subprocess leaves child living

[Thomas Dybdahl Ahle]

Hi, When I do a small program like

from subprocess import Popen
popen = Popen(["ping", "google.com"])
from time import sleep
sleep(100)

start it and kill it, the ping process lives on.
Is there a way to ensure that the ping process is always killed when the
python process is?
I can't use atexit, as ping then isn't killed when python is killed "in
the hard way"

[Stefan Sonnenberg-Carstens]

Thomas Dybdahl Ahle schrieb:

Hi, When I do a small program like

from subprocess import Popen
popen = Popen(["ping", "google.com"])
from time import sleep
sleep(100)

start it and kill it, the ping process lives on.
Is there a way to ensure that the ping process is always killed when the
python process is?
I can't use atexit, as ping then isn't killed when python is killed "in
the hard way"

Calling popen.close() perhaps ?
You basically open a pipe, which spawns a shell and the command is then
started in there.
So, if your program quits, the spawned shell is still alive, only the
pipe is dead.

[Thomas Dybdahl Ahle]

Den Tue, 05 Jun 2007 14:07:44 +0200 skrev Stefan Sonnenberg-Carstens:

Thomas Dybdahl Ahle schrieb:

>Hi, When I do a small program like

from subprocess import Popen
popen = Popen(["ping", "google.com"]) from time import sleep
sleep(100)

start it and kill it, the ping process lives on. Is there a way to
ensure that the ping process is always killed when the python process
is?
I can't use atexit, as ping then isn't killed when python is killed "in
the hard way"

Calling popen.close() perhaps ?
You basically open a pipe, which spawns a shell and the command is then
started in there.
So, if your program quits, the spawned shell is still alive, only the
pipe is dead.

Problem is - I can't do that when I get killed.
Isn't it possible to open processes in such a way like terminals? If I
kill the terminal, everything open in it will die too.

[Rob Wolfe]

Thomas Dybdahl Ahle wrote:

Problem is - I can't do that when I get killed.
Isn't it possible to open processes in such a way like terminals? If I
kill the terminal, everything open in it will die too.

On POSIX platform you can use signals and ``os.kill`` function.
Fo example:

<code>
import os, signal
from subprocess import Popen
from time import sleep

def handler(signum, frame):
print 'Signal handler called'
raise KeyboardInterrupt

signal.signal(signal.SIGTERM, handler)

try:
popen = Popen(["ping", "google.com"])
try:
sleep(100)
except KeyboardInterrupt:
pass
finally:
if popen.poll() is None:
print "killing process: %d" % popen.pid
os.kill(popen.pid, signal.SIGTERM)
</code>

--
HTH,
Rob

[Thomas Dybdahl Ahle]

Den Tue, 05 Jun 2007 07:06:15 -0700 skrev Rob Wolfe:

Thomas Dybdahl Ahle wrote:

>Problem is - I can't do that when I get killed. Isn't it possible to
open processes in such a way like terminals? If I kill the terminal,
everything open in it will die too.

On POSIX platform you can use signals and ``os.kill`` function. Fo
example:

<code>
import os, signal
from subprocess import Popen
from time import sleep

def handler(signum, frame):
print 'Signal handler called'
raise KeyboardInterrupt

signal.signal(signal.SIGTERM, handler)

try:
popen = Popen(["ping", "google.com"]) try:
sleep(100)
except KeyboardInterrupt:
pass
finally:
if popen.poll() is None:
print "killing process: %d" % popen.pid os.kill(popen.pid,
signal.SIGTERM)
</code>

But you can't ever catch sigkill.
Isn't there a way to make sure the os kills the childprocess when the
parrent dies?

[Rob Wolfe]

Thomas Dybdahl Ahle <lo****@gmail.comwrites:

But you can't ever catch sigkill.

There is no protection against sigkill.

Isn't there a way to make sure the os kills the childprocess when the
parrent dies?

If the parent dies suddenly without any notification childprocesses
become zombies and there is no way to avoid that.

--
HTH,
Rob

[Michael Bentley]

On Jun 5, 2007, at 3:01 PM, Rob Wolfe wrote:

Thomas Dybdahl Ahle <lo****@gmail.comwrites:

>But you can't ever catch sigkill.

There is no protection against sigkill.

>Isn't there a way to make sure the os kills the childprocess when the
parrent dies?

If the parent dies suddenly without any notification childprocesses
become zombies and there is no way to avoid that.

Apologies for picking nits...

But actually *that* is an orphan process. When a parent process dies
and the child continues to run, the child becomes an orphan and is
adopted by init. Orphan processes can be cleaned up on most Unices
with 'init q' (or something very similar).

