fix(gateway): write restart markers atomically and fix Windows lock collisions

tests/gateway/test_restart_notification.py

@@ -113,6 +113,36 @@ async def test_restart_command_preserves_thread_id(tmp_path, monkeypatch):
     assert data["thread_id"] == "topic_7"
 
 
+@pytest.mark.asyncio
+async def test_restart_command_uses_atomic_json_writes_for_marker_files(tmp_path, monkeypatch):
+    monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path)
+
+    calls = []
+
+    def _fake_atomic_json_write(path, payload, **kwargs):
+        calls.append((Path(path).name, payload, kwargs))
+
+    monkeypatch.setattr(gateway_run, "atomic_json_write", _fake_atomic_json_write)
+
+    runner, _adapter = make_restart_runner()
+    runner.request_restart = MagicMock(return_value=True)
+
+    source = make_restart_source(chat_id="42")
+    event = MessageEvent(
+        text="/restart",
+        message_type=MessageType.TEXT,
+        source=source,
+        message_id="m1",
+    )
+
+    await runner._handle_restart_command(event)
+
+    names = [name for name, _payload, _kwargs in calls]
+    assert names == [".restart_notify.json", ".restart_last_processed.json"]
+    assert calls[0][1]["chat_id"] == "42"
+    assert calls[1][1]["platform"] == "telegram"
+
+
 # ── _send_restart_notification ───────────────────────────────────────────
 
 

tests/gateway/test_restart_resume_pending.py

@@ -999,3 +999,65 @@ class TestStuckLoopEscalation:
 
         assert store._entries[entry.session_key].resume_pending is False
         assert not counts_file.exists()
+
+    def test_increment_restart_failure_counts_uses_atomic_json_write(
+        self, tmp_path, monkeypatch
+    ):
+        from gateway.run import GatewayRunner
+
+        source = _make_source()
+        session_key = _make_store(tmp_path).get_or_create_session(source).session_key
+
+        monkeypatch.setattr("gateway.run._hermes_home", tmp_path)
+        calls = []
+
+        def _fake_atomic_json_write(path, payload, **kwargs):
+            calls.append((path, payload, kwargs))
+
+        monkeypatch.setattr("gateway.run.atomic_json_write", _fake_atomic_json_write)
+
+        runner = object.__new__(GatewayRunner)
+        runner._increment_restart_failure_counts({session_key})
+
+        assert calls == [
+            (
+                tmp_path / ".restart_failure_counts",
+                {session_key: 1},
+                {"indent": None},
+            )
+        ]
+
+    def test_clear_restart_failure_count_uses_atomic_json_write_when_entries_remain(
+        self, tmp_path, monkeypatch
+    ):
+        import json
+
+        from gateway.run import GatewayRunner
+
+        source = _make_source()
+        session_key = _make_store(tmp_path).get_or_create_session(source).session_key
+        other_key = "agent:main:telegram:dm:other"
+        counts_file = tmp_path / ".restart_failure_counts"
+        counts_file.write_text(
+            json.dumps({session_key: 2, other_key: 1}),
+            encoding="utf-8",
+        )
+
+        monkeypatch.setattr("gateway.run._hermes_home", tmp_path)
+        calls = []
+
+        def _fake_atomic_json_write(path, payload, **kwargs):
+            calls.append((path, payload, kwargs))
+
+        monkeypatch.setattr("gateway.run.atomic_json_write", _fake_atomic_json_write)
+
+        runner = object.__new__(GatewayRunner)
+        runner._clear_restart_failure_count(session_key)
+
+        assert calls == [
+            (
+                tmp_path / ".restart_failure_counts",
+                {other_key: 1},
+                {"indent": None},
+            )
+        ]

tests/gateway/test_status.py

@@ -2,6 +2,7 @@
 
 import json
 import os
+from pathlib import Path
 from types import SimpleNamespace
 
 from gateway import status
@@ -245,6 +246,27 @@ class TestGatewayPidState:
 
 
 class TestGatewayRuntimeStatus:
+    def test_write_json_file_uses_atomic_json_write(self, tmp_path, monkeypatch):
+        monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+        calls = []
+
+        def _fake_atomic_json_write(path, payload, **kwargs):
+            calls.append((Path(path), payload, kwargs))
+
+        monkeypatch.setattr(status, "atomic_json_write", _fake_atomic_json_write)
+
+        payload = {"gateway_state": "running"}
+        target = tmp_path / "gateway_state.json"
+        status._write_json_file(target, payload)
+
+        assert calls == [
+            (
+                target,
+                payload,
+                {"indent": None, "separators": (",", ":")},
+            )
+        ]
+
     def test_write_runtime_status_overwrites_stale_pid_on_restart(self, tmp_path, monkeypatch):
         """Regression: setdefault() preserved stale PID from previous process (#1631)."""
         monkeypatch.setenv("HERMES_HOME", str(tmp_path))
@@ -349,6 +371,35 @@ class TestTerminatePid:
 
 
 class TestScopedLocks:
+    def test_windows_file_lock_uses_high_offset(self, tmp_path, monkeypatch):
+        lock_path = tmp_path / "gateway.lock"
+        handle = open(lock_path, "a+", encoding="utf-8")
+        fd = handle.fileno()
+        calls = []
+
+        def fake_locking(fd, mode, size):
+            calls.append((fd, mode, size, handle.tell()))
+
+        monkeypatch.setattr(status, "_IS_WINDOWS", True)
+        monkeypatch.setattr(
+            status,
+            "msvcrt",
+            SimpleNamespace(LK_NBLCK=1, LK_UNLCK=2, locking=fake_locking),
+            raising=False,
+        )
+
+        try:
+            assert status._try_acquire_file_lock(handle) is True
+            status._release_file_lock(handle)
+        finally:
+            handle.close()
+
+        assert calls == [
+            (fd, 1, 1, status._WINDOWS_LOCK_OFFSET),
+            (fd, 2, 1, status._WINDOWS_LOCK_OFFSET),
+        ]
+        assert lock_path.read_text(encoding="utf-8") == "\n"
+
     def test_acquire_scoped_lock_rejects_live_other_process(self, tmp_path, monkeypatch):
         monkeypatch.setenv("HERMES_GATEWAY_LOCK_DIR", str(tmp_path / "locks"))
         lock_path = tmp_path / "locks" / "telegram-bot-token-2bb80d537b1da3e3.lock"