The backend code is not thread-safe, leads to network errors when running from multi-threaded gunicorn etc.

[aivarsk]

For example, getting a value:

django-constance/constance/base.py

Line 18 in 6970708

result = self._backend.get(key)

Goes to the backend:

django-constance/constance/backends/redisd.py

Line 36 in 6970708

value = self._rd.get(self.add_prefix(key))

And there it uses the global Redis connection (self._rd) without locking it.

Which probably relates to this setting in Redis client
https://github.com/redis/redis-py/blob/master/redis/client.py#L271 and then the code fails on the retries when one thread closes the connection and the other tries to use it around here https://github.com/redis/redis-py/blob/master/redis/client.py#L638

Unfortunately I no longer have the stack traces in the slack history. We ended up always using a single-threaded WSGI runner.

I think the correct way is to lock the backend before using it, I have not checked the thread-safety guarantees by RedisPy.

[aivarsk]

Ok, I think the actual problem is in the Redis Py library. But I still think locking backend would be the safe way.

BlockingIOError: [Errno 11] Resource temporarily unavailable
  File "redis/connection.py", line 427, in read_from_socket
    bufflen = recv_into(self._sock, self._buffer)
  File "redis/_compat.py", line 75, in recv_into
    return sock.recv_into(*args, **kwargs)

KeyboardInterrupt: 
(10 additional frame(s) were not displayed)
...
  File "redis/client.py", line 898, in execute_command
    conn = self.connection or pool.get_connection(command_name, **options)
  File "redis/connection.py", line 1198, in get_connection
    if connection.can_read():
  File "redis/connection.py", line 734, in can_read
    return self._parser.can_read(timeout)
  File "redis/connection.py", line 416, in can_read
    return self.read_from_socket(timeout=timeout,
  File "redis/connection.py", line 427, in read_from_socket
    bufflen = recv_into(self._sock, self._buffer)