Migrating Elasticsearch node

Are there any special consideration for migrating elasticsearch nodes?

OS: CentOS
Old ES Version: 2.1.0
Old ES Cluster: 3 servers (5 shards each with 1 replica)
Old topology:
Server1 Node: master=true, data=true (only 1 node)
Server2 Node: master=true, data=true (only 1 node)
Server3 Node: master=false, data=true (only 1 node)

New ES Version: 2.4.0
New Topology:
Server1 Node: master=true, data=false | master=false, data=true (2 nodes)
Server2 Node: master=true, data=false | master=false, data=true (2 nodes)
Server3 Node: master=flase, data=false | master=false, data=true (2 nodes)

Are you going to perform an online migration? You'll want to use the same JVM version on all machines that are part of the same cluster. You should also keep in mind that shards won't move from a newer version of ES to an older. Finally, are you aware that a cluster with two master-eligible nodes will either not be fault tolerant (i.e. won't survive if one of the masters goes down) or will be susceptible to split brain situations? It seems people want to implement master-only nodes far more than what is warranted.

You'd be better off having 3 master nodes - https://www.elastic.co/guide/en/elasticsearch/guide/current/important-configuration-changes.html#_minimum_master_nodes

I am planning inline migration. I was planning the new master nodes pointing to the same "data" folder. Thus the follow up question.

What about the data folder?
Can the same data folder be used?
Once we have the new "data-nodes" (with the new data folder) will the data be spread across them seamlessly?

Ok. I will make three master nodes. Thanks for pointing out.

Can the same data folder be used?

No.

Once we have the new "data-nodes" (with the new data folder) will the data be spread across them seamlessly?

Yes, the shards will be distributed evenly. If your shards have roughly the same size you'll have roughly the same amount of data on each node.

Does the master node node need more memory or does the data node need more memory?

The role of being a master in a small cluster doesn't require much in terms of resources.