current peer interactions are too fragile

currently any message that is sent as part of a larger protocol is sent once and a reply awaited. If this reply is not forthcoming within a set period we enter exceptional code...

what is needed is an algorithm which sends the message and waits a while, then checks to see if the recipient OR another node performing the same role is available and tried again - up to a maximum number of times, standing off for a little time between each try to give the cluster time to repair itself...

this way, an e.g. session evacuation would continue and succeed even if the original target node died and was replaced by another in mid interaction...