cpython/Lib/test/test_asyncio/test_subprocess.py

550 lines
20 KiB
Python
Raw Normal View History

import signal
import sys
import unittest
import warnings
from unittest import mock
2014-12-18 07:29:53 -04:00
import asyncio
asyncio: sync with Tulip Issue #23347: send_signal(), kill() and terminate() methods of BaseSubprocessTransport now check if the transport was closed and if the process exited. Issue #23347: Refactor creation of subprocess transports. Changes on BaseSubprocessTransport: * Add a wait() method to wait until the child process exit * The constructor now accepts an optional waiter parameter. The _post_init() coroutine must not be called explicitly anymore. It makes subprocess transports closer to other transports, and it gives more freedom if we want later to change completly how subprocess transports are created. * close() now kills the process instead of kindly terminate it: the child process may ignore SIGTERM and continue to run. Call explicitly terminate() and wait() if you want to kindly terminate the child process. * close() now logs a warning in debug mode if the process is still running and needs to be killed * _make_subprocess_transport() is now fully asynchronous again: if the creation of the transport failed, wait asynchronously for the process eixt. Before the wait was synchronous. This change requires close() to *kill*, and not terminate, the child process. * Remove the _kill_wait() method, replaced with a more agressive close() method. It fixes _make_subprocess_transport() on error. BaseSubprocessTransport.close() calls the close() method of pipe transports, whereas _kill_wait() closed directly pipes of the subprocess.Popen object without unregistering file descriptors from the selector (which caused severe bugs). These changes simplifies the code of subprocess.py.
2015-01-29 19:05:19 -04:00
from asyncio import base_subprocess
2014-12-18 07:29:53 -04:00
from asyncio import subprocess
from test.test_asyncio import utils as test_utils
from test import support
if sys.platform != 'win32':
from asyncio import unix_events
# Program blocking
PROGRAM_BLOCKED = [sys.executable, '-c', 'import time; time.sleep(3600)']
# Program copying input to output
PROGRAM_CAT = [
sys.executable, '-c',
';'.join(('import sys',
'data = sys.stdin.buffer.read()',
'sys.stdout.buffer.write(data)'))]
asyncio: sync with Tulip Issue #23347: send_signal(), kill() and terminate() methods of BaseSubprocessTransport now check if the transport was closed and if the process exited. Issue #23347: Refactor creation of subprocess transports. Changes on BaseSubprocessTransport: * Add a wait() method to wait until the child process exit * The constructor now accepts an optional waiter parameter. The _post_init() coroutine must not be called explicitly anymore. It makes subprocess transports closer to other transports, and it gives more freedom if we want later to change completly how subprocess transports are created. * close() now kills the process instead of kindly terminate it: the child process may ignore SIGTERM and continue to run. Call explicitly terminate() and wait() if you want to kindly terminate the child process. * close() now logs a warning in debug mode if the process is still running and needs to be killed * _make_subprocess_transport() is now fully asynchronous again: if the creation of the transport failed, wait asynchronously for the process eixt. Before the wait was synchronous. This change requires close() to *kill*, and not terminate, the child process. * Remove the _kill_wait() method, replaced with a more agressive close() method. It fixes _make_subprocess_transport() on error. BaseSubprocessTransport.close() calls the close() method of pipe transports, whereas _kill_wait() closed directly pipes of the subprocess.Popen object without unregistering file descriptors from the selector (which caused severe bugs). These changes simplifies the code of subprocess.py.
