Add native traces to `RubySpy::get_stack_trace`

[Maaarcocr]

I've been needing to profile a ruby application that uses a variety of ruby extensions (written in C & rust) and also knowing how time is spent in the native extension a lot of the logic resides there.

I've taken inspiration from https://www.benfrederickson.com/profiling-native-python-extensions-with-py-spy/ and their source code to figure out how I could do it with rbspy. I have a prototype that works, and I have 2 questions:

When I get a StackTrace from ruby I also get a thread id, but it does not seem to match any thread id from remoteprocess::Process::threads in RubySpy. Is that expected? Is there a way to get a current thread ID that matches one from remoteprocess?

Would there be any interest in merging this into main here? Right now my changes live in my fork. I've made it such this capability is only turned on if a feature flag is set to true.

[acj]

Hey, thanks for working on this.

I have a prototype that works

🎉

When I get a StackTrace from ruby I also get a thread id, but it does not seem to match any thread id from remoteprocess::Process::threads in RubySpy. Is that expected? Is there a way to get a current thread ID that matches one from remoteprocess?

Hmm, I'm not sure. The thread ID in the StackTrace is read from rb_thread_struct, which might be an internal ID that the Ruby VM controls rather than the OS's thread ID that remoteprocess returns. Can you say more about the problem you're trying to solve? Also, which ruby version are you using to test this?

There's some structural similarity between this and the work in #327 -- might be worth a look.

Would there be any interest in merging this into main here?

Yes, it would be great to support this capability. Feel free to open a draft PR when you're ready, and then we can work out the details. Worst case, we can keep it behind the feature flag until it's working reliably.

[Maaarcocr]

Can you say more about the problem you're trying to solve? Also, which ruby version are you using to test this?

right so, currently (this is my understanding of rbspy, it may be wrong) rbspy would get the traces only for the current thread every time it samples the ruby process. Ideally I would like to do the same with native traces, i.e. only get the native traces for the thread which is marked as active by the RubyVM and just ignore all the other system threads.

If I do not do this, then it becomes difficult to know how to interweave the native trace and the ruby trace, as we could be in a situation where we have multiple native traces (one for each OS thread from remoteprocess) and just 1 ruby trace.

to make my prototype work I currently just assume that there is only 1 thread, but that would not work in the general case.

[Maaarcocr]

Feel free to open a draft PR when you're ready, and then we can work out the details

here you go: https://github.com/rbspy/rbspy/pull/373!

[acj]

Thanks! I needed to take a break during the holiday, but I should have time to review the PR this next week. Looking forward to it.