Namespacing
If you have something like:
1
2
3
4
5
class FiveCardStudPlayer : public rclcpp::Node {
public:
FiveCardStudPlayer() : Node("five_card_stud_player"){
...
}}
And you launch 2 instances of five_card_stud_player, ROS2 Humble won’t crash the nodes. Instead, it will have two nodes with the colliding names
1
2
3
/five_card_stud_dealer
/five_card_stud_player
/five_card_stud_player ← collision!
You can address this issue by using NodeOptions
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
class MyNode : public rclcpp::Node {
public:
MyNode(const rclcpp::NodeOptions & opts)
: Node("my_node", opts)
{ /* … */ }
};
int main(int argc, char ** argv)
{
rclcpp::init(argc, argv);
auto args = rclcpp::remove_ros_arguments(argc, argv);
rclcpp::NodeOptions opts;
opts.arguments(args);
auto node = std::make_shared<MyNode>(opts);
rclcpp::spin(node);
rclcpp::shutdown();
return 0;
}
std::make_shared<rclcpp::Node>(DEFAULT_NAME, opts)will automatically change the name of the nodenode->get_name()will reflect the change- Then you can call it with
ros2 run card_deck_game five_card_stud_player --ros-args --remap __node:=<unique_name>
Weird Issues
-
You could have two instances of the same launch files of the same action server running. They could throw this error
1
[capacity_manager_node-1] [ERROR] [1750698354.440384729] [commander.capacity_manager.rclcpp_action]: unknown result response, ignoring...
- In ROS2, you are able to see two nodes with the exact same namespace and name. Not sure if that’s a design choice
- Across multiple processes you can launch two nodes with the same fully-qualified name and namespace; the DDS discovery layer just sees two matching participants.
- This could be deadly when there’s one dead but not-cleaned node. You might still see its topics but they are dead. Is there a way to avoid that?
- If a node dies without calling
rcl_shutdown(e.g. hitting SIGKILL), its DDS participant is never disposed cleanly.- A “node” dies just means
A single rclcpp::Node / rclpy.node.Node object.- All publishers, subscriptions, services, actions, timers that belong only to that node are destroyed.
- The process dies when
rclcpp::shutdown()is called.
- A “node” dies just means
- Most DDS vendors keep endpoints in the graph until a liveliness timeout expires (default is 20 s – 1 min, vendor-specific). During that window:
ros2 node listandros2 topic liststill show the dead participant.
- If a node dies without calling
- In ROS2, you are able to see two nodes with the exact same namespace and name. Not sure if that’s a design choice