2015-01-29 19:05:19 -04:00
class TestSubprocessTransport(base_subprocess.BaseSubprocessTransport):
def _start(self, *args, **kwargs):
self._proc = mock.Mock()
self._proc.stdin = None
self._proc.stdout = None
self._proc.stderr = None
self._proc.pid = -1
asyncio: sync with Tulip Issue #23347: send_signal(), kill() and terminate() methods of BaseSubprocessTransport now check if the transport was closed and if the process exited. Issue #23347: Refactor creation of subprocess transports. Changes on BaseSubprocessTransport: * Add a wait() method to wait until the child process exit * The constructor now accepts an optional waiter parameter. The _post_init() coroutine must not be called explicitly anymore. It makes subprocess transports closer to other transports, and it gives more freedom if we want later to change completly how subprocess transports are created. * close() now kills the process instead of kindly terminate it: the child process may ignore SIGTERM and continue to run. Call explicitly terminate() and wait() if you want to kindly terminate the child process. * close() now logs a warning in debug mode if the process is still running and needs to be killed * _make_subprocess_transport() is now fully asynchronous again: if the creation of the transport failed, wait asynchronously for the process eixt. Before the wait was synchronous. This change requires close() to *kill*, and not terminate, the child process. * Remove the _kill_wait() method, replaced with a more agressive close() method. It fixes _make_subprocess_transport() on error. BaseSubprocessTransport.close() calls the close() method of pipe transports, whereas _kill_wait() closed directly pipes of the subprocess.Popen object without unregistering file descriptors from the selector (which caused severe bugs). These changes simplifies the code of subprocess.py.
2015-01-29 19:05:19 -04:00
class SubprocessTransportTests(test_utils.TestCase):
def setUp(self):
super().setUp()
asyncio: sync with Tulip Issue #23347: send_signal(), kill() and terminate() methods of BaseSubprocessTransport now check if the transport was closed and if the process exited. Issue #23347: Refactor creation of subprocess transports. Changes on BaseSubprocessTransport: * Add a wait() method to wait until the child process exit * The constructor now accepts an optional waiter parameter. The _post_init() coroutine must not be called explicitly anymore. It makes subprocess transports closer to other transports, and it gives more freedom if we want later to change completly how subprocess transports are created. * close() now kills the process instead of kindly terminate it: the child process may ignore SIGTERM and continue to run. Call explicitly terminate() and wait() if you want to kindly terminate the child process. * close() now logs a warning in debug mode if the process is still running and needs to be killed * _make_subprocess_transport() is now fully asynchronous again: if the creation of the transport failed, wait asynchronously for the process eixt. Before the wait was synchronous. This change requires close() to *kill*, and not terminate, the child process. * Remove the _kill_wait() method, replaced with a more agressive close() method. It fixes _make_subprocess_transport() on error. BaseSubprocessTransport.close() calls the close() method of pipe transports, whereas _kill_wait() closed directly pipes of the subprocess.Popen object without unregistering file descriptors from the selector (which caused severe bugs). These changes simplifies the code of subprocess.py.
2015-01-29 19:05:19 -04:00
self.loop = self.new_test_loop()
self.set_event_loop(self.loop)
def create_transport(self, waiter=None):
protocol = mock.Mock()
protocol.connection_made._is_coroutine = False
protocol.process_exited._is_coroutine = False
transport = TestSubprocessTransport(
self.loop, protocol, ['test'], False,
None, None, None, 0, waiter=waiter)
return (transport, protocol)
def test_proc_exited(self):
waiter = asyncio.Future(loop=self.loop)
transport, protocol = self.create_transport(waiter)
transport._process_exited(6)
self.loop.run_until_complete(waiter)
self.assertEqual(transport.get_returncode(), 6)
self.assertTrue(protocol.connection_made.called)
self.assertTrue(protocol.process_exited.called)
self.assertTrue(protocol.connection_lost.called)
self.assertEqual(protocol.connection_lost.call_args[0], (None,))
self.assertFalse(transport.is_closing())
asyncio: sync with Tulip Issue #23347: send_signal(), kill() and terminate() methods of BaseSubprocessTransport now check if the transport was closed and if the process exited. Issue #23347: Refactor creation of subprocess transports. Changes on BaseSubprocessTransport: * Add a wait() method to wait until the child process exit * The constructor now accepts an optional waiter parameter. The _post_init() coroutine must not be called explicitly anymore. It makes subprocess transports closer to other transports, and it gives more freedom if we want later to change completly how subprocess transports are created. * close() now kills the process instead of kindly terminate it: the child process may ignore SIGTERM and continue to run. Call explicitly terminate() and wait() if you want to kindly terminate the child process. * close() now logs a warning in debug mode if the process is still running and needs to be killed * _make_subprocess_transport() is now fully asynchronous again: if the creation of the transport failed, wait asynchronously for the process eixt. Before the wait was synchronous. This change requires close() to *kill*, and not terminate, the child process. * Remove the _kill_wait() method, replaced with a more agressive close() method. It fixes _make_subprocess_transport() on error. BaseSubprocessTransport.close() calls the close() method of pipe transports, whereas _kill_wait() closed directly pipes of the subprocess.Popen object without unregistering file descriptors from the selector (which caused severe bugs). These changes simplifies the code of subprocess.py.
