cpu stuck in memory profiling

[alangenfeld]

Webserver application (uvicorn & starlette) stuck at 100% CPU in
ddtrace/profiling/collector/memalloc.py:139 / heap_tracker_untrack_thawed (_memalloc_heap.c:80)

py-spy top
https://gist.github.com/alangenfeld/4dd201717208f7bde63ff26552d80096

py-spy top -n
https://gist.github.com/alangenfeld/fb51310ccf085f67fe717ebb2df9dee8

Which version of dd-trace-py are you using?

observed in 0.57.2 and 0.57.1

Which version of pip are you using?

pip 21.0.1

Which version of the libraries are you using?

https://gist.github.com/alangenfeld/73336333915dd9e7e2b0a1592e38e24e

How can we reproduce your problem?

Unclear - observed in one of our webservers but not the other

What is the result that you get?

CPU pegged at 100% at heap_tracker_untrack_thawed (_memalloc_heap.c:80)

What is the result that you expected?

Profiling with reasonable overhead

[alangenfeld]

looks likely fixed by #3162

[jd]

@alangenfeld if you can give it a try, that'd be appreciated!

[alangenfeld]

Is there going to be a release soon or would I need to try from master?

[Flydiverny]

We are also seeing this in 0.56.0, probably after enabling DD_PROFILING_ENABLED, so we are attempting to disable this to see if this helps. :)

[jd]

The fixes have been released in 0.57.3 (and will be in 0.58 too, for the record). You can upgrade and see if that fixes everything for you. 🙏

If not, please reopen this!