is_running() should be called before kill()

[giampaolo]

From billiej...@gmail.com on July 13, 2009 17:01:25

I think it makes sense to call is_running() before kill() as a safety
measure since the original process could be gone in meantime and the same
PID assigned to a different process.

Original issue: http://code.google.com/p/psutil/issues/detail?id=58

[giampaolo]

From jlo...@gmail.com on July 13, 2009 18:17:11

That doesn't really save us from anything, since the process could disappear between
calling is_running() and calling kill() anyway. A race condition is a race condition,
even if you do it really fast ;)

Also, bear in mind that on Windows, we're checking if the PID is running in the C
code before calling TerminateProcess() and again when we open the process handle. On
UNIX platforms we use the os.kill() function with a try/except watching for ESRCH
being thrown in the event the process does not exist. 

I think we're covered just fine on this front without adding additional code.

[giampaolo]

From billiej...@gmail.com on July 14, 2009 00:52:09

I didn't mean to do this to prevent race conditions, also because NoSuchProcess is
already raised in case the process disappears on us, and there's also a test case
(test_zombie_process) showing how that already works.
I wanted to do this for checking the process identity to make sure that we aren't
going to kill another process we didn't mean to.

is_running() can help us on doing this as it creates a new Process object for the pid
and compares the "hash" for it (cmdline, path, etc...) with the hash of the current
object before executing the action (kill()).

Hope you can see my point now.

[giampaolo]

From jlo...@gmail.com on July 14, 2009 06:33:03

There's still a race condition, since such a scenario could happen in between calling
is_running() and kill() on the process. It's unlikely since the OS usually doesn't
reuse PIDs that quickly, at least as far as I know that's the case for the OSes we
support so far, but it's still theoretically possible. 

However, I can see this being at least more accurate than a use case where someone
creates a Process object and then an hour later issues kill() against it. It could
potentially be a performance issue at some point but we can worry about that later if
it comes up I guess.

[giampaolo]

From billiej...@gmail.com on July 14, 2009 07:08:44

Calling is_running() before kill() surely requires a considerable additional amount
of time compared to the current implementation, in fact I was thinking that we could
also speed up the __eq__ implementation somehow.

Anyway, I don't see kill() as a kind function that is called so often to represent an
actual performance problem, imho.

[giampaolo]

From billiej...@gmail.com on July 14, 2009 07:51:24

Fixed as r415 .

Status: Fixed
Labels: -Progress-0in4

[giampaolo]

From billiej...@gmail.com on September 03, 2009 12:48:09

Status: FixedInSVN

[giampaolo]

From billiej...@gmail.com on September 17, 2009 01:57:42

Status: Fixed

[giampaolo]

From g.rodola on March 02, 2013 03:49:55

Updated csets after the SVN -> Mercurial migration: r415 == revision f6713a409197