This was actually an issue in previous versions, in (I believe 0.7) after a timeout (reconnect_timeout and reconnect_timeout_wan in options) a node will be completely removed.