stack traces in flamegraph appear without callers about py-spy HOT 5 CLOSED

I think that this might be an artifact of how I'm collecting the stack trace: I'm not pausing the python program, and the stack trace is gotten by traversing a linked list of frame objects from the target python processes memory. If the python program returns from a function call while this is happening, it could mess up this traversal and lead to results like you saw.

I've been thinking about adding an option to pause the python process while the stack is collected - which should prevent this (with the downside of potentially slowing the process down slightly).

from py-spy.

I think I've managed to replicate this - it seems to mainly happen with deep-ish callstacks, where the python functions aren't taking any significant time.

It'd be great to figure out a way to discard these callstacks so we can keep the profiler from blocking the python program - but I can't see any way of doing it right now (just with a higher error rate during sampling). The problem is that the end of the a legitimate call stack is indicated by a null pointer on the 'next' frame - which is the same thing that is happening here. Going to see if pausing the python program with ptrace etc resolves this

from py-spy.

This commit seems to fix: e25656c

from py-spy.

fixed in v0.1.9

from py-spy.

Sweet, thanks. Updated and my use-case seems to be working fine now.

Also, the slow-down is just 3% in the case I tried.

Tested on macos 10.13

from py-spy.