Add tracing instrumentation using tracetools

[christophebedard]

With this instrumentation, rmw_zenoh_cpp is supported out-of-the-box by tools that process ROS 2 trace data, such as Eclipse Trace Compass.

It's pretty straightforward except for a few things:

1. Getting the rmw_publisher_t * for the rmw_publish tracepoint and rmw_service_t * for the rmw_send_response tracepoint
   • Add them to PublisherData and ServiceData, respectively.
2. Getting the source timestamp for the rmw_publish and rmw_send_response tracepoints
   • Getting the source timestamp of a message being published is needed for tracking messages from publisher to subscription or replies from service to client.
   • Since Add timestamp to rmw_publish tracepoint ros2_tracing#74, we're getting the source timestamp for a published message from rmw. This means we don't need to instrument the underlying middleware itself, which we used to have to do.
   • Remove create_map_and_set_sequence_num(), since it's already pretty simple and it makes extracting the source timestamp easier.
3. Getting the/a subscription rmw GID for the rmw_subscription_init tracepoint
   • See the comment before the tracepoint.

I'm open to suggestions.

---

action-ros-ci-repos-override: https://gist.githubusercontent.com/christophebedard/18561011c03603652ad49701412e5f41/raw/9e9d0bc9bbf0c36b7d5419bc5797ef17aa90d6fa/ros2.repos

[christophebedard]

Oh, looks like this repo aims to support Iron, Jazzy, and Rolling on the rolling branch. I will deal with that after a first round of review.

[christophebedard]

@Yadunund @clalancette any interest in having this?

[Yadunund]

@christophebedard we'd love to have this! We're sort of freezing the rolling branch at the moment to focus effortts on getting the #276 merged before we add new features. So let's wait for that to get in before merging this PR . By then Iron would have ben sunsetted so we won't have to worry about failing Iron CI job.

[Yadunund]

@christophebedard do you mind resolving the conflicts after rebasing to the latest rolling branch?

[christophebedard]

I rebased, but now the tracing tests in test_tracetools sometimes time out. It seems to happen during rclcpp::shutdown(), which might be related to #329.

[christophebedard]

I rebased & resolved the new merge conflicts.

I also added client/service instrumentation, see ros2/ros2_tracing#145.

I rebased, but now the tracing tests in test_tracetools sometimes time out. It seems to happen during rclcpp::shutdown(), which might be related to #329.

I ran the tests quite a few times and didn't get any failures. Not sure what changed in the code, other than the instrumentation.

[clalancette]

I've left 3 more minor things to fix, then I'm happy with it.

[clalancette]

One more fix, then I'm happy with this.

[clalancette]

This looks great. Thanks for all of the iteration, and for being patient with us as we revamped other things. I'm going to go ahead and merge now.

[cybaol]

@christophebedard TRACETOOLS_TRACEPOINT seems break jazzy build (replicated on archlinux).

[christophebedard]

Make sure to build the tracetools package (from the ros2/ros2_tracing repo) from source. If you rely on the binaries, rmw_zenoh will indeed not build.

[christophebedard]

Oh, I guess this does break Jazzy builds. This will only work with Rolling (built from source for now).

The new service/client tracepoints would need to be excluded on Jazzy.

[christophebedard]

So CI passed for this PR because the build_and_test_binary_jazzy job wasn't actually testing Jazzy: https://github.com/ros2/rmw_zenoh/actions/runs/12422154058/job/34683299857. It's my fault, I set action-ros-ci-repos-override: to use the ros2_tracing branch that enables tracing tests for rmw_zenoh_cpp (ros2/ros2_tracing#140). The same job is indeed now failing on rolling: https://github.com/ros2/rmw_zenoh/actions/runs/12432407339/job/34711738348.

Since this repo doesn't use distro-specific branches yet, could we add distro-specific #defines`? Since I broke the build, I've opened a PR with a quick-and-dirty solution: #355.