Avoid stacktrace on process exit in Client.__del__()

Client.close() may call ConnectionPool.release() or
ConnectionPool.disconnect(); both methods may end up calling
os.getpid() (through ConnectionPool._checkpid() or threading.Lock()
(through ConnectionPool.reset()). As mentioned in the Python
documentation [1], at interpreter shutdown, module globals (in this
case, the os and threading module references) may be deleted or set to
None before __del__() methods are called. This causes an
AttributeError to be raised when trying to run e.g. os.getpid(); while
the error is ignored by the interpreter, the traceback is still
printed out to stderr.

Closes redis#3014

[1] https://docs.python.org/3/reference/datamodel.html#object.__del__

redis/client.py

@@ -517,7 +517,18 @@ def __exit__(self, exc_type, exc_value, traceback):
         self.close()
 
     def __del__(self):
-        self.close()
+        try:
+            self.close()
+        except AttributeError as exc:
+            if exc.name in ("getpid", "Lock"):
+                # cf https://github.com/redis/redis-py/issues/3014
+                # Most likely means the "os" or "threading" reference in
+                # connection.py was unloaded before this method was called; this
+                # can happen on process exit, cf the warning in
+                # https://docs.python.org/3/reference/datamodel.html#object.__del__
+                pass
+            else:
+                raise
 
     def close(self):
         # In case a connection property does not yet exist