Hi
Jimmy Chen !
There has similar requirements in out company (Alibaba. in China). we construct a data syncing pipeline between A and B(A and B just as your mentioned). like A <=> B. And the only thing we should take action on one data center is gone. is just switching the use request and net traffic to another one. Since they have the same mirror data.
In this scenario. A and B don't need to deploy as an ReplicaSet each other. That means no heartbeats and primary election between the long distance network....