[BUG] Gateway in a bridge-network container is unreachable from the host, and one gateway shows up as four nodes

[bburda]

Bug report

Steps to reproduce

1. Install ros-jazzy-ros2-medkit-gateway in a container on a bridge network and publish the port (-p 8080:8080).
2. In the container run ros2 launch ros2_medkit_gateway bringup.launch.py.
3. On the host run curl http://localhost:8080/api/v1/areas.
4. In the container run ros2 node list and ros2 param list /ros2_medkit_gateway.

Expected behavior

• The host reaches the API through the published port, or the docs name the launch argument that makes it work.
• One gateway process shows up as one node.

Actual behavior

• The host gets curl: (56) Recv failure: Connection reset by peer. The startup log says REST API at 127.0.0.1:8080, so the gateway listens on the container's loopback. The troubleshooting page only lists "Connection refused". The README tells native users to set cors.allowed_origins, which the launch files already do.
• ros2 node list shows /ros2_medkit_gateway four times and warns about duplicate names. The log has Publisher already registered for node name: 'ros2_medkit_gateway' three times, and ros2 param list /ros2_medkit_gateway prints every parameter four times. The fault-service clients, the lifecycle state reader and the subscription executor each create an rclcpp::Node, and all of them take the -r __node:=ros2_medkit_gateway that launch_ros adds to the whole process.

Environment

• ros2_medkit version: 0.6.0 (apt) and main
• ROS 2 distro: Jazzy
• OS: Ubuntu 24.04

Additional information

The gateway unit suite also times out now and then under ctest -j16. test_discovery_manager, test_ros2_topic_data_provider and test_ros2_subscription_executor hang when TearDown cancels the executor before the spin thread has entered spin(). test_node_death_integration reads fault reports after a fixed sleep and sometimes sees a report one tick late.