secondary namenode not configured

[Kevin W Monroe]

When restarting a hadoop cluster from the hdfs-master using /usr/lib/hadoop/sbin/[stop|start]-dfs.sh, the systems attempts to stop/start the secondary namenode process on the master. This is because we have no configuration set for the secondary, and it defaults to 0.0.0.0 (effectively localhost). We must set this on the hdfs-master's /etc/hadoop/conf/hdfs-site.xml:

<property>
 <name>dfs.secondary.http.address</name>
 <value>secondary-namenode-0:50090</value>
 <description>Enter your Secondary NameNode hostname.</description>
</property>

With this value set, the system will ssh to the secondary and stop/start the process there. This means we'll also need the hdfs-master to have password-less ssh access to the hdfs user on secondary-namenode-0 -- we already do this for datanodes, so we'll just need something similar for the secondary relation.