I've looked all over and haven't found anyone asking this question.
My elasticsearch environment has each elasticsearch role running on different VMs. Example: 3 master nodes running on 3 separate VMs, 3 client nodes running on 3 separate VMs, 10 data nodes running on 10 separate VMs.
With this architecture if I want to do a full rolling restart (one at a time regardless of role) of the cluster is it necessary to disable shard allocation when bouncing each master and client node? My knowledge of elasticsearch makes me think I should only need to disable shard allocation when bouncing the data nodes.
Edit: https://www.elastic.co/guide/en/elasticsearch/guide/2.x/_rolling_restarts.html for reference.