when a datanode is terminated, all its data is unreachable; it is the equivalent of a node crashing.

When a datanode is decommissioned, the namenode needs to know about it; it needs to pull the data off onto different nodes for a clean shutdown. We should provide workflow components to do this.

There is also the option of having a node do a clean shutdown when terminated. Here the termination time would be long -the time to move off all data-, and it could fail partway through. This is why a separate decommision operation would seem better