Docker swarm node communication issues

[ytk929]

Network Environment： One ubuntu server is userd as the swarm manager node, The ip address of one of its network interfaces is 192.168.200.82, and the route "to 10.3.71.0/24" is configured via netplan, and the default route is on another interface. One CentOS server, only one network interface, the ip address is 10.3.71.15, as a swarm worker. The network administrator maps 10.64.128.100 to the manager node through the proxy server so that worker 10.3.71.5 can access the manager, and then maps ports 12377/tcp, 17946/tcp, 7946/udp, and 4798/udp from 10.64.128.100 to ports 2377/tcp, 7946/tcp, 7946/udp, and 4798/udp on 192.168.200.82.

Qustion： The containers between the manager node and the worker node cannot communicate with each other. The worker node can successfully join the swarm by executing the command 「 docker swarm join-token 10.64.128.100:12377」 and the manager node can deploy the service to the worker node, and all containers join the same overlay mode network. All containers in the manager node can communicate with each other, and the same in the worker node. But the containers between the manager node and the worker node cannot communicate with each other.

Investigation： On the server 192.168.200.82, I can successfully telnet to port 7946/tcp of 10.3.71.15, and successfully nc to port 7946/udp and 4798/udp of 10.3.71.15. On the server 10.3.71.15, I can successfully telnet to port 12377/tcp and 17946/tcp of 10.64.128.100, and successfully nc to port 7946/udp and 4798/udp of 10.64.128.100. On the server 10.3.71.15, I can see the swarm manager address is 192.168.200.82 by "docker info". I use tcpdump to see that 10.3.71.15 is communicating with port 7946/tcp and 7946/udp and 4798/udp of 192.168.200.82 , then I use iptables tool to redirect the traffic of 7946/tcp and 7946/udp and 4798/udp of 192.168.200.82 to port 17946/tcp and 7946/udp and 4798/udp of 10.64.128.100. I use tcpdump again and find that 10.3.71.15 is communicating with port 17946/tcp and 7946/udp and 4798/udp of 10.64.128.100, but the containers between the manager node and the worker node still cannot communicate with each other. I want to ask what might be the reason?