Install LTTng and related packages on Linux#690
Conversation
Signed-off-by: Christophe Bedard <christophe.bedard@apex.ai>
6077844 to
d721c3c
Compare
Signed-off-by: Christophe Bedard <christophe.bedard@apex.ai>
| python3-importlib-metadata \ | ||
| $(if test ${EL_RELEASE} = 8; then echo python3-importlib-resources; fi) \ | ||
| python3-lark-parser \ | ||
| $(if test ${EL_RELEASE} != 8; then echo python3-lttng; fi) \ |
There was a problem hiding this comment.
I think lttng-ust should work, but I'm not completely sure about python3-lttng.
However, looking at this, I think it should work: https://github.com/ros/rosdistro/blob/bbf74994da40a4905b25bd27fab44b22c5ec9b2c/rosdep/python.yaml#L7651
There was a problem hiding this comment.
RHEL 8 has python3-lttng, but the problem is that it's too old and tracetools breaks if it's present. I removed it in #657. Until tracetools handles the old version gracefully (which I'd still very much like to see), this is the best workaround.
There was a problem hiding this comment.
Old versions can definitely be handled better, but this only installs python3-lttng on RHEL 9.
clalancette
left a comment
There was a problem hiding this comment.
If we keep our view focused to Rolling, then these changes look good to me.
However, these Dockerfiles are also used for Humble (Ubuntu 22.04/RHEL-8) and Foxy (Ubuntu 20.04/RHEL-8). In both cases, we'll be adding packages that weren't there before. What is ros2_tracing going to do in those circumstances? I don't want to enable it in the stable distributions (particularly Foxy), for fear it might break something.
Oh, I missed that. That's indeed not great, because then I can do |
Yeah, that is probably a good idea. I'll also suggest putting this in a separate block in the Ubuntu Dockerfile, similar to ci/linux_docker_resources/Dockerfile Line 97 in 7cbb9f6 |
Signed-off-by: Christophe Bedard <christophe.bedard@apex.ai>
Done! |
clalancette
left a comment
There was a problem hiding this comment.
This looks good to me, though I'll want to see Linux CI for all of Foxy, Humble, and Rolling on both Ubuntu and RHEL. I'd also like to get one more approval from @cottsay .
Oh, actually, I guess I need to run it, since we can only do tests on local branches (not forks). I'll wait to hear from Scott before kicking that off. |
Signed-off-by: Christophe Bedard <christophe.bedard@apex.ai>
|
I thought we didn't need |
Signed-off-by: Christophe Bedard <christophe.bedard@apex.ai>
|
I also noticed that there was no rosdep rule for |
|
All right. I'm only going to run the latest change against Rolling again, given that it is protected under the same checks as the previous code. |
|
The test failures on RHEL don't seem to be related, but I can't find the same failures in recent jobs. |
That one is a flaky test; you can see a failure of it in https://ci.ros2.org/view/nightly/job/nightly_linux-rhel_repeated/1448/#showFailuresLink . I really don't think it is involved here at all. So I think we are good with this change. Once we merge this in, that means that we'll be building tracepoints in by default for Rolling in our nightly CI jobs and our nightly packaging (tarball) jobs. What needs to happen to have it enabled in the debians? Do we just need to do a release of |
Specifically after merging ros2/ros2_tracing#31 (which I think we should merge right after this), yes! LTTng will be picked up as a dependency (otherwise the build will fail), and the tracepoints will be compiled in. Well, that's what I think will happen, anyway. I can check the testing repo and validate once it's re-built. |
|
Next step before merging would be to run CI using this branch and ros2/ros2_tracing#31. See ros2/ros2_tracing#31 (comment) |
Awesome, the jobs there look good. So what is the next step here? Do you want to merge the ros2_tracing PR, then I can merge this one? |
Since packages will now depend on LTTng, and since CI doesn't seem to run |
OK, got it. So, the only thing is that I'm on PTO, not checking email, starting about now until Monday (April 3). Despite the fact that I think we've done sufficient testing here, I'm somewhat reluctant to merge something like this and disappear. So either we should have @cottsay merge it and keep an eye on it tomorrow, or I can do it first thing Monday morning when I'm back. @cottsay what do you think? |
|
Waiting until Monday as a precaution is fine with me. |
Part of ros2/ros2#1177
See ros2/ros2_tracing#31
Signed-off-by: Christophe Bedard christophe.bedard@apex.ai