CA-400275 Tolerate a restarting fairlock service while connecting

.gitignore

@@ -5,6 +5,7 @@
 *.o
 *.swp
 *.pyc
+*.kate-swp
 tests/faultinjection/pfilter
 tests/faultinjection/libipq.c
 tests/faultinjection/libipq/libipq.h

misc/fairlock/fairlock.py

@@ -48,21 +48,29 @@ def _ensure_service(self):
             if time.time() > timeout:
                 raise FairlockServiceTimeout(f"Timed out starting service {service}")
 
+    def _connect_and_recv(self):
+        while True:
+            self.sock.connect(self.sockname)
+            # Merely being connected is not enough. Read a small blob of data.
+            b = self.sock.recv(10)
+            if len(b) > 0:
+                return True
+            # If we got a zero-length return, it means the service exited while we
+            # were waiting. Any timeout we put here would be a max wait time to acquire
+            # the lock, which is dangerous.
+            self._ensure_service()
+
     def __enter__(self):
         if self.connected:
             raise FairlockDeadlock(f"Deadlock on Fairlock resource '{self.name}'")
 
         self.sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
         self.sock.setblocking(True)
         try:
-            self.sock.connect(self.sockname)
-            # Merely being connected is not enough. Read a small blob of data.
-            self.sock.recv(10)
+            self._connect_and_recv()
         except (FileNotFoundError, ConnectionRefusedError):
             self._ensure_service()
-            self.sock.connect(self.sockname)
-            # Merely being connected is not enough. Read a small blob of data.
-            self.sock.recv(10)
+            self._connect_and_recv()
 
         self.sock.send(f'{os.getpid()} - {time.monotonic()}'.encode())
         self.connected = True

tests/test_fairlock.py

@@ -23,6 +23,7 @@ def test_first_lock(self):
         mock_sock = mock.MagicMock()
         self.mock_socket.socket.return_value = mock_sock
         mock_sock.connect.side_effect = [FileNotFoundError(), 0]
+        mock_sock.recv.side_effect = [b'Foop']
         self.mock_os.system.side_effect = [0, 1, 0]
         self.mock_time.time.side_effect = [0, 0, 0]
 
@@ -38,6 +39,7 @@ def test_first_lock_timeout(self):
         mock_sock = mock.MagicMock()
         self.mock_socket.socket.return_value = mock_sock
         mock_sock.connect.side_effect = [FileNotFoundError(), 0]
+        mock_sock.recv.side_effect = [b'Foop']
         self.mock_os.system.side_effect = [0, 1, 1, 1, 0]
         self.mock_time.time.side_effect = [0, 1, 3]
 
@@ -53,6 +55,7 @@ def test_second_lock(self):
         mock_sock = mock.MagicMock()
         self.mock_socket.socket.return_value = mock_sock
         mock_sock.connect.side_effect = [0]
+        mock_sock.recv.side_effect = [b'Foop']
 
         with Fairlock("test"):
             print("Hello World")
@@ -68,6 +71,8 @@ def test_two_locks(self):
         self.mock_socket.socket.side_effect = [mock_sock1, mock_sock2]
         mock_sock1.connect.side_effect = [FileNotFoundError(), 0]
         mock_sock2.connect.side_effect = [FileNotFoundError(), 0]
+        mock_sock1.recv.side_effect = [b'Foop']
+        mock_sock2.recv.side_effect = [b'Foop']
         self.mock_os.system.side_effect = [0, 1, 0, 0, 1, 0]
         self.mock_time.time.side_effect = [0, 0, 0, 0, 0, 0]
 
@@ -81,15 +86,16 @@ def test_double_lock_deadlock(self):
         Test double usage of the same lock
         """
         mock_sock = mock.MagicMock()
-        self.mock_socket.socket.side_effect = [mock_sock]
+        self.mock_socket.socket.side_effect = [mock_sock]        
         mock_sock.connect.side_effect = [FileNotFoundError(), 0]
+        mock_sock.recv.side_effect = [b'Foop', b'Foop']
         self.mock_os.system.side_effect = [0, 1, 0, 0, 1, 0]
         self.mock_time.time.side_effect = [0, 0, 0, 0, 0, 0]
 
         with self.assertRaises(FairlockDeadlock) as err:
             with Fairlock("test") as l:
                 n = Fairlock("test")
-                self.assertEquals(l, n)
+                self.assertEqual(l, n)
                 # Real code would use another 'with Fairlock("test")' here but we cannot
                 # do that because it insists on having a code block as a body, which would
                 # then not be reached, causing a "Test code not fully covered" failure