2015-01-29 19:05:19 -04:00
self.assertIsNone(transport._loop)
self.assertIsNone(transport._proc)
self.assertIsNone(transport._protocol)
# methods must raise ProcessLookupError if the process exited
self.assertRaises(ProcessLookupError,
transport.send_signal, signal.SIGTERM)
self.assertRaises(ProcessLookupError, transport.terminate)
self.assertRaises(ProcessLookupError, transport.kill)
transport.close()
def test_subprocess_repr(self):
waiter = asyncio.Future(loop=self.loop)
transport, protocol = self.create_transport(waiter)
transport._process_exited(6)
self.loop.run_until_complete(waiter)
self.assertEqual(
repr(transport),
"<TestSubprocessTransport pid=-1 returncode=6>"
)
transport._returncode = None
self.assertEqual(
repr(transport),
"<TestSubprocessTransport pid=-1 running>"
)
transport._pid = None
transport._returncode = None
self.assertEqual(
repr(transport),
"<TestSubprocessTransport not started>"
)
transport.close()
asyncio: sync with Tulip Issue #23347: send_signal(), kill() and terminate() methods of BaseSubprocessTransport now check if the transport was closed and if the process exited. Issue #23347: Refactor creation of subprocess transports. Changes on BaseSubprocessTransport: * Add a wait() method to wait until the child process exit * The constructor now accepts an optional waiter parameter. The _post_init() coroutine must not be called explicitly anymore. It makes subprocess transports closer to other transports, and it gives more freedom if we want later to change completly how subprocess transports are created. * close() now kills the process instead of kindly terminate it: the child process may ignore SIGTERM and continue to run. Call explicitly terminate() and wait() if you want to kindly terminate the child process. * close() now logs a warning in debug mode if the process is still running and needs to be killed * _make_subprocess_transport() is now fully asynchronous again: if the creation of the transport failed, wait asynchronously for the process eixt. Before the wait was synchronous. This change requires close() to *kill*, and not terminate, the child process. * Remove the _kill_wait() method, replaced with a more agressive close() method. It fixes _make_subprocess_transport() on error. BaseSubprocessTransport.close() calls the close() method of pipe transports, whereas _kill_wait() closed directly pipes of the subprocess.Popen object without unregistering file descriptors from the selector (which caused severe bugs). These changes simplifies the code of subprocess.py.
2015-01-29 19:05:19 -04:00
class SubprocessMixin:
2014-02-18 23:56:15 -04:00
def test_stdin_stdout(self):
args = PROGRAM_CAT
async def run(data):
proc = await asyncio.create_subprocess_exec(
*args,
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
loop=self.loop)
# feed data
proc.stdin.write(data)
await proc.stdin.drain()
proc.stdin.close()
# get output and exitcode
data = await proc.stdout.read()
exitcode = await proc.wait()
return (exitcode, data)
task = run(b'some data')
task = asyncio.wait_for(task, 60.0, loop=self.loop)
exitcode, stdout = self.loop.run_until_complete(task)
self.assertEqual(exitcode, 0)
self.assertEqual(stdout, b'some data')
def test_communicate(self):
args = PROGRAM_CAT
async def run(data):
proc = await asyncio.create_subprocess_exec(
*args,
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
loop=self.loop)
stdout, stderr = await proc.communicate(data)
return proc.returncode, stdout
task = run(b'some data')
task = asyncio.wait_for(task, 60.0, loop=self.loop)
exitcode, stdout = self.loop.run_until_complete(task)
self.assertEqual(exitcode, 0)
self.assertEqual(stdout, b'some data')
def test_shell(self):
create = asyncio.create_subprocess_shell('exit 7',
loop=self.loop)
proc = self.loop.run_until_complete(create)
exitcode = self.loop.run_until_complete(proc.wait())
self.assertEqual(exitcode, 7)
def test_start_new_session(self):
