control-connection: restore local Host.listen_address during topology refresh

[dkropachev]

Problem

Since a0cde2e019, control-connection topology refresh no longer assigns system.local.listen_address to the local Host.

The attribute remains part of the public Host contract:

python-driver/cassandra/pool.py

Lines 95 to 113 in 83bdd46

	listen_address = None
	"""
	listen address configured for the node, *if available*:
	'system.local.listen_address'
	This is only available in the ``system.local`` table for newer versions of Cassandra. It is also not
	queried if :attr:`~.Cluster.token_metadata_enabled` is ``False``. Usually the same as ``broadcast_address``
	unless configured differently in cassandra.yaml.
	"""
	listen_port = None
	"""
	listen port configured for the node, *if available*:
	'system.local.listen_port'
	This is only available in the ``system.local`` table for newer versions of Cassandra. It is also not
	queried if :attr:`~.Cluster.token_metadata_enabled` is ``False``.

Current refresh copies broadcast fields but omits local listen metadata:

python-driver/cassandra/cluster.py

Lines 4209 to 4225 in 83bdd46

	host.host_id = host_id
	host.broadcast_address = _NodeInfo.get_broadcast_address(row)
	host.broadcast_port = _NodeInfo.get_broadcast_port(row)
	host.broadcast_rpc_address = _NodeInfo.get_broadcast_rpc_address(row)
	host.broadcast_rpc_port = _NodeInfo.get_broadcast_rpc_port(row)
	host.release_version = row.get("release_version")
	host.dse_version = row.get("dse_version")
	host.dse_workload = row.get("workload")
	host.dse_workloads = row.get("workloads")
	if row is local_row:
	connection._control_connection_host_id = host_id
	tokens = row.get("tokens", None)
	if partitioner and tokens and self._token_meta_enabled:
	token_map[host] = tokens
	self._cluster.metadata.update_host(host, old_endpoint=endpoint)

Code before the regression populated it:

python-driver/cassandra/cluster.py

Lines 3819 to 3843 in a3adf76

	host = self._cluster.metadata.get_host(connection.original_endpoint)
	if host:
	datacenter = local_row.get("data_center")
	rack = local_row.get("rack")
	self._update_location_info(host, datacenter, rack)
	# support the use case of connecting only with public address
	if isinstance(self._cluster.endpoint_factory, SniEndPointFactory):
	new_endpoint = self._cluster.endpoint_factory.create(local_row)
	if new_endpoint.address:
	host.endpoint = new_endpoint
	host.host_id = local_row.get("host_id")
	found_host_ids.add(host.host_id)
	found_endpoints.add(host.endpoint)
	host.listen_address = local_row.get("listen_address")
	host.listen_port = local_row.get("listen_port")
	host.broadcast_address = _NodeInfo.get_broadcast_address(local_row)
	host.broadcast_port = _NodeInfo.get_broadcast_port(local_row)
	host.broadcast_rpc_address = _NodeInfo.get_broadcast_rpc_address(local_row)
	host.broadcast_rpc_port = _NodeInfo.get_broadcast_rpc_port(local_row)

Regression commit:

a0cde2e

Impact

When CQL RPC address changes but REST/listen address stays unchanged, Host.listen_address remains None. Consumers fall back to the relocated RPC endpoint and contact the wrong REST address.

Tested with ScyllaDB 2026.2.6 (c06236b53803b96e7d2d21b4fad144a9eef2c3ba) and python-driver PR #944 (83bdd4605468f73bfcee80588aa8f1be5f87d9ef). Driver 3.29.7 passes; PR #944 fails.

Related test from testpy

https://github.com/scylladb/scylladb/blob/c06236b53803b96e7d2d21b4fad144a9eef2c3ba/test/cluster/test_change_rpc_address.py#L40-L89

The test changes only the CQL RPC address, then obtains the REST address from Host.listen_address:

https://github.com/scylladb/scylladb/blob/c06236b53803b96e7d2d21b4fad144a9eef2c3ba/test/pylib/rest_client.py#L775-L781

Actual: Host.listen_address is None; REST request targets the relocated RPC IP and fails with ConnectionRefusedError.

Expected: local Host.listen_address matches system.local.listen_address after topology refresh.