Managing host failures in a traffic forwarding system

Methods and apparatus for handling failure of servers in traffic forwarding (TF) systems between networks. A TF system may include units each including multiple servers. Outbound and inbound traffic for a local network may be distributed among the units according to a routing technique, with each unit responsible for an allocated portion of the traffic. Servers in a unit may participate in a health check protocol to detect servers that are not healthy. If the healthy servers in a unit drops below a threshold at which the unit cannot reliably handle its allocated portion of the traffic, the servers may automatically take the unit out of service, for example by stopping advertisement of routes, and the traffic may be reallocated across the remaining units. This may help prevent congestion-related delays, high latency, packet losses, and other problems on connections through the unhealthy unit.