Is there a way to measure time spent in GC?

[sarahec]

I'm benchmarking and performance tuning some ML code, and garbage collection seems to be the largest drag on performance. Is there a way to measure time spent in GC?

[emeryberger]

Not within Scalene. My intuition for GC in Python is that it's not usually a problem because most of the time I'd expect reference counting to handle most of the garbage (except for cycles, which I wouldn't expect in a case where you are using an ML library). I'd be interested to see what's going on! Anyway, looks like py-spy might be helpful here: see benfred/py-spy#389.

[sarahec]

Thank you :) I'll keep looking.

[emeryberger]

Might try this, though it's not exactly the nicest UI:

https://stackoverflow.com/a/6115258/335756