The detection and handling of network disruptions between slave HCs and masters is not working properly.
1) Network disruptions are not being detected until an operation is invoked from the master to the slave, and the detection will not happen until the TCP layer give up trying to transmit the message, which could take up to 20 mins with default linux tcp settings.
2) If no operation is invoked, or the TCP level detection has not happened yet, the master still regards the slave as registered. So if the network heals and the slave tries to reregister, the re-registration attempt will be denied.
To solve this I am going to re-implement a variant of the ping-pong stuff we had in 7.0.