Fix slow connections by using multicast mDNS queries for cache sharing (#1394)

Author: bdraco

aioesphomeapi/host_resolver.py

@@ -11,7 +11,7 @@
 import socket
 from typing import TYPE_CHECKING, Any, cast
 
-from zeroconf import IPVersion
+from zeroconf import DNSQuestionType, IPVersion
 from zeroconf.asyncio import AsyncServiceInfo, AsyncZeroconf
 
 from .core import ResolveAPIError, ResolveTimeoutAPIError
@@ -66,7 +66,16 @@ async def _async_zeroconf_get_service_info(
 ) -> AsyncServiceInfo:
     info = _make_service_info_for_short_host(short_host)
     try:
-        await info.async_request(aiozc.zeroconf, int(timeout * 1000))
+        # Use QM (multicast) questions to ensure multicast responses that all
+        # zeroconf instances on the network can observe and cache. RFC 6762 Section
+        # 5.4 specifies that responders SHOULD send multicast responses to QU
+        # (unicast) questions if they haven't multicast recently (within 1/4 TTL),
+        # but ESP32 mDNS doesn't implement this - it always sends unicast to QU.
+        # By using QM, we ensure the dashboard observes responses when the CLI does
+        # resolution, preventing the dashboard's cache from going stale.
+        await info.async_request(
+            aiozc.zeroconf, int(timeout * 1000), question_type=DNSQuestionType.QM
+        )
     except Exception as exc:
         raise ResolveAPIError(
             f"Error resolving mDNS {short_host} via mDNS: {exc}"