) to DN5 > - Enable all tables > - Start a new namenode on DN5 aka NN2 > - Reconfigure DN5 to point to NN2 > - Configure a new HBase cluster on top of the new HDFS > > It works on a small cluster but is it enough to have a consistent copy ? > > Any hints ? Is there a best-practice ? > > Thanks -- Harsh J