Zombies on the other hand, are those processes that have completed
execution but still have an entry in the process table. The cause of
zombies AFAIK, is a parent that has failed to call wait(2). To clean
up zombies, you can send a SIGCHLD signal to the parent process --
probably with 'kill -17' (but use 'kill -l' to find out what it is on
your system).

hth,
Michael
---
"I would rather use Java than Perl. And I'd rather be eaten by a
crocodile than use Java." — Trouser

[Thomas Dybdahl Ahle]

Den Tue, 05 Jun 2007 22:01:44 +0200 skrev Rob Wolfe:

Thomas Dybdahl Ahle <lo****@gmail.comwrites:

>But you can't ever catch sigkill.

There is no protection against sigkill.

>Isn't there a way to make sure the os kills the childprocess when the
parrent dies?

If the parent dies suddenly without any notification childprocesses
become zombies and there is no way to avoid that.

If I call "kill -9 pythonpid" on the python code you posted earlier, the
terminal running the script will continue pinging forever.
If it was a harder program like a chessengine or such, it would continue
consuming 100% cpu time.
Zombies would be fine to me.

[Thomas Dybdahl Ahle]

Den Tue, 05 Jun 2007 15:46:39 -0500 skrev Michael Bentley:

But actually *that* is an orphan process. When a parent process dies
and the child continues to run, the child becomes an orphan and is
adopted by init. Orphan processes can be cleaned up on most Unices with
'init q' (or something very similar).

Is it not possible to tell python that this process should not be adopted
by init, but die with its parrent?
Just like terminals seem to do it..

[Michael Bentley]

On Jun 5, 2007, at 4:17 PM, Thomas Dybdahl Ahle wrote:

Den Tue, 05 Jun 2007 15:46:39 -0500 skrev Michael Bentley:

>But actually *that* is an orphan process. When a parent process dies
and the child continues to run, the child becomes an orphan and is
adopted by init. Orphan processes can be cleaned up on most
Unices with
'init q' (or something very similar).

Is it not possible to tell python that this process should not be
adopted
by init, but die with its parrent?
Just like terminals seem to do it..

Well, the way you posed the original question:

from subprocess import Popen
popen = Popen(["ping", "google.com"])
from time import sleep
sleep(100)

is really what adoption by init is designed to handle. Here you've
created a child but have not waited for its return value. Like a
good adoptive parent, init will wait(2) for the child.

I think if you really looked into it you'd find that the terminal had
called wait(2) before it was killed. Similarly, if you start a long-
running subprocess in python and wait for it to return -- killing the
parent will slaughter the child as well.

hth,
Michael
---
Let the wookie win.

[Michael Bentley]

On Jun 5, 2007, at 5:13 PM, Michael Bentley wrote:

>
On Jun 5, 2007, at 4:17 PM, Thomas Dybdahl Ahle wrote:

>Den Tue, 05 Jun 2007 15:46:39 -0500 skrev Michael Bentley:

>>But actually *that* is an orphan process. When a parent process
dies
and the child continues to run, the child becomes an orphan and is
adopted by init. Orphan processes can be cleaned up on most
Unices with
'init q' (or something very similar).

Is it not possible to tell python that this process should not be
adopted
by init, but die with its parrent?
Just like terminals seem to do it..

Well, the way you posed the original question:

>from subprocess import Popen
popen = Popen(["ping", "google.com"])
from time import sleep
sleep(100)

is really what adoption by init is designed to handle. Here you've
created a child but have not waited for its return value. Like a
good adoptive parent, init will wait(2) for the child.

I think if you really looked into it you'd find that the terminal had
called wait(2) before it was killed. Similarly, if you start a long-
running subprocess in python and wait for it to return -- killing the
parent will slaughter the child as well.

I guess I should have verified my suspicions before speaking up -- I
was worng -- even if you are waiting for a return code, the child
will persist as a child of init.

regards,
Michael
---
The Rules of Optimization are simple.
Rule 1: Don't do it.
Rule 2 (for experts only): Don't do it yet.
-Michael A. Jackson

[Nick Craig-Wood]

Thomas Dybdahl Ahle <lo****@gmail.comwrote:

But you can't ever catch sigkill.
Isn't there a way to make sure the os kills the childprocess when the
parrent dies?

Not as far as I know.

If you've got a pipe open to the child then killing the parent should
deliver SIGPIPE to the child which may (or may not) kill it. At least
it got some sort of notification.

--
Nick Craig-Wood <ni**@craig-wood.com-- http://www.craig-wood.com/nick

[Thomas Dybdahl Ahle]

Den Tue, 05 Jun 2007 17:41:47 -0500 skrev Michael Bentley:

On Jun 5, 2007, at 5:13 PM, Michael Bentley wrote:

>On Jun 5, 2007, at 4:17 PM, Thomas Dybdahl Ahle wrote:

>>Den Tue, 05 Jun 2007 15:46:39 -0500 skrev Michael Bentley:

But actually *that* is an orphan process. When a parent process dies
and the child continues to run, the child becomes an orphan and is
adopted by init. Orphan processes can be cleaned up on most Unices
with
'init q' (or something very similar).