# start the new process in a new session
create = asyncio.create_subprocess_shell('exit 8',
start_new_session=True,
loop=self.loop)
proc = self.loop.run_until_complete(create)
exitcode = self.loop.run_until_complete(proc.wait())
self.assertEqual(exitcode, 8)
def test_kill(self):
args = PROGRAM_BLOCKED
create = asyncio.create_subprocess_exec(*args, loop=self.loop)
proc = self.loop.run_until_complete(create)
proc.kill()
returncode = self.loop.run_until_complete(proc.wait())
if sys.platform == 'win32':
self.assertIsInstance(returncode, int)
# expect 1 but sometimes get 0
else:
self.assertEqual(-signal.SIGKILL, returncode)
def test_terminate(self):
args = PROGRAM_BLOCKED
create = asyncio.create_subprocess_exec(*args, loop=self.loop)
proc = self.loop.run_until_complete(create)
proc.terminate()
returncode = self.loop.run_until_complete(proc.wait())
if sys.platform == 'win32':
self.assertIsInstance(returncode, int)
# expect 1 but sometimes get 0
else:
self.assertEqual(-signal.SIGTERM, returncode)
@unittest.skipIf(sys.platform == 'win32', "Don't have SIGHUP")
def test_send_signal(self):
# bpo-31034: Make sure that we get the default signal handler (killing
# the process). The parent process may have decided to ignore SIGHUP,
# and signal handlers are inherited.
old_handler = signal.signal(signal.SIGHUP, signal.SIG_DFL)
try:
code = 'import time; print("sleeping", flush=True); time.sleep(3600)'
args = [sys.executable, '-c', code]
create = asyncio.create_subprocess_exec(*args,
stdout=subprocess.PIPE,
loop=self.loop)
proc = self.loop.run_until_complete(create)
async def send_signal(proc):
# basic synchronization to wait until the program is sleeping
line = await proc.stdout.readline()
self.assertEqual(line, b'sleeping\n')
proc.send_signal(signal.SIGHUP)
returncode = await proc.wait()
return returncode
returncode = self.loop.run_until_complete(send_signal(proc))
self.assertEqual(-signal.SIGHUP, returncode)
finally:
signal.signal(signal.SIGHUP, old_handler)
def prepare_broken_pipe_test(self):
# buffer large enough to feed the whole pipe buffer
large_data = b'x' * support.PIPE_MAX_SIZE
# the program ends before the stdin can be feeded
create = asyncio.create_subprocess_exec(
sys.executable,
'-c', 'print("hello", flush=True)',
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
loop=self.loop)
proc = self.loop.run_until_complete(create)
return (proc, large_data)
def test_stdin_broken_pipe(self):
proc, large_data = self.prepare_broken_pipe_test()
async def write_stdin(proc, data):
await proc.stdout.readline()
proc.stdin.write(data)
await proc.stdin.drain()
coro = write_stdin(proc, large_data)
# drain() must raise BrokenPipeError or ConnectionResetError
with test_utils.disable_logger():
self.assertRaises((BrokenPipeError, ConnectionResetError),
self.loop.run_until_complete, coro)
self.loop.run_until_complete(proc.wait())
def test_communicate_ignore_broken_pipe(self):
proc, large_data = self.prepare_broken_pipe_test()
# communicate() must ignore BrokenPipeError when feeding stdin
with test_utils.disable_logger():
self.loop.run_until_complete(proc.communicate(large_data))
self.loop.run_until_complete(proc.wait())
def test_pause_reading(self):
limit = 10
size = (limit * 2 + 1)
async def test_pause_reading():
code = '\n'.join((
'import sys',
'sys.stdout.write("x" * %s)' % size,
'sys.stdout.flush()',
))
connect_read_pipe = self.loop.connect_read_pipe
async def connect_read_pipe_mock(*args, **kw):
transport, protocol = await connect_read_pipe(*args, **kw)
transport.pause_reading = mock.Mock()
transport.resume_reading = mock.Mock()
return (transport, protocol)
self.loop.connect_read_pipe = connect_read_pipe_mock
proc = await asyncio.create_subprocess_exec(
sys.executable, '-c', code,
stdin=asyncio.subprocess.PIPE,
stdout=asyncio.subprocess.PIPE,
limit=limit,
loop=self.loop)
stdout_transport = proc._transport.get_pipe_transport(1)
stdout, stderr = await proc.communicate()
