Diagnose the LB_FAILED Event

make sure your pool members have a route back via F5, sounds like a routing issue. LB_FAILED happens when the TCP handshake to the pool member is not completed successfully.

Try enabling auto-snat

I’m guessing you are using gateway_icmp as a health monitor (which is connectionless). This will mark your pool up, however a TCP connection might fail as it require a working return path via the F5