Is it not possible to tell python that this process should not be
adopted
by init, but die with its parrent?
Just like terminals seem to do it..

Well, the way you posed the original question:

>>from subprocess import Popen
popen = Popen(["ping", "google.com"]) from time import sleep
sleep(100)

is really what adoption by init is designed to handle. Here you've
created a child but have not waited for its return value. Like a good
adoptive parent, init will wait(2) for the child.

I think if you really looked into it you'd find that the terminal had
called wait(2) before it was killed. Similarly, if you start a long-
running subprocess in python and wait for it to return -- killing the
parent will slaughter the child as well.

I guess I should have verified my suspicions before speaking up -- I was
worng -- even if you are waiting for a return code, the child will
persist as a child of init.

Bugger.
I know that if you use forkpty, the child get slaughtered, but using the
subprocess module just seems much easier than forking.

[Michael Bentley]

On Jun 6, 2007, at 7:11 AM, Thomas Dybdahl Ahle wrote:

Den Tue, 05 Jun 2007 17:41:47 -0500 skrev Michael Bentley:

>On Jun 5, 2007, at 5:13 PM, Michael Bentley wrote:

>>On Jun 5, 2007, at 4:17 PM, Thomas Dybdahl Ahle wrote:

Den Tue, 05 Jun 2007 15:46:39 -0500 skrev Michael Bentley:

But actually *that* is an orphan process. When a parent
process dies
and the child continues to run, the child becomes an orphan and is
adopted by init. Orphan processes can be cleaned up on most
Unices
with
'init q' (or something very similar).

Is it not possible to tell python that this process should not be
adopted
by init, but die with its parrent?
Just like terminals seem to do it..

Well, the way you posed the original question:

from subprocess import Popen
popen = Popen(["ping", "google.com"]) from time import sleep
sleep(100)

is really what adoption by init is designed to handle. Here you've
created a child but have not waited for its return value. Like a
good
adoptive parent, init will wait(2) for the child.

I think if you really looked into it you'd find that the terminal
had
called wait(2) before it was killed. Similarly, if you start a
long-
running subprocess in python and wait for it to return -- killing
the
parent will slaughter the child as well.

I guess I should have verified my suspicions before speaking up --
I was
worng -- even if you are waiting for a return code, the child will
persist as a child of init.

Bugger.
I know that if you use forkpty, the child get slaughtered, but
using the
subprocess module just seems much easier than forking.

Yeah, and if there is some way (there may be, but I don't know) to
make your python script the head of a process group (or session --
I'm not very clear on the distinction) -- children would be
automatically slain. That's how a terminal session manages to get
its children cleaned up I think...

If it is important that children get terminated upon death of a
parent, you probably do have to fork.

g'luck!
Michael
---
"Neurons are far more valuable than disk space, screen lines, or CPU
cycles." - Ben Finney

[reed]

On Jun 5, 7:58 am, Thomas Dybdahl Ahle <[e-mail]:

Hi, When I do a small program like

from subprocess import Popen
popen = Popen(["ping", "google.com"])
from time import sleep
sleep(100)

start it and kill it, the ping process lives on.
Is there a way to ensure that the ping process is always killed when the
python process is?
I can't use atexit, as ping then isn't killed when python is killed "in
the hard way"

pid = popen.pid
pidfile = open('/usr/local/var/somefile.pid', 'w')
pidfile.write('pid')
pidfile.close()
then you can check if it is still running when your ?program? restarts
and can kill it.

maybe not the perfect answer, but it answers an imperfect question.

[Thomas Dybdahl Ahle]

Den Thu, 07 Jun 2007 07:00:53 +0000 skrev reed:

On Jun 5, 7:58 am, Thomas Dybdahl Ahle <[e-mail]:

>Hi, When I do a small program like

from subprocess import Popen
popen = Popen(["ping", "google.com"]) from time import sleep
sleep(100)

start it and kill it, the ping process lives on. Is there a way to
ensure that the ping process is always killed when the python process
is?
I can't use atexit, as ping then isn't killed when python is killed "in
the hard way"

pid = popen.pid
pidfile = open('/usr/local/var/somefile.pid', 'w') pidfile.write('pid')
pidfile.close()

then you can check if it is still running when your ?program? restarts
and can kill it.

If it restarts yeah.

maybe not the perfect answer, but it answers an imperfect question.

Any details you need?