# The child process produced more than limit bytes of output,
# the stream reader transport should pause the protocol to not
# allocate too much memory.
return (stdout, stdout_transport)
# Issue #22685: Ensure that the stream reader pauses the protocol
# when the child process produces too much data
stdout, transport = self.loop.run_until_complete(test_pause_reading())
self.assertEqual(stdout, b'x' * size)
self.assertTrue(transport.pause_reading.called)
self.assertTrue(transport.resume_reading.called)
def test_stdin_not_inheritable(self):
# asyncio issue #209: stdin must not be inheritable, otherwise
# the Process.communicate() hangs
async def len_message(message):
code = 'import sys; data = sys.stdin.read(); print(len(data))'
proc = await asyncio.create_subprocess_exec(
sys.executable, '-c', code,
stdin=asyncio.subprocess.PIPE,
stdout=asyncio.subprocess.PIPE,
stderr=asyncio.subprocess.PIPE,
close_fds=False,
loop=self.loop)
stdout, stderr = await proc.communicate(message)
exitcode = await proc.wait()
return (stdout, exitcode)
output, exitcode = self.loop.run_until_complete(len_message(b'abc'))
self.assertEqual(output.rstrip(), b'3')
self.assertEqual(exitcode, 0)
def test_empty_input(self):
async def empty_input():
code = 'import sys; data = sys.stdin.read(); print(len(data))'
proc = await asyncio.create_subprocess_exec(
sys.executable, '-c', code,
stdin=asyncio.subprocess.PIPE,
stdout=asyncio.subprocess.PIPE,
stderr=asyncio.subprocess.PIPE,
close_fds=False,
loop=self.loop)
stdout, stderr = await proc.communicate(b'')
exitcode = await proc.wait()
return (stdout, exitcode)
output, exitcode = self.loop.run_until_complete(empty_input())
self.assertEqual(output.rstrip(), b'0')
self.assertEqual(exitcode, 0)
def test_cancel_process_wait(self):
# Issue #23140: cancel Process.wait()
async def cancel_wait():
proc = await asyncio.create_subprocess_exec(
*PROGRAM_BLOCKED,
loop=self.loop)
# Create an internal future waiting on the process exit
task = self.loop.create_task(proc.wait())
self.loop.call_soon(task.cancel)
try:
await task
except asyncio.CancelledError:
pass
# Cancel the future
task.cancel()
# Kill the process and wait until it is done
proc.kill()
await proc.wait()
self.loop.run_until_complete(cancel_wait())
def test_cancel_make_subprocess_transport_exec(self):
async def cancel_make_transport():
coro = asyncio.create_subprocess_exec(*PROGRAM_BLOCKED,
loop=self.loop)
task = self.loop.create_task(coro)
self.loop.call_soon(task.cancel)
try:
await task
except asyncio.CancelledError:
pass
# ignore the log:
# "Exception during subprocess creation, kill the subprocess"
with test_utils.disable_logger():
self.loop.run_until_complete(cancel_make_transport())
def test_cancel_post_init(self):
async def cancel_make_transport():
coro = self.loop.subprocess_exec(asyncio.SubprocessProtocol,
*PROGRAM_BLOCKED)
task = self.loop.create_task(coro)
self.loop.call_soon(task.cancel)
try:
await task
except asyncio.CancelledError:
pass
# ignore the log:
# "Exception during subprocess creation, kill the subprocess"
with test_utils.disable_logger():
self.loop.run_until_complete(cancel_make_transport())
test_utils.run_briefly(self.loop)
def test_close_kill_running(self):
async def kill_running():
create = self.loop.subprocess_exec(asyncio.SubprocessProtocol,
*PROGRAM_BLOCKED)
transport, protocol = await create
kill_called = False
def kill():
nonlocal kill_called
kill_called = True
orig_kill()
proc = transport.get_extra_info('subprocess')
orig_kill = proc.kill
proc.kill = kill
returncode = transport.get_returncode()
transport.close()
await transport._wait()
return (returncode, kill_called)
