Clients will poll the server. Depending on the size of the fleet, this interval goes as low as 10s in production deployments.