How to shutdown and reinitialize a publisher node in ROS 2

Next Generation ROS
1

I have a use case of shutting down and reinitializing the publisher in ROS 2,

i have included my init function inside the while loop when i execute the publisher i get this error

terminate called after throwing an instance of 'rclcpp::exceptions::RCLError'
  what():  failed to create interrupt guard condition: rcl_init() has not been called, at /home/administrator/ros2_ws/src/ros2/rcl/rcl/src/rcl/guard_condition.c:61
Aborted (core dumped)

my code:

int main(int argc, char * argv[] )
{
auto node = rclcpp::Node::make_shared("talker1");
auto Pub = node->create_publisher<std_msgs::msg::String>("chatter1",rmw_qos_profile_default);
rclcpp::WallRate loop_rate(1);
size_t count = 1;
auto msg = std::make_shared<std_msgs::msg::String>();

while(1)
    {
    	   rclcpp::init(argc, argv);
		

		msg->data = "Hello World" + to_string(count++);
            
    		std::cout << msg->data << ":msg_no:" << count << std::endl;
    		Pub->publish(msg); 
loop_rate.sleep();
	
    }

  rclcpp::shutdown();
  return 0;
}

Can anyone help me with this?

1 Like
2

I reformatted you code to be easier to read and added some missing includes that prevent it from compiling:

#include <iostream>
#include <memory>
#include <string>

#include "rclcpp/rclcpp.hpp"

#include "std_msgs/msg/string.hpp"

using namespace std::chrono_literals;

int main(int argc, char * argv[] )
{
  auto node = rclcpp::Node::make_shared("talker1");
  auto Pub = node->create_publisher<std_msgs::msg::String>("chatter1", rmw_qos_profile_default);
  rclcpp::WallRate loop_rate(1);
  size_t count = 1;
  auto msg = std::make_shared<std_msgs::msg::String>();

  while(1) {
    rclcpp::init(argc, argv);

    msg->data = "Hello World" + std::to_string(count++);

    std::cout << msg->data << ":msg_no:" << count << std::endl;
    Pub->publish(msg); 
    loop_rate.sleep();
  }

  rclcpp::shutdown();
  return 0;
}

The first issue I see (and the reason for your error) is that you are trying to create a node before rclcpp::init() has been called, which is implied in the error message though it could be more specific to rclcpp that it currently is:

what(): failed to create interrupt guard condition: rcl_init() has not been called, at …

But even if you move the node creation into the while loop, and after the rclcpp::init(), the next issue I see is that you are actually calling init many times, while only calling shutdown once. This will definitely be an issue, and I modified your example to see what would happen :slight_smile:

#include <iostream>
#include <memory>
#include <string>

#include "rclcpp/rclcpp.hpp"

#include "std_msgs/msg/string.hpp"

using namespace std::chrono_literals;

int main(int argc, char * argv[] )
{
  rclcpp::WallRate loop_rate(1);
  size_t count = 1;
  auto msg = std::make_shared<std_msgs::msg::String>();

  while(1) {
    rclcpp::init(argc, argv);

    auto node = rclcpp::Node::make_shared("talker1");
    auto Pub = node->create_publisher<std_msgs::msg::String>("chatter1", rmw_qos_profile_default);

    msg->data = "Hello World" + std::to_string(count++);

    std::cout << msg->data << ":msg_no:" << count << std::endl;
    Pub->publish(msg); 
    loop_rate.sleep();
  }

  rclcpp::shutdown();
  return 0;
}

When I run it I get:

% /Users/william/ros2_ws/install_release/lib/demo_nodes_cpp/talker
Hello World1:msg_no:2
libc++abi.dylib: terminating with uncaught exception of type std::runtime_error: failed to initialize rmw implementation: rcl_init called while already initialized, at /Users/william/ros2_ws/src/ros2/rcl/rcl/src/rcl/rcl.c:68

The important thing being:

rcl_init called while already initialized

Which again could be more rclcpp explicit, but should also be enough of an indication as to what the issue is.

Finally, however, I’d like to ask why you think you need to init and shutdown repeatedly? It’s potentially a valid use case, but I get from your example why that’s necessary.

Assuming for a second that’s actually what you want to do, I modified your example one more time to move the shutdown into the while loop as well:

#include <iostream>
#include <memory>
#include <string>

#include "rclcpp/rclcpp.hpp"

#include "std_msgs/msg/string.hpp"

using namespace std::chrono_literals;

int main(int argc, char * argv[] )
{
  rclcpp::WallRate loop_rate(1);
  size_t count = 1;
  auto msg = std::make_shared<std_msgs::msg::String>();

  while(1) {
    rclcpp::init(argc, argv);

    auto node = rclcpp::Node::make_shared("talker1");
    auto Pub = node->create_publisher<std_msgs::msg::String>("chatter1", rmw_qos_profile_default);

    msg->data = "Hello World" + std::to_string(count++);

    std::cout << msg->data << ":msg_no:" << count << std::endl;
    Pub->publish(msg); 
    loop_rate.sleep();

    rclcpp::shutdown();
  }

  return 0;
}

But, that results in this still:

terminating with uncaught exception of type std::runtime_error: failed to initialize rmw implementation: rcl_init called while already initialized

Unfortunately doing init-shutdown-init does not currently work in rclcpp and is a known issue, see:

This does work one layer down in rcl (which is used by both rclcpp and rclpy), but specifically in rclcpp, there was some disagreement about how to deterministically shutdown in a multi-threaded environment, and so it was not resolved yet. It can be and should be but it’s not the case at the moment.

So if you truly do need to init and shutdown repeatedly in the same process we’ll have to first fix rclcpp to support that case, which would be a great contribution if you or anyone else has time to help with it :smile:.

1 Like
3

Just for fun I did this last iteration in Python a well, which works:

import sys

import rclpy
from rclpy.node import Node

from std_msgs.msg import String


def main():
    msg = String()
    i = 0

    while True:
        rclpy.init()

        node = Node('talker')
        pub = node.create_publisher(String, 'chatter')
        msg.data = 'Hello World: {0}'.format(i)
        i += 1
        node.get_logger().info('Publishing: "{0}"'.format(msg.data))
        pub.publish(msg)

        node.destroy_node()
        rclpy.shutdown()


if __name__ == '__main__':
    main()
4

Funny you should mention this python alternative. For a potential use case, I was just using this style of iterative initialisation and shutdown within a single thread to do sros2 access control verification this week. Interesting to know that had I used rclcpp to benchmark my policies rather than rclpy I would have hit this current issue. I also found using the try_shutdown method more forgiving to use should you ever lose track of the state of things like when developing from in an interactive python shell session.