Certainly sounds reasonable. I can work something up this week most likely, unless you're wanting to do this yourself and opening this issue just for tracking.

I'm not sure I'll have bandwidth soon (and Python is not a language I'm super familiar with) so I'd certainly appreciate it if you made the change. Thanks!

@samuelkarp I just pushed #3926 for this, I think it covers the request but please take a look when you have a moment and let us know if there is anything missing.

I think that looks like it would work. To invoke a user would need to pass --plugin-option containerd.stackdump? The only other bit is if we also wanted to have stack dumps from the shims (containerd-shim-runc-v2) which would just be written to the journal, but that seems a more rare use-case.

To invoke a user would need to pass --plugin-option containerd.stackdump?

Yup.

The only other bit is if we also wanted to have stack dumps from the shims (containerd-shim-runc-v2)

That was actually what prompted me to make the changes to get_process_pids(), so we weren't signalling those processes 🙃

That was actually what prompted me to make the changes to get_process_pids(), so we weren't signalling those processes 🙃

That's still probably valuable; getting a stack dump from each shim would be fairly noisy and only useful in rare cases. Maybe if we put that behind a second option that might be better than having everything behind containerd.stackdump?

@TurboTurtle Thank you!