Actually both / all.
In the systemd unit(s)
Setup caching agent: nano -w /etc/systemd/system/check_mk@.service
Change
ExecStart=-/usr/bin/check_mk_agent
to
ExecStart=-/usr/bin/check_mk_caching_agent
Adjust Client Socket MAX connections - nano -w /etc/systemd/system/check_mk.socket
comment out MaxConnectionsPerSource=3 (So there is no limit)
# MaxConnectionsPerSource=3
And then since the default rules still give crits/warning when services fail, I adjusted the systemd rule to ignore failed check_mk service failures. Alternately you can clear the failures on the monitored client via
systemctl reset-failed