We are testing out a topology in the lab, with 2 PA-2020 in an active/passive HA cluster. They are between 2 pairs of Cisco switches and should play a role of redundant in-line firewalls. The connection to the switches is with FO modules on ports e1/13 and e1/14 (these ports are in a monitor group).
What we have noticed is some strange behaviour, and it is the same with PANOS 4.1.9. and 4.1.11.
If we pull out the cable on port e1/13 on the primary/active device, the firewalls failover, and the secondary/passive device becomes secondary/active. The now primary/passive will go to a non-functional state, and after a minute to passive, and will then again move to the active state. Of course, the cable is still unplugged, so the failover happens again, and the secondary device becomes active once more. The process will continue until the primary device moves into a suspended state (3 times by default). The data traffic is highly effected with the failovering and spanning-tree recalculations on the Cisco switches.
When we disable the preemption, this does not happen, and failovering worked perfectly through different scenarios.
So, my question is - should the preemption be disabled in vwire active/passive HA? I have not found any reference or configuration best practice for this kind of topology in any document.
To me it seems logical that the firewall should check the state of the monitored interfaces (or path) before trying to resume its active role, even with preemption enabled.