fix: simplify GMS layout state and harden GPU-backed flows (#7006)

Signed-off-by: mohammedabdulwahhab <furkhan324@berkeley.edu>
Signed-off-by: Hannah Zhang <hannahz@nvidia.com>
Co-authored-by: mohammedabdulwahhab <furkhan324@berkeley.edu>
Co-authored-by: hhzhang16 <54051230+hhzhang16@users.noreply.github.com>

tests/fault_tolerance/gpu_memory_service/test_failover_lock.py → tests/gms/common/test_failover_lock.py

@@ -20,7 +20,6 @@ pytestmark = [
     pytest.mark.pre_merge,
     pytest.mark.unit,
     pytest.mark.gpu_0,
-    pytest.mark.fault_tolerance,
 ]
 
 

tests/gms/common/test_gms_harness.py

+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+from __future__ import annotations
+
+from types import SimpleNamespace
+
+import pytest
+
+from tests.gms.harness.gms import GMSServerProcess
+from tests.utils.managed_process import ManagedProcess
+
+pytestmark = [pytest.mark.pre_merge, pytest.mark.unit, pytest.mark.gpu_0]
+
+
+@pytest.fixture
+def request_stub():
+    return SimpleNamespace(node=SimpleNamespace(name="gms-harness"))
+
+
+def test_server_process_refuses_foreign_live_socket(monkeypatch, request_stub):
+    server = GMSServerProcess(request_stub, device=0, tag="weights")
+
+    monkeypatch.setattr("tests.gms.harness.gms.os.path.exists", lambda path: True)
+    monkeypatch.setattr("tests.gms.harness.gms._socket_has_live_gms", lambda path: True)
+
+    with pytest.raises(RuntimeError, match="already active"):
+        server.__enter__()
+
+
+def test_server_process_unlinks_only_stale_socket_on_exit(monkeypatch, request_stub):
+    server = GMSServerProcess(request_stub, device=0, tag="weights")
+    unlinked: list[str] = []
+
+    monkeypatch.setattr(
+        ManagedProcess,
+        "__exit__",
+        lambda self, exc_type, exc_val, exc_tb: False,
+    )
+    monkeypatch.setattr("tests.gms.harness.gms.os.path.exists", lambda path: True)
+    monkeypatch.setattr("tests.gms.harness.gms.os.unlink", unlinked.append)
+
+    monkeypatch.setattr("tests.gms.harness.gms._socket_has_live_gms", lambda path: True)
+    server.__exit__(None, None, None)
+    assert unlinked == []
+
+    monkeypatch.setattr(
+        "tests.gms.harness.gms._socket_has_live_gms", lambda path: False
+    )
+    server.__exit__(None, None, None)
+    assert unlinked == [server.socket_path]