I experimented with allocating a client on a server node by installing a fresh instance of Netmaker server (10.101.0.1) on a node and by creating two clients, the first (10.101.0.2) on the server node and the second (10.101.0.3) on a normal node. The connection from the server client to the normal client was reliable.

server # ping -c100 10.101.0.3 | grep loss
100 packets transmitted, 100 received, 0% packet loss, time 99123ms

The connection from the normal client to the server client was not reliable.

normal # ping -c100 10.101.0.2 | grep loss
100 packets transmitted, 43 received, 57% packet loss, time 99658ms

I found that the server client was listening on port

51115

.

server # wg show nm-default | grep listening
  listening port: 51115

But the port on which the normal client was was trying to contact the server client was alternating between

51115

(the correct port) and

51821

(one of the ports dedicated to the Netmaker server). Packets were lost unless the port was

51115

.

normal # wg show nm-default | grep 10.101.0.2 -B1 | head -n 1 ; ping -c1 10.101.0.2 | grep loss
  endpoint: [SERVER IP ADDRESS]:51821
1 packets transmitted, 0 received, 100% packet loss, time 0ms

normal # wg show nm-default | grep 10.101.0.2 -B1 | head -n 1 ; ping -c1 10.101.0.2 | grep loss
  endpoint: [SERVER IP ADDRESS]:51115
1 packets transmitted, 1 received, 0% packet loss, time 0ms

normal # wg show nm-default | grep 10.101.0.2 -B1 | head -n 1 ; ping -c1 10.101.0.2 | grep loss
  endpoint: [SERVER IP ADDRESS]:51821
1 packets transmitted, 0 received, 100% packet loss, time 0ms

It appears that the Netmaker control program that runs on the server has a logic error and does not consistently configure normal clients to access the listening port of the server client.