Disable the ZHA bellows UART thread when connecting to a TCP coordinator (#88202)

Disable the bellows UART thread when connecting to a TCP coordinator

homeassistant/components/zha/core/const.py

@@ -139,6 +139,7 @@ CONF_ENABLE_QUIRKS = "enable_quirks"
 CONF_FLOWCONTROL = "flow_control"
 CONF_RADIO_TYPE = "radio_type"
 CONF_USB_PATH = "usb_path"
+CONF_USE_THREAD = "use_thread"
 CONF_ZIGPY = "zigpy_config"
 
 CONF_CONSIDER_UNAVAILABLE_MAINS = "consider_unavailable_mains"

homeassistant/components/zha/core/gateway.py

@@ -40,7 +40,9 @@ from .const import (
     ATTR_SIGNATURE,
     ATTR_TYPE,
     CONF_DATABASE,
+    CONF_DEVICE_PATH,
     CONF_RADIO_TYPE,
+    CONF_USE_THREAD,
     CONF_ZIGPY,
     DATA_ZHA,
     DATA_ZHA_BRIDGE_ID,
@@ -167,6 +169,15 @@ class ZHAGateway:
         app_config[CONF_DATABASE] = database
         app_config[CONF_DEVICE] = self.config_entry.data[CONF_DEVICE]
 
+        # The bellows UART thread sometimes propagates a cancellation into the main Core
+        # event loop, when a connection to a TCP coordinator fails in a specific way
+        if (
+            CONF_USE_THREAD not in app_config
+            and RadioType[radio_type] is RadioType.ezsp
+            and app_config[CONF_DEVICE][CONF_DEVICE_PATH].startswith("socket://")
+        ):
+            app_config[CONF_USE_THREAD] = False
+
         app_config = app_controller_cls.SCHEMA(app_config)
 
         for attempt in range(STARTUP_RETRIES):

tests/components/zha/test_gateway.py

@@ -287,3 +287,39 @@ async def test_gateway_initialize_failure_transient(
 
     # Initialization immediately stops and is retried after TransientConnectionError
     assert mock_new.call_count == 2
+
+
+@patch(
+    "homeassistant.components.zha.core.gateway.ZHAGateway.async_load_devices",
+    MagicMock(),
+)
+@patch(
+    "homeassistant.components.zha.core.gateway.ZHAGateway.async_load_groups",
+    MagicMock(),
+)
+@pytest.mark.parametrize(
+    ("device_path", "thread_state", "config_override"),
+    [
+        ("/dev/ttyUSB0", True, {}),
+        ("socket://192.168.1.123:9999", False, {}),
+        ("socket://192.168.1.123:9999", True, {"use_thread": True}),
+    ],
+)
+async def test_gateway_initialize_bellows_thread(
+    device_path, thread_state, config_override, hass, coordinator
+):
+    """Test ZHA disabling the UART thread when connecting to a TCP coordinator."""
+    zha_gateway = get_zha_gateway(hass)
+    assert zha_gateway is not None
+
+    zha_gateway.config_entry.data = dict(zha_gateway.config_entry.data)
+    zha_gateway.config_entry.data["device"]["path"] = device_path
+    zha_gateway._config.setdefault("zigpy_config", {}).update(config_override)
+
+    with patch(
+        "bellows.zigbee.application.ControllerApplication.new",
+        new=AsyncMock(),
+    ) as mock_new:
+        await zha_gateway.async_initialize()
+
+    assert mock_new.mock_calls[0].args[0]["use_thread"] is thread_state