1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
|
# This is a variant of the very old (early 90's) file
# Demo/threads/bug.py. It simply provokes a number of threads into
# trying to import the same module "at the same time".
# There are no pleasant failure modes -- most likely is that Python
# complains several times about module random having no attribute
# randrange, and then Python hangs.
import _imp as imp
import os
import importlib
import sys
import time
import shutil
import threading
import unittest
from unittest import mock
from test.support import (
verbose, run_unittest, TESTFN, reap_threads,
forget, unlink, rmtree, start_threads)
def task(N, done, done_tasks, errors):
try:
# We don't use modulefinder but still import it in order to stress
# importing of different modules from several threads.
if len(done_tasks) % 2:
import modulefinder
import random
else:
import random
import modulefinder
# This will fail if random is not completely initialized
x = random.randrange(1, 3)
except Exception as e:
errors.append(e.with_traceback(None))
finally:
done_tasks.append(threading.get_ident())
finished = len(done_tasks) == N
if finished:
done.set()
def mock_register_at_fork(func):
# bpo-30599: Mock os.register_at_fork() when importing the random module,
# since this function doesn't allow to unregister callbacks and would leak
# memory.
return mock.patch('os.register_at_fork', create=True)(func)
# Create a circular import structure: A -> C -> B -> D -> A
# NOTE: `time` is already loaded and therefore doesn't threaten to deadlock.
circular_imports_modules = {
'A': """if 1:
import time
time.sleep(%(delay)s)
x = 'a'
import C
""",
'B': """if 1:
import time
time.sleep(%(delay)s)
x = 'b'
import D
""",
'C': """import B""",
'D': """import A""",
}
class Finder:
"""A dummy finder to detect concurrent access to its find_spec()
method."""
def __init__(self):
self.numcalls = 0
self.x = 0
self.lock = threading.Lock()
def find_spec(self, name, path=None, target=None):
# Simulate some thread-unsafe behaviour. If calls to find_spec()
# are properly serialized, `x` will end up the same as `numcalls`.
# Otherwise not.
assert imp.lock_held()
with self.lock:
self.numcalls += 1
x = self.x
time.sleep(0.01)
self.x = x + 1
class FlushingFinder:
"""A dummy finder which flushes sys.path_importer_cache when it gets
called."""
def find_spec(self, name, path=None, target=None):
sys.path_importer_cache.clear()
class ThreadedImportTests(unittest.TestCase):
def setUp(self):
self.old_random = sys.modules.pop('random', None)
def tearDown(self):
# If the `random` module was already initialized, we restore the
# old module at the end so that pickling tests don't fail.
# See http://bugs.python.org/issue3657#msg110461
if self.old_random is not None:
sys.modules['random'] = self.old_random
@mock_register_at_fork
def check_parallel_module_init(self, mock_os):
if imp.lock_held():
# This triggers on, e.g., from test import autotest.
raise unittest.SkipTest("can't run when import lock is held")
done = threading.Event()
for N in (20, 50) * 3:
if verbose:
print("Trying", N, "threads ...", end=' ')
# Make sure that random and modulefinder get reimported freshly
for modname in ['random', 'modulefinder']:
try:
del sys.modules[modname]
except KeyError:
pass
errors = []
done_tasks = []
done.clear()
t0 = time.monotonic()
with start_threads(threading.Thread(target=task,
args=(N, done, done_tasks, errors,))
for i in range(N)):
pass
completed = done.wait(10 * 60)
dt = time.monotonic() - t0
if verbose:
print("%.1f ms" % (dt*1e3), flush=True, end=" ")
dbg_info = 'done: %s/%s' % (len(done_tasks), N)
self.assertFalse(errors, dbg_info)
self.assertTrue(completed, dbg_info)
if verbose:
print("OK.")
def test_parallel_module_init(self):
self.check_parallel_module_init()
def test_parallel_meta_path(self):
finder = Finder()
sys.meta_path.insert(0, finder)
try:
self.check_parallel_module_init()
self.assertGreater(finder.numcalls, 0)
self.assertEqual(finder.x, finder.numcalls)
finally:
sys.meta_path.remove(finder)
def test_parallel_path_hooks(self):
# Here the Finder instance is only used to check concurrent calls
# to path_hook().
finder = Finder()
# In order for our path hook to be called at each import, we need
# to flush the path_importer_cache, which we do by registering a
# dedicated meta_path entry.
flushing_finder = FlushingFinder()
def path_hook(path):
finder.find_spec('')
raise ImportError
sys.path_hooks.insert(0, path_hook)
sys.meta_path.append(flushing_finder)
try:
# Flush the cache a first time
flushing_finder.find_spec('')
numtests = self.check_parallel_module_init()
self.assertGreater(finder.numcalls, 0)
self.assertEqual(finder.x, finder.numcalls)
finally:
sys.meta_path.remove(flushing_finder)
sys.path_hooks.remove(path_hook)
def test_import_hangers(self):
# In case this test is run again, make sure the helper module
# gets loaded from scratch again.
try:
del sys.modules['test.test_importlib.threaded_import_hangers']
except KeyError:
pass
import test.test_importlib.threaded_import_hangers
self.assertFalse(test.test_importlib.threaded_import_hangers.errors)
def test_circular_imports(self):
# The goal of this test is to exercise implementations of the import
# lock which use a per-module lock, rather than a global lock.
# In these implementations, there is a possible deadlock with
# circular imports, for example:
# - thread 1 imports A (grabbing the lock for A) which imports B
# - thread 2 imports B (grabbing the lock for B) which imports A
# Such implementations should be able to detect such situations and
# resolve them one way or the other, without freezing.
# NOTE: our test constructs a slightly less trivial import cycle,
# in order to better stress the deadlock avoidance mechanism.
delay = 0.5
os.mkdir(TESTFN)
self.addCleanup(shutil.rmtree, TESTFN)
sys.path.insert(0, TESTFN)
self.addCleanup(sys.path.remove, TESTFN)
for name, contents in circular_imports_modules.items():
contents = contents % {'delay': delay}
with open(os.path.join(TESTFN, name + ".py"), "wb") as f:
f.write(contents.encode('utf-8'))
self.addCleanup(forget, name)
importlib.invalidate_caches()
results = []
def import_ab():
import A
results.append(getattr(A, 'x', None))
def import_ba():
import B
results.append(getattr(B, 'x', None))
t1 = threading.Thread(target=import_ab)
t2 = threading.Thread(target=import_ba)
t1.start()
t2.start()
t1.join()
t2.join()
self.assertEqual(set(results), {'a', 'b'})
@mock_register_at_fork
def test_side_effect_import(self, mock_os):
code = """if 1:
import threading
def target():
import random
t = threading.Thread(target=target)
t.start()
t.join()
t = None"""
sys.path.insert(0, os.curdir)
self.addCleanup(sys.path.remove, os.curdir)
filename = TESTFN + ".py"
with open(filename, "wb") as f:
f.write(code.encode('utf-8'))
self.addCleanup(unlink, filename)
self.addCleanup(forget, TESTFN)
self.addCleanup(rmtree, '__pycache__')
importlib.invalidate_caches()
__import__(TESTFN)
del sys.modules[TESTFN]
@reap_threads
def test_main():
old_switchinterval = None
try:
old_switchinterval = sys.getswitchinterval()
sys.setswitchinterval(1e-5)
except AttributeError:
pass
try:
run_unittest(ThreadedImportTests)
finally:
if old_switchinterval is not None:
sys.setswitchinterval(old_switchinterval)
if __name__ == "__main__":
test_main()
|