Ben Frederickson

this could be useful, but will only let you get a single stack trace (like the output you get from passing --dump on the command line) instead of something like...

Fwiw, with python 3.11 we can get the OS thread id directly from python, and will be able to grab it from a dockerized process from the host container. We...

Can you try with the ```--idle``` option? py-spy samples 100x a second - but by default doesn't write out frames where the CPU isn't active. This is causing issue #3...

It does seem like we are getting tripped up with gevent's use of greenlets - and that is potentially causing the idle detection code to go awry =(. For your...

merlin-core and hugectr should both be fine for the backlog automation (I see https://github.com/NVIDIA-Merlin/core/issues/104 and https://github.com/NVIDIA-Merlin/HugeCTR/issues/341 in the backlog project).

> Sam's interpreter swaps out the normal Python virtual machine. Is py-spy looking for a specific active function to determine that it is looking at a running Python interpreter? The...

Yeah - it looks like it's picking up the libpython from the embedded interpreter, but the python version from the main executable =(. When you're looking at the logs -...

I haven't seen a segfault here before =( I'm guessing it's probably segfaulting in libunwind trying to get the native trace from your program but can't really be sure -...

Have you tried the subprocess support built in to py-spy? if you give the subprocesses argument to py-spy it should profile all the child processes

Can you try ``` py-spy record --output mpitestoptimizer.scope --format speedscope --subprocesses --r 25 -- mpirun -np 6 python optimize.py ``` instead? We want to get py-spy to profile the mpirun...