How to establish zmq connections between docker containers using docker compose option network_mode: host? - docker

Question

I want to establish a connection using the draft server/client pattern of ZeroMQ across different docker containers. For that, I have a zmq.server that is running in docker_container_server. In addition, I have a zmq.client which runs in docker_container_client. Both containers are based on different docker images. Despite using the network_mode: host option with docker compose, this does not work. If I do the same within one container, it just works fine.
The docker-compose files are the following.
docker-compose-client.yaml:
services:
client:
privileged: true
image: client-image
container_name: client
command: tail -F anything
network_mode: host
environment:
- "ROS_MASTER_URI=http://127.0.0.1:11311"
- "ROS_HOSTNAME=127.0.0.1"
build:
context: ./client
network: host
dockerfile: Dockerfile
volumes:
- /var/run/docker.sock:/var/run/docker.sock
docker-compose-server.yaml:
services:
server:
image: server-image
privileged: true
container_name: server
network_mode: host
command: tail -F anything
volumes:
- /var/run/docker.sock:/var/run/docker.sock
build:
context: .
network: host
dockerfile: Dockerfile.server
The server application uses pyzmq as follows:
import zmq
class ServerExample():
def __init__(self):
self.stop = False
self.ctx = zmq.Context.instance()
self.url = 'tcp://127.0.0.1:5555'
self.server = self.ctx.socket(zmq.SERVER)
self.server.bind(self.url)
try:
while not self.stop:
msg = self.server.recv(copy=False) # stucks in this line
print(msg.routing_id)
if __name__ == '__main__':
server = ServerExample()
The client uses cppzmq as follows: (I also tried a pyzmq-client. This also works within a container but not across different containers.)
#include <zmq.hpp>
class ClientExample
{
private:
zmq::socket_t socket;
zmq::context_t ctx;
public:
ClientExample()
{
socket = zmq::socket_t(ctx, zmq::socket_type::client);
socket_.connect("tcp://127.0.0.1:5555");
char[] some_data;
socket_.send(zmq::buffer(some_data)); // not received by server
}
};
The containers are started by docker compose up server and docker compose up client.
When I do everything in one container and have a look on netstat -t | grep 5555, I get the same result in the container and on the host: (In this case, there are two clients and one server)
tcp 0 2856 localhost:56674 localhost:5555 ESTABLISHED
tcp 0 2856 localhost:5555 localhost:47666 ESTABLISHED
tcp 0 0 localhost:5555 localhost:56674 ESTABLISHED
That is why I think that the docker compose option network_mode: host is properly working in my case.
When I start the docker_container_server and docker_container_client, the execution stucks in the line where the server should receive a message. netstat -t | grep 5555 ouputs both on the host and in the containers just nothing.
The versions I use:
Ubuntu 20.04
Docker Compose version v2.12.2
Docker Engine Version 20.10.21
libzmq v4.3.4
cppzmq v4.9.0
pyzmq v25.0.0b1
I searched a lot through the internet, read the documentations of docker, docker compose, zmq etc. I came to the conclusion, that it should work with the network_mode: host option and I think I confirmed that the network_mode: host option works as it should. The code itself should not contain any mistakes, as it works fine within one container. But I didn't figure out, why the communication doesn't work across containers.