#upgrading hadoop from 0.2 to 2 
#upgrade require re-kickstarts (3.2.21)
#roll osg-6.1.1-2 (3.2.21)

#upgrade roll (including creation of group zookeeper if needed)
/opt/rocks/bin/rocks add attr OSG_zookeeperuid value=406
/opt/rocks/bin/rocks add attr OSG_zookeepergid value=406
groupadd -g 406 zookeeper
rocks remove roll osg
rocks add roll osg-6.1.1-2.x86_64.disk1.iso  #include path if needed
rocks enable roll osg
cd /export/rocks/install
rocks create distro

#reinstall rocks-osg-command (make sure comes from latest roll version )
yum clean all
yum install rocks-osg-command

#if mounted hadoop dismount first
rocks run host compute compute='umount /mnt/hadoop'
#stop hadoop in all nodes (in particular namenode)
rocks run host compute compute='service hadoop stop'

#edit extend-compute.xml to turn off old hadoop services
#rocks create distro if needed
# backup '$HADOOP_SCRATCH/dfs/name' on namenode

#re-kicksstart nodes

#recover  '$HADOOP_SCRATCH/dfs/name' on namenode

#on namenode
service hadoop-hdfs-namenode upgrade #only first time later start

#on datanode
service hadoop-hdfs-datanode start

#on secondary
#http://www.cloudera.com/content/cloudera/en/documentation/cdh4/v4-3-0/CDH4-Release-Notes/cdh4ki_topic_2_2.html
mv /home/hadoop/current /home/hadoop/current_0.20bak
mv $HADOOP_SCRATCH/current $HADOOP_SCRATCH/current_0.2bak
service hadoop-hdfs-secondarynamenode start


#not required, just in case to finalize upgrade (since $HADOOP_SCRATCH/dfs/name/previous does not exist)
#hdfs dfsadmin -finalizeUpgrade

#note:mar 24,2015
#after -finalizeUpgrade in production, I found that restarting nodes (hdfs-hadoop-datannode) give me back about 100Gb/node
#data nodes had $HADOOP_SCRATCH/previuos dirs that were processed then removed. 
# for upgrades is a big deal this last command AND restart nodes.