What Vaibhav said; just take out the regionservers gently by shutting each down using hbase-daemon.sh stop -- preferably not the hosts carrying -ROOT- and .META if you can avoid it (less churn if you don't have to take these down). Once down and their regions are deployed elsewhere, you'll want to turn off the datanodes. Here you need to be more careful making sure you don't decommission too many at a time (Your block replication is 3 is my guess.. this means that you could shut down 3 nodes and thereby removing all copies of a particular block). Check over in the hadoop mailing lists to see if the science of decommissioning has improved since the days when you'd up replication all around cluster -- perhaps running the hdfs balancer to speed block movement -- then checking with fsck as you took down a node or two at a time making sure you'd not done bodily injury to files up in hdfs.
St.Ack On Wed, Oct 21, 2009 at 2:25 PM, Vaibhav Puranik <[email protected]> wrote: > Rolling Restart would be the best option in my opinion. > > http://wiki.apache.org/hadoop/Hbase/RollingRestart > > Regards, > Vaibhav Puranik > GumGum > > On Wed, Oct 21, 2009 at 2:22 PM, Farshad Kazemzadeh > <[email protected]>wrote: > > > > > Hello, > > > > I am trying to take down 9 nodes out of our 20 node Hadoop/HBase cluster. > > What is the best way to do this without having to shut down the entire > > cluster and have things to continue to run without a hiccup? > > > > Thank you in advance, > > FarshadK > > _________________________________________________________________ > > Hotmail: Trusted email with Microsoft’s powerful SPAM protection. > > http://clk.atdmt.com/GBL/go/177141664/direct/01/ > > >
