5.5.4. Failover and Failback Procedures

In the event that a transport failure is detected with a peer, it is necessary for all pending request messages to be forwarded to an alternate agent, if possible. This is commonly referred to as "failover".

In order for a Diameter node to perform failover procedures, it is necessary for the node to maintain a pending message queue for a given peer. When an answer message is received, the corresponding request is removed from the queue. The Hop-by-Hop Identifier field is used to match the answer with the queued request.

When a transport failure is detected, if possible, all messages in the queue are sent to an alternate agent with the T flag set. On booting a Diameter client or agent, the T flag is also set on any remaining records in non-volatile storage that are still waiting to be transmitted. An example of a case where it is not possible to forward the message to an alternate server is when the message has a fixed destination, and the unavailable peer is the message's final destination (see Destination-Host AVP). Such an error requires that the agent return an answer message with the 'E' bit set and the Result-Code AVP set to DIAMETER_UNABLE_TO_DELIVER.

It is important to note that multiple identical requests or answers MAY be received as a result of a failover. The End-to-End Identifier field in the Diameter header along with the Origin-Host AVP MUST be used to identify duplicate messages.

As described in Section 2.1, a connection request should be periodically attempted with the failed peer in order to re-establish the transport connection. Once a connection has been successfully established, messages can once again be forwarded to the peer. This is commonly referred to as "failback".