'UnixDomainSocketConnection' object has no attribute '_command_packer'

[TheUbuntuGuy]

Version: v4.5.0

Platform: Debian Bullseye using Python v3.9.2

Description:
The following code which used to work on up to and including v4.4.2 now crashes with the stack trace below on v4.5.0.

redis_conn = redis.Redis(
        unix_socket_path="/var/run/redis/redis-server.sock",
        decode_responses=True)
redis_conn.ping()

Traceback (most recent call last):
  File "<redacted>", line 142, in main
    redis_conn.ping()
  File "/usr/local/lib/python3.9/dist-packages/redis/commands/core.py", line 1194, in ping
    return self.execute_command("PING", **kwargs)
  File "/usr/local/lib/python3.9/dist-packages/redis/client.py", line 1258, in execute_command
    return conn.retry.call_with_retry(
  File "/usr/local/lib/python3.9/dist-packages/redis/retry.py", line 46, in call_with_retry
    return do()
  File "/usr/local/lib/python3.9/dist-packages/redis/client.py", line 1259, in <lambda>
    lambda: self._send_command_parse_response(
  File "/usr/local/lib/python3.9/dist-packages/redis/client.py", line 1234, in _send_command_parse_response
    conn.send_command(*args)
  File "/usr/local/lib/python3.9/dist-packages/redis/connection.py", line 916, in send_command
    self._command_packer.pack(*args),
AttributeError: 'UnixDomainSocketConnection' object has no attribute '_command_packer'

[alee]

also seeing this in our apps, had to pin to 4.4.2 to prevent our django app from crashing anytime it attempted to use the cache

[woutdenolf]

I guess this is the issue:

class Connection:
    def __init__(self, ...):
        ...
        self._command_packer = self._construct_command_packer(command_packer)

class SSLConnection(Connection):
    def __init__(self, ...):
        ...
        super().__init__(...)

class UnixDomainSocketConnection(Connection):
    def __init__(self, ...):
         # does not call _construct_command_packer nor super().__init__()

[woutdenolf]

Seems like the bug was introduced just before the 4.5.0 release: #2570

@prokazov Could you take a look please?

Strange that this does not get picked up by the tests.

[prokazov]

Looks like there is no synchronous tests that run any command on SSLConnection/UnixDomainSocketConnection.

[woutdenolf]

Actually SSLConnection does call super().__init__ so that one should be fine.

[woutdenolf]

The fact that UnixDomainSocketConnection.__init__ does not call super().__init__ is probably worth fixing. I see lots of copy&paste so seems doable? Perhaps introduce an AbstractConnection with abstract methods _connect, repr_pieces and _error_message?

[woutdenolf]

Connection has host and port while UnixDomainSocketConnection has path.

Connection has in addition socket_connect_timeout, socket_keepalive , socket_keepalive_options and socket_type.

Lastly there is if retry or retry_on_error: vs. if retry_on_error:. This discrepancy is probably also a mistake #2377 @barshaul ?

The rest is identical.

[prokazov]

@woutdenolf @TheUbuntuGuy @chayim @dvora-h
Here is the fix, please review it:
#2583

@chayim @dvora-h
I tried to add a full end-to-end Unix socket test but it took a little bit too much time so I felt back to a unit one.
Last thing I tried was something like:
touch /tmp/redis.sock and add it to volumes in tox.ini[master], which still resulted to "connection refused" ("docker exec -it master ls /tmp/" confirmed the presence of this file in the container) and I gave up.