test_socket: testCmsgTrunc0() deadlock on ARM Raspbian 3.x

[vstinner]

test_socket adds a lock on addCleanup:

class ThreadSafeCleanupTestCase:
    """Subclass of unittest.TestCase with thread-safe cleanup methods.

    This subclass protects the addCleanup() and doCleanups() methods
    with a recursive lock.
    """

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self._cleanup_lock = threading.RLock()

    def addCleanup(self, *args, **kwargs):
        with self._cleanup_lock:
            return super().addCleanup(*args, **kwargs)

    def doCleanups(self, *args, **kwargs):
        with self._cleanup_lock:
            return super().doCleanups(*args, **kwargs)

Problem: what happens if a thread calls addCleanup() while the main thread is calling doCleanups()? Well, a deadlock.

ARM Raspbian 3.x:

0:40:31 load avg: 3.05 [337/467/1] test_socket worker non-zero exit code (Exit code 1) -- running (1): test_math (1 min 4 sec)
Timeout (0:40:00)!
Thread 0xf5dea440 (most recent call first):
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/test/test_socket.py", line 230 in addCleanup
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/test/test_socket.py", line 3576 in newFDs
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/test/test_socket.py", line 3611 in createAndSendFDs
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/test/test_socket.py", line 3841 in _testCmsgTrunc0
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/test/test_socket.py", line 402 in clientRun

Thread 0xf7acb040 (most recent call first):
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/threading.py", line 348 in wait
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/threading.py", line 648 in wait
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/unittest/case.py", line 597 in _callCleanup
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/unittest/case.py", line 673 in doCleanups
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/test/test_socket.py", line 235 in doCleanups
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/unittest/case.py", line 640 in run
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/unittest/case.py", line 692 in __call__
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/unittest/suite.py", line 122 in run
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/unittest/suite.py", line 84 in __call__
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/unittest/suite.py", line 122 in run
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/unittest/suite.py", line 84 in __call__
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/unittest/suite.py", line 122 in run
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/unittest/suite.py", line 84 in __call__
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/unittest/runner.py", line 240 in run
  File "/var/lib/buildbot/workers/3.x.gps-raspbian.nondebug/build/Lib/test/support/__init__.py", line 1155 in _run_suite
  ...

build: https://buildbot.python.org/all/#/builders/424/builds/5065

Linked PRs

• gh-110167: Increase support.LOOPBACK_TIMEOUT to 10 seconds #110413
• gh-110167: Fix test_socket deadlock in doCleanups() #110416
• [3.12] gh-110167: Fix test_socket deadlock in doCleanups() (GH-110416) #110423
• [3.11] gh-110167: Fix test_socket deadlock in doCleanups() (GH-110416) #110424
• [3.12] gh-110167: Increase support.LOOPBACK_TIMEOUT to 10 seconds (#110413) #110427
• [3.11] [3.12] gh-110167: Increase support.LOOPBACK_TIMEOUT to 10 seconds (GH-110413) (GH-110427) #110440

[vstinner]

Patch to trigger the bug:

diff --git a/Lib/test/test_socket.py b/Lib/test/test_socket.py
index 09605f7e77..d3638bccc2 100644
--- a/Lib/test/test_socket.py
+++ b/Lib/test/test_socket.py
@@ -215,6 +215,8 @@ def setUp(self):
         self.serv = socket.socket(socket.AF_INET, socket.SOCK_DGRAM, socket.IPPROTO_UDPLITE)
         self.port = socket_helper.bind_port(self.serv)
 
+CLEANUP_EVENT = threading.Event()
+
 class ThreadSafeCleanupTestCase:
     """Subclass of unittest.TestCase with thread-safe cleanup methods.
 
@@ -232,6 +234,7 @@ def addCleanup(self, *args, **kwargs):
 
     def doCleanups(self, *args, **kwargs):
         with self._cleanup_lock:
+            CLEANUP_EVENT.set()
             return super().doCleanups(*args, **kwargs)
 
 class SocketCANTest(unittest.TestCase):
@@ -2818,7 +2821,7 @@ class SendrecvmsgBase(ThreadSafeCleanupTestCase):
 
     # Time in seconds to wait before considering a test failed, or
     # None for no timeout.  Not all tests actually set a timeout.
-    fail_timeout = support.LOOPBACK_TIMEOUT
+    fail_timeout = 1e-9
 
     def setUp(self):
         self.misc_event = threading.Event()
@@ -3573,6 +3576,7 @@ def newFDs(self, n):
         fds = []
         for i in range(n):
             fd, path = tempfile.mkstemp()
+            CLEANUP_EVENT.wait()
             self.addCleanup(os.unlink, path)
             self.addCleanup(os.close, fd)
             os.write(fd, str(i).encode())