From fd96e138b64aa9f2266e971700df7bfd63cc41d5 Mon Sep 17 00:00:00 2001
From: web3blind <264741654+web3blind@users.noreply.github.com>
Date: Tue, 21 Jul 2026 07:29:52 +0000
Subject: [PATCH] fix(gateway): hard-exit CLI runner after graceful teardown

---
 hermes_cli/gateway.py                         | 22 ++++-
 .../hermes_cli/test_gateway_run_hard_exit.py  | 87 +++++++++++++++++++
 2 files changed, 107 insertions(+), 2 deletions(-)
 create mode 100644 tests/hermes_cli/test_gateway_run_hard_exit.py

diff --git a/hermes_cli/gateway.py b/hermes_cli/gateway.py
index 69f6464e806b..7f81072515f7 100644
--- a/hermes_cli/gateway.py
+++ b/hermes_cli/gateway.py
@@ -4975,6 +4975,17 @@ def run_gateway(verbose: int = 0, quiet: bool = False, replace: bool = False, fo
     except Exception as _be:
         logger.debug("respawn-storm breaker check failed (non-fatal): %s", _be)
 
+    def _hard_exit_after_gateway_teardown(code: int) -> None:
+        # ``hermes gateway run`` enters through this CLI wrapper, not through
+        # ``gateway.run.main()``.  Mirror that module's wedge-proof exit path:
+        # once start_gateway() has completed graceful teardown, bypass Python
+        # finalization so non-daemon worker threads (notably in-flight cron
+        # ThreadPoolExecutor jobs) cannot keep the old gateway alive and delay a
+        # service-managed /restart by minutes.
+        from gateway.run import _exit_after_graceful_shutdown
+
+        _exit_after_graceful_shutdown(code)
+
     success = False
     try:
         success = asyncio.run(start_gateway(replace=replace, verbosity=verbosity))
@@ -4994,7 +5005,13 @@ def run_gateway(verbose: int = 0, quiet: bool = False, replace: bool = False, fo
             code=getattr(e, "code", None),
             traceback=_traceback.format_exc(),
         )
-        raise
+        if e.code is None:
+            _code = 0
+        elif isinstance(e.code, int):
+            _code = e.code
+        else:
+            _code = 1
+        _hard_exit_after_gateway_teardown(_code)
     except BaseException as e:
         # Absolutely everything else: Exception, asyncio.CancelledError,
         # even exotic BaseException subclasses. We want the cause logged.
@@ -5007,8 +5024,9 @@ def run_gateway(verbose: int = 0, quiet: bool = False, replace: bool = False, fo
         raise
     if not success:
         _exit_diag("gateway.exit_nonzero")
-        sys.exit(1)
+        _hard_exit_after_gateway_teardown(1)
     _exit_diag("gateway.exit_clean")
+    _hard_exit_after_gateway_teardown(0)
 
 
 # =============================================================================
diff --git a/tests/hermes_cli/test_gateway_run_hard_exit.py b/tests/hermes_cli/test_gateway_run_hard_exit.py
new file mode 100644
index 000000000000..be69b0915e77
--- /dev/null
+++ b/tests/hermes_cli/test_gateway_run_hard_exit.py
@@ -0,0 +1,87 @@
+"""Regression tests for CLI gateway run exit behavior.
+
+``hermes gateway run`` enters through hermes_cli.gateway, not gateway.run.main().
+After graceful teardown it must use the same hard-exit backstop as gateway.run.main()
+so Python finalization does not wait on non-daemon worker threads (for example
+in-flight cron ThreadPoolExecutor jobs) and delay service-managed restarts.
+"""
+
+from __future__ import annotations
+
+import types
+
+import pytest
+
+
+class _HardExitObserved(BaseException):
+    def __init__(self, code: int):
+        super().__init__(code)
+        self.code = code
+
+
+def _prepare(monkeypatch):
+    import hermes_cli.gateway as gateway_cli
+    import gateway.run as gateway_run
+
+    monkeypatch.setattr(gateway_cli, "_guard_official_docker_root_gateway", lambda: None)
+    monkeypatch.setattr(gateway_cli, "_guard_named_profile_under_multiplexer", lambda force=False: None)
+    monkeypatch.setattr(gateway_cli, "_guard_supervised_gateway_conflict", lambda force=False: None)
+    monkeypatch.setattr(gateway_cli, "_guard_existing_gateway_process_conflict", lambda replace=False: None)
+    monkeypatch.setattr(gateway_cli, "supports_systemd_services", lambda: False)
+    monkeypatch.setattr(gateway_cli.sys, "stdin", types.SimpleNamespace(isatty=lambda: False))
+    monkeypatch.setenv("HERMES_GATEWAY_EXIT_DIAG", "0")
+
+    async def _start_gateway(*args, **kwargs):  # pragma: no cover - never awaited by fake run
+        return True
+
+    def _hard_exit(code: int) -> None:
+        raise _HardExitObserved(code)
+
+    monkeypatch.setattr(gateway_run, "start_gateway", _start_gateway)
+    monkeypatch.setattr(gateway_run, "_exit_after_graceful_shutdown", _hard_exit)
+    return gateway_cli
+
+
+def test_run_gateway_hard_exits_after_clean_return(monkeypatch):
+    gateway_cli = _prepare(monkeypatch)
+
+    def _fake_run(coro):
+        coro.close()
+        return True
+
+    monkeypatch.setattr(gateway_cli.asyncio, "run", _fake_run)
+
+    with pytest.raises(_HardExitObserved) as excinfo:
+        gateway_cli.run_gateway()
+
+    assert excinfo.value.code == 0
+
+
+def test_run_gateway_hard_exits_after_service_restart_systemexit(monkeypatch):
+    gateway_cli = _prepare(monkeypatch)
+
+    def _fake_run(coro):
+        coro.close()
+        raise SystemExit(75)
+
+    monkeypatch.setattr(gateway_cli.asyncio, "run", _fake_run)
+
+    with pytest.raises(_HardExitObserved) as excinfo:
+        gateway_cli.run_gateway()
+
+    assert excinfo.value.code == 75
+
+
+def test_run_gateway_hard_exits_after_failed_return(monkeypatch):
+    gateway_cli = _prepare(monkeypatch)
+
+    def _fake_run(coro):
+        coro.close()
+        return False
+
+    monkeypatch.setattr(gateway_cli.asyncio, "run", _fake_run)
+
+    with pytest.raises(_HardExitObserved) as excinfo:
+        gateway_cli.run_gateway()
+
+    assert excinfo.value.code == 1