My production Hadoop 0.20.2 cluster has been running Pri. namenode and
Jobtracker together in the same box for long time. it's time to
separate them. I am trying to write up a procedure on this to make
sure that it won't cause any issues.
- Decrease TTL of DNS entry for value of mapred.job.tracker
- Prepare new name in DNS entry for value of mapred.job.tracker
- Change value of mapred.job.tracker (in mapred-site.xml) to new
server and propagate the new config to whole cluster and clients (
where Hadoop jobs are submitted from )
- Create directory defined in mapred.local.dir on new server.
- Make sure that new jobtracker server is not in slave or master file.
- run stop-mapred.sh on current jobtracker server
- run start-mapred.sh on new jobtracker server
Do I miss anything ?