There are use cases where customer has to implement High Availability ( HA) architecture for Mashery Local specially in relation to On-Premise ( Swarm / Kubernetes) 

Best-Practice-to-implement-High-Availability-in-Mashery-Local

Attached document can be used as a reference point which is a little more geared towards Swarm/On-Prem but has other considerations called out. Kubernetes handles resiliency, it seems, a bit better than Swarm out of the gate and our solution takes advantage of that.

Is the use case is running in a single point of presence?  If so, having multiple clusters won’t help as much as having additional nodes in K8s to sustain any potential failures/outages or planned maintenance events. If  running in multiple POPs and are running active/active across both, then one would ideally have each POPs local capacity requirements plus 50-100% extra in case the other POP entirely goes down and this one needs to be able to handle the additional processing. This also assumes that customer has that level of business continuity extended to the underlying services being served in each POP.

If traffic patterns are steady, it’s easier to size for that volume of request + 10-20% for spikes and outliers. If it’s very spikey, evaluation needs to be done whether capacity available for more rapid adaptation to those potential changes vs.  leveraging a setup that will auto-provision additional nodes/Mashery Local components where and as needed.