Clarification on proxy backend states (Zombie vs Dead) and possible packet loss

[Erdal Emlik]

Hello FreeRADIUS team,



We are running FreeRADIUS as a proxy setup, and we are trying to better understand how backend states (Alive, Zombie, Dead) behave in practice.



In our current configuration:



response_window = 5s (default is 20s)
zombie_period = 20s (default is 40s)

From what we observe:



If a backend does not reply within response_window, the proxy marks it as Zombie.
The backend remains Zombie until the zombie_period expires. If no valid responses are received during this time, it becomes Dead.
If there are no Alive backends left, the proxy prefers sending to a Zombie backend instead of dropping requests.
Periodic Status-Check packets can bring a Zombie backend back to Alive.


👉 My questions / doubts:



In the Zombie state, we sometimes see packet loss (packets not reaching the backend). Is this the expected behavior, since the proxy still forwards traffic to a potentially unstable backend?
When the documentation says “if the home servers are still down” in the context of robust-proxy-accounting, does “down” strictly mean Dead, or are Zombie servers also considered “down”?
In our understanding, if a backend is in Zombie state, packets may still be sent but can get lost due to intermittent connectivity (before the backend ever becomes Dead). Is this correct?
What would be the recommended approach to avoid accounting packet loss in such scenarios ( i know network must be stable but shit happens :(…)— should we rely on robust-proxy-accounting to queue locally until the backend is fully Alive again?

Any clarification on this behavior would be very helpful.

Thanks in advance,