WhiteListRoundRobinPolicy does not support unix domain sockets

[kbr-scylla]

Most prominent example of this is when we try to use cqlsh to connect to Scylla maintenance socket (currently queued on next): scylladb/scylladb#16489

$ cqlsh /tmp/scylla-workdir/cql.m 
Traceback (most recent call last):
  File "/home/kbr/dev/scylla/tools/cqlsh/bin/cqlsh.py", line 2706, in <module>
    main(*read_options(sys.argv[1:], os.environ))
  File "/home/kbr/dev/scylla/tools/cqlsh/bin/cqlsh.py", line 2647, in main
    shell = Shell(hostname,
            ^^^^^^^^^^^^^^^
  File "/home/kbr/dev/scylla/tools/cqlsh/bin/cqlsh.py", line 491, in __init__
    profiles[EXEC_PROFILE_DEFAULT].load_balancing_policy = WhiteListRoundRobinPolicy([self.hostname])
                                                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kbr/dev/scylla/pyenv/lib64/python3.11/site-packages/cassandra/policies.py", line 425, in __init__
    self._allowed_hosts_resolved = [endpoint[4][0] for a in self._allowed_hosts
                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kbr/dev/scylla/pyenv/lib64/python3.11/site-packages/cassandra/policies.py", line 426, in <listcomp>
    for endpoint in socket.getaddrinfo(a, None, socket.AF_UNSPEC, socket.SOCK_STREAM)]
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.11/socket.py", line 962, in getaddrinfo
    for res in _socket.getaddrinfo(host, port, family, type, proto, flags):
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
socket.gaierror: [Errno -2] Name or service not known

A workaround is possible (using HostFilterPolicy), but since cqlsh uses WhiteListRoundRobinPolicy, we need to fix it.

[sylwiaszunejko]

@kbr-scylla
This issue have two parts:

• firstly WhiteListRoundRobinPolicy does not support unix domain sockets (this PR address that: Add support for unix domain sockets to WhiteListRoundRobinPolicy #287)
• secondly cqlsh does not work not only because WhiteListRoundRobinPolicy, but also because in cqlsh there is no mention that this contact point is UnixSocketEndPoint as it should be, you can see that on that example:

from cassandra.cluster import Cluster
from cassandra.connection import UnixSocketEndPoint
from cassandra.policies import HostFilterPolicy, RoundRobinPolicy

socket = "<node's workdir>/cql.m"
cluster = Cluster([UnixSocketEndPoint(socket)],
                  # Driver tries to connect to other nodes in the cluster, so we need to filter them out.
                  load_balancing_policy=HostFilterPolicy(RoundRobinPolicy(), lambda h: h.address == socket))
session = cluster.connect()

with WhiteListRoundRobinPolicy it should look like that:

from cassandra.cluster import Cluster
from cassandra.connection import UnixSocketEndPoint
from cassandra.policies import HostFilterPolicy, RoundRobinPolicy

socket = "<node's workdir>/cql.m"
cluster = Cluster([UnixSocketEndPoint(socket)],
                  # Driver tries to connect to other nodes in the cluster, so we need to filter them out.
                  load_balancing_policy=WhiteListRoundRobinPolicy([UnixSocketEndPoint(socket)])
session = cluster.connect()

so there is a need to introduce a way to cqlsh to recognize if hostname is a unix domain socket and if it is it should wrap self.hostname (here: https://github.com/scylladb/scylla-cqlsh/blob/426fa0eaa72b7d46115188b1af9cdd983b923241/bin/cqlsh.py#L488,L491) in UnixSocketEndPoint.

[sylwiaszunejko]

@kbr-scylla @avelanarius I addressed second part of this issue in scylladb/scylla-cqlsh#67