Each Radius server is described by an FSM (Finite State Machine).

A Radius server can be in one of six states:

• Initial — the state when fastPCRF starts;
• Alive — the Radius server is accessible, requests can be sent to it;
• Dead — the Radius server is inaccessible, requests cannot be sent;
• Alive-ping-sent — a ping has been sent to the Radius server in the Alive state, waiting for a response;
• Dead-ping-sent — a ping has been sent to the Radius server in the Dead state, waiting for a response;
• Disabled — the server is administratively prohibited.

The main states are Alive and Dead, while the other states are transitional. A Radius server is considered accessible if it is in the Alive or Alive-ping-sent states. In other states, the Radius server is considered inaccessible and requests cannot be sent to it.

Ping — is a Status-Server request, or if the Radius server does not support Status-Server (the `radius_status_server` option), an Access-Request with User-Name and Password equal to the `radius_ping_user_name` and `radius_ping_user_password` options, respectively.

Every 10 seconds, a watchdog task runs to check whether a ping should be sent, whether a response to the ping has been received, and based on this, it changes the server's state or leaves it unchanged.

When fastPCRF starts, all Radius servers are in the Initial state. fastPCRF sends a ping to them and transitions to the Dead-ping-sent state, waiting for a ping response.

In the Alive state, the server is pinged only if no responses have been received from it within the last `radius_keepalive` seconds. If the server is responding to requests and there is continuous request-response exchange, pings are not sent. When a ping is sent, the server transitions to the Alive-ping-sent state (waiting for a ping response).

In the Dead state, the server is pinged every `radius_dead_timeout` seconds: a ping is sent, and the server transitions to the Dead-ping-sent state. For example, if `radius_dead_timeout=3600` (1 hour), the Radius server cannot transition to the Alive state earlier than one hour after transitioning to the Dead state.

In the DeadDisabled or Alive states, the watchdog notifies fastPCRF that a particular server has become unavailable or available, respectively.
Important! Reservation of Radius servers is implemented according to the Master-Slaves scheme: only one Radius server is considered operational (master), the rest are backups (regardless of their state). Radius servers are specified in a list (each Radius server in a separate `radius_server` option), with the most preferred server being the one specified in the first `radius_server` option in `fastpcrf.conf`, followed by the less preferred ones, and so on. If the server with number N > 1 is the master, it means all servers before it (1..N-1) are unavailable (Dead). If any of the servers 1..N-1 becomes alive (Alive), it immediately becomes the master, and all requests go to it. Responses from the previous master are accepted.

If the current master becomes unavailable, the first server in the Alive state is searched for and becomes the master.

Authorization Radius servers do not receive any messages when they gain or lose master status. However, accounting servers do: if an accounting server loses master status, it is sent Accounting-Off; if an accounting server gains master status, it is sent Accounting-On. The following attributes are sent in Accounting-On/Off:

• NAS attributes: NAS-IP-Address, NAS-IPv6-Address, NAS-Identifier. Either only one of these attributes is sent (in the priority order listed here — the highest priority is NAS-IP-Address), or if the `radius_add_all_nas_ids=1` option is set, all known attributes are sent;
• Acct-Status-Type = Accounting-On or Accounting-Off;
• Acct-Session-Id = 0 — this acct-session-id value cannot be encountered by any acct session.