Sophos XG wrong balancing behaviour

        I've a Sophos XG210 (SFOS 19.5.0 GA-Build197) cluster, with a load balancing rule (with Sticky IP Method) pointing a couple of servers on port 443; in the rule I activated the health check with a TCP probe on port 443.

The balancing works fine, but when I try to shutdown the destination service on one of the servers, the health check does not work as expected, sending traffic toward to actives and NOT actives. Sniffing the traffic, I see the probe (that correctly receives a RST) and I don't have any idea. Suggestions?