Is your layout like this?
- DC1 has an LTM with a pool containing the 3 servers in DC1;
- DC2 has a different LTM with a pool containing the 3 servers in DC2;
- The GTM cluster contains the LTMs, and the LTM objects on the GTM use the bigip monitor;
- Each LTM uses a monitor to detect the application health on the servers in its pool;
- A system outside of the BIG-IPs decides when to turn up a server instance.
If so, when you say a server instance is "up" does that mean it responds to the monitor, and when it is "down", it does not?
If this is the case, then BIG-IP monitoring solves your problem. Let's say all servers in DC2 are "down". The LTM Virtual Server fronting them will also be "down" (because all of the pool members for that VS will be "down"). If a GTM pool has that LTM Virtual Server in it, the GTM would not supply the VIP on LTM in DC2 as an A record for a WIP using that GTM pool. That is, the GTM will not direct traffic the DC2 LTM VS. As soon as any servers in DC2 are declared "up" by the LTM health monitor, the GTM would start using the VS VIP for the LTM in DC2. That is, it would start directing traffic to the DC2 LTM VS (and continue directing traffic to the DC1 LTM VS, if any of the DC1 servers are still "up").