Allow enabling syscalls through ros2 trace or the Trace action

[christophebedard]

This adds support for enabling syscall tracing with the ros2 trace command or the Trace launch action.

While system calls are separate from kernel events, both can be traced using the LTTng kernel tracer: https://lttng.org/docs/v2.13/#doc-tracing-the-linux-kernel. Kernel events are "tracepoints" while system calls are their own kind of trace event. This explains why this PR adds a separate option for syscalls instead of just using the kernel events option. See the changes to lttngpy::enable_events in lttngpy/src/lttngpy/event.cpp.

How to use:

1. ros2 trace command:

   $ ros2 trace --syscall openat  # interactive
   $ ros2 trace start my-session-name --syscall openat  # non-interactive
   

2. Launch files (Python, XML, and YAML): syscalls arg

Note that, when enabling a syscall, e.g., openat, the resulting trace has syscall entry and exit events: syscall_entry_openat and syscall_exit_openat.

[christophebedard]

@mjcarroll would you be willing to review this?

[christophebedard]

There's something wrong with the GitHub CI binary jobs: https://github.com/ros2/ros2_tracing/actions/runs/11205771796/job/31145663869?pr=137#step:4:1989. I see the same failures in the nightly GitHub CI binary jobs. Not really related to these changes.

[fujitatomoya]

overall, lgtm with green CI. had a couple of comments.

[christophebedard]

• Linux
• Linux-aarch64
• Linux-rhel
• Windows

[christophebedard]

Sorry, I realized that the way I had written the decorator meant it was evaluated during test collection, at the very beginning. This means that, if no session daemon was running before the test was run, then the test would not be allowed to try to spawn a session daemon and therefore the test would definitely be skipped. I modified the decorator so that it's evaluated before running the test: https://github.com/ros2/ros2_tracing/compare/485dca3ac871e2a26d603b2916ac3aa6281ebf28..e12b8044bfd1b8a15209df44c10e0f1e47bcf5b0.

Can you take another look?

[christophebedard]

Argh this isn't even completely working.

[christophebedard]

OK, I fixed the decorator and I think this is all I can really do. Implementing the decorator this way ensures it's evaluated right before running the test, after the test class' @unittest.skipIf has been evaluated. Otherwise, running on Windows fails, since lttngpy is empty. See the previous Windows CI job: Windows .

Also, when running with colcon test on my system, test_kernel_tracing runs, because I have a root session daemon running and my user is part of the tracing group. However, if you try to run the test using a build system that does sandboxing, it fails because there's no root session daemon in it, and spawning one would be more tricky. I think this is fine for now.

New CI:

• Linux
• Linux-aarch64
• Linux-rhel
• Windows

[fujitatomoya]

lgtm, thanks for the explanation. lgtm with green CI.

[christophebedard]

Thanks for the review @fujitatomoya!

CI looks good.