Tolerate a restarting server after the native update

The first ping after `herdr update --handoff` was sent exactly once. A
server still binding its socket turned into "更新/热切换未完成" even
though the update had succeeded. Retry connection errors for up to 20
seconds before giving up; a healthy server still answers immediately.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

herdr_hub/restart.py

@@ -32,6 +32,29 @@ def request(socket_path: str, method: str, params: dict, timeout: float = 10) ->
     return result
 
 
+def ping_when_reachable(socket_path: str, *, timeout: float = 20) -> dict:
+    """Ping a server that may still be restarting right after an update.
+
+    The native updater can return while the new server is still binding the
+    socket; a single refused connection must not be reported as a failed
+    update. The first attempt is immediate so a healthy server adds no delay.
+    """
+    deadline: float | None = None
+    while True:
+        try:
+            return request(socket_path, "ping", {}, timeout=2)
+        except (OSError, ValueError, RuntimeError) as error:
+            now = time.monotonic()
+            if deadline is None:
+                deadline = now + timeout
+            elif now >= deadline:
+                raise RuntimeError(
+                    f"更新后 {timeout:g} 秒内无法连接 Herdr 服务器：{error}；"
+                    "请检查 Session 状态，不会自动重启"
+                ) from error
+            time.sleep(.2)
+
+
 def installed_version(binary: str) -> str:
     version = subprocess.check_output([binary, "--version"], text=True, timeout=10).strip()
     if not version.startswith("herdr ") or not version[6:].strip():
@@ -62,7 +85,7 @@ def update_and_handoff(binary: str, session: str) -> None:
     # Let the native updater perform its opt-in handoff, scoped to this session.
     subprocess.run([binary, "--session", session, "update", "--handoff"], check=True, timeout=300)
     version = installed_version(binary)
-    current = request(socket_path, "ping", {})
+    current = ping_when_reachable(socket_path)
     if current.get("version") != version:
         if binary_identity(binary) != identity_before:
             raise RuntimeError(

tests/test_restart.py

@@ -65,6 +65,33 @@ class HandoffTests(unittest.TestCase):
                 update_and_handoff("/fake/herdr", "named")
         self.assertEqual([c.args[1] for c in call.call_args_list].count("server.live_handoff"), 1)
 
+    def test_server_restarting_after_native_update_is_polled_not_failed(self):
+        old = {"version": "0.9.0", "capabilities": {"live_handoff": True}}
+        with patch("herdr_hub.restart.subprocess.check_output", return_value="status: running\nsocket: /tmp/fake.sock\n"), patch(
+            "herdr_hub.restart.request",
+            side_effect=[old, ConnectionRefusedError(), FileNotFoundError(), {"version": "0.9.1"}],
+        ) as call, patch("herdr_hub.restart.subprocess.run"), patch(
+            "herdr_hub.restart.installed_version", return_value="0.9.1"
+        ), patch("herdr_hub.restart.time.sleep"), patch(
+            "herdr_hub.restart.time.monotonic", side_effect=[0, 0.2]
+        ), contextlib.redirect_stdout(io.StringIO()):
+            update_and_handoff("/fake/herdr", "named")
+        self.assertEqual([c.args[1] for c in call.call_args_list], ["ping"] * 4)
+
+    def test_server_unreachable_after_update_is_reported_without_restart(self):
+        old = {"version": "0.9.0", "capabilities": {"live_handoff": True}}
+        with patch("herdr_hub.restart.subprocess.check_output", return_value="status: running\nsocket: /tmp/fake.sock\n"), patch(
+            "herdr_hub.restart.request",
+            side_effect=[old, ConnectionRefusedError(), ConnectionRefusedError()],
+        ) as call, patch("herdr_hub.restart.subprocess.run"), patch(
+            "herdr_hub.restart.installed_version", return_value="0.9.1"
+        ), patch("herdr_hub.restart.time.sleep"), patch(
+            "herdr_hub.restart.time.monotonic", side_effect=[0, 25]
+        ), contextlib.redirect_stdout(io.StringIO()):
+            with self.assertRaisesRegex(RuntimeError, "无法连接 Herdr 服务器"):
+                update_and_handoff("/fake/herdr", "named")
+        self.assertNotIn("server.live_handoff", [c.args[1] for c in call.call_args_list])
+
     def test_real_helper_updates_old_server_even_when_binary_already_latest(self):
         self._run_real_helper()