Unable to connect via SSH through two proxy jumps

Describe the bug

It seems there's an issue where trying to connect via SSH through two or more proxy jumps fails, resulting in a connection timeout error.

To Reproduce

Let's say we have the following SSH configuration file:

Host jumper1
  Hostname 10.10.10.1
  User devops

Host jumper2
  Hostname 10.20.10.1
  Port 30028
  User devops
  StrictHostKeyChecking no
  UserKnownHostsFile /dev/null
  ProxyJump jumper1

Host test
  Hostname 10.20.10.2
  User devops
  ProxyJump jumper2

When setting ssh_config_file in inventory.py and trying to connect to the test server (which belongs to the test_group) using the following command, a timeout error occurs.

$ pyinfra inventory.py --limit test_group exec -- hostname
--> Loading config...
--> Loading inventory...
--> Connecting to hosts...
    [test] Could not connect ([Errno 60] Operation timed out)

--> Disconnecting from hosts...
--> pyinfra error: No hosts remaining!

I think the problem might be occurring in the following line where config and kwargs are merged, with sock=None contained in kwargs being overwritten to config. https://github.com/pyinfra-dev/pyinfra/blob/9ce7ac44ef6e9d5d8d8a926836e8d294cccbff29/pyinfra/connectors/sshuserclient/client.py#L150

When printing the config variable before and after the line, it was observed that the config.sock parameter had been changed to None in the second hop.

$ pyinfra inventory.py --limit test_group exec -- hostname
--> Loading config...
--> Loading inventory...
--> Connecting to hosts...
--- debug: New hop
Initial config={'port': 22, 'sock': None, 'username': 'devops'}
Updated config={'port': 22, 'sock': None, 'username': 'devops'}
--- debug: New hop
Initial config={'port': 30028, 'sock': <paramiko.Channel 0 (open) window=2097152 -> <paramiko.Transport at 0x33b6fc0 (cipher ***, *** bits) (active; 1 open channel(s))>>, 'username': 'devops'}
Updated config={'port': '30028', 'sock': None, 'username': 'devops'}
    [test] Could not connect ([Errno 60] Operation timed out)

--> Disconnecting from hosts...
--> pyinfra error: No hosts remaining!

When I changed the line as follows:

config.update({k: v for k, v in kwargs.items() if k not in config})

I confirmed that the connection is established successfully, as shown in the following output. However, there seems to be a secondary issue where an EOF-related error occurs at the end.

$ pyinfra inventory.py --limit test_group exec -- hostname
--> Loading config...
--> Loading inventory...
--> Connecting to hosts...
---debug: New hop
Initial config={'port': 22, 'sock': None, 'username': 'devops'}
Updated config={'port': 22, 'sock': None, 'username': 'devops'}
--- debug: New hop
Initial config={'port': 30028, 'sock': <paramiko.Channel 0 (open) window=2097152 -> <paramiko.Transport at 0x6886d80 (cipher ****, **** bits) (active; 1 open channel(s))>>, 'username': 'devops'}
Updated config={'port': 30028, 'sock': <paramiko.Channel 0 (open) window=2097152 -> <paramiko.Transport at 0x6886d80 (cipher ****, **** bits) (active; 1 open channel(s))>>, 'username': 'devops'}
    No host key for [10.20.10.1]:30028 found in known_hosts
--- debug: New hop
Initial config={'port': 22, 'allow_agent': False, 'look_for_keys': False, 'username': 'devops', 'timeout': 10, 'pkey': PKey(alg=****, bits=****, fp=****:****), 'sock': <paramiko.Channel 0 (open) window=2097152 in-buffer=32 -> <paramiko.Transport at 0x70b2810 (cipher ****, **** bits) (active; 1 open channel(s))>>}
Updated config={'port': 22, 'allow_agent': False, 'look_for_keys': False, 'username': 'devops', 'timeout': 10, 'pkey': PKey(alg=****, bits=****, fp=****:****), 'sock': <paramiko.Channel 0 (open) window=2097152 in-buffer=32 -> <paramiko.Transport at 0x70b2810 (cipher ****, **** bits) (active; 1 open channel(s))>>}
    [test] Connected

--> Preparing operations...
    [test] Ready: shell

--> Beginning operation run...
--> Starting operation: server.shell (hostname)
[test] harbor1
    [test] Success

--> Results:
    Operation                 Hosts   Success   Error   No Change
    server.shell (hostname)   1       1         -       -

--> Disconnecting from hosts...
Exception ignored in atexit callback: <function _join_lingering_threads at 0x105eb3920>
Traceback (most recent call last):
  File "****/paramiko/transport.py", line 149, in _join_lingering_threads
    thr.stop_thread()
  File "****/paramiko/transport.py", line 1920, in stop_thread
    self.packetizer.close()
  File "****/paramiko/packet.py", line 228, in close
    self.__socket.close()
  File "****/paramiko/channel.py", line 669, in close
    self.transport._send_user_message(m)
  File "****/paramiko/transport.py", line 1988, in _send_user_message
    self._send_message(data)
  File "****/paramiko/transport.py", line 1964, in _send_message
    self.packetizer.send_message(data)
  File "****/paramiko/packet.py", line 468, in send_message
    self.write_all(out)
  File "****/paramiko/packet.py", line 382, in write_all
    raise EOFError()
EOFError:

Operation code & usage: Included above
Target system information: Apple M1 Pro
Example using the @docker connector (helps isolate the problem): I don't think it is relevant with this issue

Expected behavior

It should be able to connect through two or more jump proxies.

pyinfra-dev / pyinfra