Using a single Prometheus node with redundant drives rather than redundant nodes with single drives is an odd choice to me. Why replicate at block level when layer 7 has support?
(It also requires more hardware.)
(I'm the author of the linked-to entry.)