Detecting a connection interruption
Recall that clients maintain two kinds of connections to the server environment:
A web socket connection the ICE Server for management and provisioning functions. When this connection has failed, the user’s avatar (visible in the top-right corner of the ICE Desktop and ICE Mobile applications) will appear drawn with an animated glowing yellow ring.An SSL connection to the Rallypoint (or, potentially, many SSL connections to many Rallypoints depending on deployment); used to convey audio traffic on channels configured to use a Rallypoint. When this connection has failed channels reliant on the connection will display a 'Limited connectivity' message with a slash through the channel’s cloud icon.In the simplest case, clients are made aware of a connection interruption whenever their underlying operating system informs them that the socket has been closed. However, it’s possible for a loss of service to occur that does not cause the socket connection to be terminated at all, or terminated in a timely manner.
To recover from such failures, both connections utilize a form of ping-pong messages that are sent back and forth between the client and server components to detect such outages. When a client sends a ping to the server and does not receive a timely 'pong' message in response the client will presume the server/Rallypoint to be unreachable or out of service and attempt to reconnect to an alternate ICE Server ingress address or to an alternate Rallypoint. Similarly, requests made of the ICE Server that are not fulfilled also trigger the client to assume an outage has occurred.
The amount of time required by the client to detect a connection interruption depends on the type of failure and the component whose link has failed. Failures in which the client’s operating system detects a socket closure typically occur quite quickly—within a matter of seconds (if not milliseconds). Failures in which the socket remains open but the server becomes unresponsive typically take 30 seconds to a minute to detect. In the case of an oversubscribed server component whose ping-pong messages may be received intermittently can result in marginal service or delays in detecting an outage.