# Ignore "Close running child process: kill ..." log
with test_utils.disable_logger():
returncode, killed = self.loop.run_until_complete(kill_running())
self.assertIsNone(returncode)
# transport.close() must kill the process if it is still running
self.assertTrue(killed)
test_utils.run_briefly(self.loop)
def test_close_dont_kill_finished(self):
async def kill_running():
create = self.loop.subprocess_exec(asyncio.SubprocessProtocol,
*PROGRAM_BLOCKED)
transport, protocol = await create
proc = transport.get_extra_info('subprocess')
# kill the process (but asyncio is not notified immediately)
proc.kill()
proc.wait()
proc.kill = mock.Mock()
proc_returncode = proc.poll()
transport_returncode = transport.get_returncode()
transport.close()
return (proc_returncode, transport_returncode, proc.kill.called)
# Ignore "Unknown child process pid ..." log of SafeChildWatcher,
# emitted because the test already consumes the exit status:
# proc.wait()
with test_utils.disable_logger():
result = self.loop.run_until_complete(kill_running())
test_utils.run_briefly(self.loop)
proc_returncode, transport_return_code, killed = result
self.assertIsNotNone(proc_returncode)
self.assertIsNone(transport_return_code)
# transport.close() must not kill the process if it finished, even if
# the transport was not notified yet
self.assertFalse(killed)
# Unlike SafeChildWatcher, FastChildWatcher does not pop the
# callbacks if waitpid() is called elsewhere. Let's clear them
# manually to avoid a warning when the watcher is detached.
if (sys.platform != 'win32' and
isinstance(self, SubprocessFastWatcherTests)):
asyncio.get_child_watcher()._callbacks.clear()
def test_popen_error(self):
# Issue #24763: check that the subprocess transport is closed
# when BaseSubprocessTransport fails
if sys.platform == 'win32':
target = 'asyncio.windows_utils.Popen'
else:
target = 'subprocess.Popen'
with mock.patch(target) as popen:
exc = ZeroDivisionError
popen.side_effect = exc
create = asyncio.create_subprocess_exec(sys.executable, '-c',
'pass', loop=self.loop)
with warnings.catch_warnings(record=True) as warns:
with self.assertRaises(exc):
self.loop.run_until_complete(create)
self.assertEqual(warns, [])
def test_read_stdout_after_process_exit(self):
async def execute():
code = '\n'.join(['import sys',
'for _ in range(64):',
' sys.stdout.write("x" * 4096)',
'sys.stdout.flush()',
'sys.exit(1)'])
fut = asyncio.create_subprocess_exec(
sys.executable, '-c', code,
stdout=asyncio.subprocess.PIPE,
loop=self.loop)
process = await fut
while True:
data = await process.stdout.read(65536)
if data:
await asyncio.sleep(0.3, loop=self.loop)
else:
break
self.loop.run_until_complete(execute())
if sys.platform != 'win32':
# Unix
class SubprocessWatcherMixin(SubprocessMixin):
2014-02-18 23:56:15 -04:00
Watcher = None
def setUp(self):
super().setUp()
policy = asyncio.get_event_loop_policy()
self.loop = policy.new_event_loop()
self.set_event_loop(self.loop)
watcher = self.Watcher()
watcher.attach_loop(self.loop)
policy.set_child_watcher(watcher)
self.addCleanup(policy.set_child_watcher, None)
2014-02-18 23:56:15 -04:00
class SubprocessSafeWatcherTests(SubprocessWatcherMixin,
test_utils.TestCase):
2014-02-18 23:56:15 -04:00
Watcher = unix_events.SafeChildWatcher
2014-02-18 23:56:15 -04:00
class SubprocessFastWatcherTests(SubprocessWatcherMixin,
test_utils.TestCase):
2014-02-18 23:56:15 -04:00
Watcher = unix_events.FastChildWatcher
2014-02-18 23:56:15 -04:00
else:
# Windows
class SubprocessProactorTests(SubprocessMixin, test_utils.TestCase):
2014-02-18 23:56:15 -04:00
def setUp(self):
super().setUp()
self.loop = asyncio.ProactorEventLoop()
self.set_event_loop(self.loop)
if __name__ == '__main__':
unittest.main()