We have been running out of disk recently with some indexes requiring
more than 400GB of space per index replica. Actual disk space
requirements are double that as we run with a replica. On top of that
the idea is that 5 of 6 elasticsearch nodes have enough space for all
our data so that we are resilient to losing a node.
Napkin math:
400 * 10 * 2 = ~8TB of disk
400 * 7 * 2 = ~5.6TB of disk
Each of the six ES nodes has 1TB of disk allocated to ES so 5.6TB should
get us just under the limit. Then for handling a node outage weekends
tend to not have as many records so our actual usage should be a little
lower.
Change-Id: Ie677bd47a9886870bc83876d2407742133299861
bd752a0bfe