Can ros2_tracing trace rclpy?

[Binohow]

Hello, I want to know if ros2_tracing support to trace rclpy? Lttng tools is too weak for python.

[christophebedard]

Hi! Unfortunately, rclpy is not instrumented; you can only trace rclcpp.

Lttng tools is too weak for python.

What do you mean?

[christophebedard]

Since this has been requested many times, I've opened an issue as a feature request: #15.

[Binohow]

Hi! Unfortunately, rclpy is not instrumented; you can only trace rclcpp.

Lttng tools is too weak for python.

What do you mean?

Thanks for your answer! Can we use Lttng tracer to implement this feature? I am very interested in how to implement this function. I want to participate in this meaning work, but I don't where to start. Which tracer should I choose first?

Hello, I want to know if ros2_tracing support to trace rclpy? Lttng tools is too weak for python.

[christophebedard]

LTTng can trace Python applications: https://lttng.org/docs/v2.13/#doc-python-application

[vtran5]

Hello, maybe a related question, can ros2_tracing support rclc? If not, is there any guide on how to implement it myself? From the design document, I understand that I would need to manually add the tracepoint in rclc following the 6-step guide in the Adding instrumentation. Is there anything else that I need to do or considered? Do I need to add all the equivalent TRACEPOINT in rclcpp to rclc in order for it to work properly?

[christophebedard]

It could, but it would require a bit of work. The proper way would be to design a new set of instrumentation points for rclc, since the rclcpp instrumentation might not necessarily all apply to rclc (and vice versa). Once the instrumentation is defined, you can indeed go through the Adding instrumentation section of the ros2_tracing design document to actually create the tracepoints.

Like I said, that would be the "proper" way. Also, I'm not sure if those tracepoints should be here in ros2_tracing/tracetools, since rclc is more of a separate client library.

However, you could try to just re-use the existing rclcpp tracepoints by adding them to rclc as if it was rclcpp. I did it a couple years ago with the main rclcpp tracepoints, so it might need to be updated, but it can be a good starting point: christophebedard/rclc@090a0cb. Someone also did that for rclpy: #15 (comment). This is more of a hack, though, as you can see from my comments in that rclc commit. You would probably need to process trace data coming from rclc differently compared to trace data coming from rclcpp.

[christophebedard]

Since this issue was mostly just a question, and since I've opened a standalone issue for the actual feature request (#15), I'll close this.