LTM Traffic Groups Failover - Is Layer 2 needed?

> Could you advise if the LTM will allow a floating traffic-group IP to be floated to a unit which isn't in the same L2 broadcast domain?

Yes - L2 adjacency is not required. Typically you are relying on ARP upstream of the BigIP, so the L2 is the same, but this is not required, as long as the upstream devices can sent the traffic to the currently active BigIP correctly.

Network-based HA and config-sync communications are routable. The HA vlan is intended to prevent HA and mirror traffic from impacting service traffic by isolating it from the internal/external networks.

However, you generally do want to have the BigIP peers having the same views and connectivity into the network for seamless failover, and the inter-device communication needs to be fast enough to prevent unexpected failovers. By default, the heartbeat timeout is 3 seconds.

K7249: Configuring the network failover timer

> Does the LTM issue a GARP when taking over the floating IP?

The LTM does GARP for the floating self-IP and all virtual IPs when failover occurs.

> I guess that if the above isn't possible, an MLAG/VPC up to a leaf pair would be the way to go and just connect both F5s to the same leaf pair.

if you have Advanced Routing licensed, you can advertise Floating Self-IPs/Virtual IPs via BGP directly from the BigIP

AskF5 | Manual Chapter: Dynamic Routing