HA primary gets stuck spinning on a perpetually ready external socket
This stems from https://support.isc.org/Ticket/Display.html?id=15378.
It is possible for the HA primary server to get into a state where an external socket is apparently marked ready and never cleared/satisfied. This causes the server to starve DHCP socket reading by endlessly detecting the external socket and eating 100% CPU. This condition is most easily reproduced using a 3 server HA hot-standby setup: primary, standby, and backup. It does not appear to be DHCP traffic volume issue, rather it appears simply be a code hole in the socket management.
I have attached the primary server config file I used. (The other servers files are the same except for HA server name).
I have only tested with v6, but the customer supplied top output shows both v4 and v6 servers eating CPU. I am pretty confident the issue is common to both.
NOTE: Further analysis demonstrates this condition is possible, but far less likely in two server HA setups.