Splitting HA across datacentres

The only thing you really need to worry about as distances grow is the potential for split-brain situations (active-active) due to congestion or network latency.

I think many people already use F5 equipment in this kind of setup, as far as I know, you only really need to increase the Network Failover Timeout to account for the increased distance. The default is 3 seconds, which seems mighty enough for even your proposed distance, but I'd play it safe and raise this to 6 seconds.

tmsh modify sys db failover.nettimeoutsec value 6

For anything else, just use the normal common-sense approach.