Tomoya Fujita

@agoeckner rmw_cyclonedds also returns `TIMEOUT` when the same situation happens. so technically this problem could be observed too. https://github.com/ros2/rmw_cyclonedds/blob/76572ebf1032f359ed302d7017471a295738fa1e/rmw_cyclonedds_cpp/src/rmw_node.cpp#L4541-L4542 > Is there any word on a solution for FastRTPS? Currently...

I believe this is related to buildin discovery process in DDS implementation.

> We're having this issue occur when using Python clients. the error comes from rcl/rmw that means any client rclcpp, rclpy or other community packages relies on rcl possibly have...

@royito55 @agoeckner > i think both rclcpp and rclpy, this process is handled by executor thread once executable service is ready. that means probably we can protect spin with try...

@longjie0723 thanks for sharing the verification details. after all, this problem could happen eventually if there is racy condition to establish the connection between server and client with different topics....

> `Error in destruction of rcl action client handle: the Node Handle was destructed too early. You will leak memory` only explanation is, object pointed with `node_handle` is gone since...

> have you experienced this as well elsewhere? No, actually i was trying to reproduce the issue with minimal action client/server with some change that stresses server like million times....

@tgreier thanks for checking on this. how about the following error? are you still facing it often? > `Error in destruction of rcl action client handle: the Node Handle was...

@clalancette thank you very much for the explanation! > To do that, what we need to do is to add a reader/writer lock around the hashmap. yeah agree, that can...

just sharing my thought, > The implementation could live in rcl or rmw, I'm not sure what would be better. i believe that it is better to be implemented in...