I worked on up gradation of production Hadoop cluster along with the application. The process we followed is:
- We build a small cluster from desired distribution of hadoop which support YARN.
- We tested entire code like Map/Reduce program, Pig program hive script, sqoop script etc in new cluster and did the required changes in code. Generally there is not too much code side change. Just need to validated supported function, methods etc.
- Verified the output from old output, If your output match with old output then your code is good to go.If there is mismatch then you need to fix your code.
- Upgraded the hadoop cluster.
- Deployed the code on upgraded cluster.
Note: You cannot replace only MapReduce version 1 with YARN. You also need to upgrade pig, hive, sqoop etc. I suggests pic up the HDP stack which you need and then upgrade your cluster. All supported tools, utility and software should be upgraded with supported version of Hadoop.