GTM resolves to inactive ip address

In the GTM pool load balancing methods, there is primary method, an alternate method, and a fallback method. If the primary method cannot be used (such as global availability while all pool members are down), then the GTM will fall back to the alternate method, and if that cannot be used, it will fallback to the fallback method. If the fallback method is "return to DNS", then it will respond with records from the BIND configuration, which often results in RR among all addresses in the pool. Change the fallback to "drop request" (or the equivalent - I don't recall the exact option), and DNS requests to that pool should be dropped when all members are failing.