Process.children() is slow

[pitrou]

>>> import psutil
>>> p = psutil.Process()
>>> p.children()
[]
>>> %timeit p.children()
36.1 ms ± 268 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

It seems half of that is calling process_iter():

>>> %timeit l = list(psutil.process_iter())
20.5 ms ± 232 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

[giampaolo]

Hey Antoine! =)
Process.children() is inevitably slow because it has to obtain pid and ppid for all processes:

psutil/psutil/__init__.py

Lines 862 to 864 in 3a3598e

	for p in process_iter():
	try:
	if p.ppid() == self.pid:

We can't get away from that.

With that said, it is true that we're doing more than just that because we also call create_time(), even more than once (here and here).

We can probably avoid that and define a utility function similar to what we have on windows:

psutil/psutil/__init__.py

Lines 851 to 854 in 3a3598e

	if hasattr(_psplatform, 'ppid_map'):
	# Windows only: obtain a {pid:ppid, ...} dict for all running
	# processes in one shot (faster).
	ppid_map = _psplatform.ppid_map()

...which just returns a {pid; ppid, ...} map for all processes in one shot.

[pitrou]

Does that function have to be consistent? The problem is if the processes change while iterating on e.g. /proc.

[giampaolo]

A process which "changes" while iterating means it will disappear, in which case we'll just skip it.
They will also be skipped in case of AccessDenied (doc doesn't mention this - I will file a ticket for it).

[pitrou]

I see, so we can just scan quickly over /proc?

[giampaolo]

Yes. Same goes for all other platforms except Windows which already does this in C.

[giampaolo]

Addresed in #1185 and #1186. Closing.