From 2489f15040b3234ec637365e325bfc704ebd73db Mon Sep 17 00:00:00 2001
From: salch-cred <141555468+salch-cred@users.noreply.github.com>
Date: Wed, 9 Sep 2026 20:26:04 +0530
Subject: [PATCH] fix(serve): ensure parent-death watchdog detects exited
 Desktop parent (#80204)

---
 hermes_cli/web_server_lifecycle.py            | 19 ++++--
 .../hermes_cli/test_serve_parent_watchdog.py  | 63 +++++++++++++++++++
 2 files changed, 76 insertions(+), 6 deletions(-)

diff --git a/hermes_cli/web_server_lifecycle.py b/hermes_cli/web_server_lifecycle.py
index 14b554628c..ccb26d425c 100644
--- a/hermes_cli/web_server_lifecycle.py
+++ b/hermes_cli/web_server_lifecycle.py
@@ -75,7 +75,7 @@ def _process_start_marker(pid: int) -> str:
     marker = result.stdout.strip()
     if result.returncode == 0 and marker:
         return f"ps:{marker}"
-    if result.returncode == 1 and not marker:
+    if result.returncode != 0 and (not marker or "no such process" in result.stderr.lower() or "not found" in result.stderr.lower()):
         raise ProcessLookupError(pid)
     raise OSError(f"ps could not inspect PID {pid}: {result.stderr.strip()}")
 
@@ -278,12 +278,19 @@ def _is_serve_orphaned(
     try:
         if expected_start_marker is not None:
             probe = process_start_marker or _process_start_marker
-            actual_marker = probe(int(desktop_pid))
-            if _parent_start_markers_match(actual_marker, expected_start_marker):
-                return False
-            if _parent_start_marker_mismatch_is_conclusive(actual_marker, expected_start_marker):
+            try:
+                actual_marker = probe(int(desktop_pid))
+            except ProcessLookupError:
                 return True
-            # Inconclusive marker: degrade to PID liveness instead of exiting.
+            except Exception:
+                actual_marker = None
+
+            if actual_marker is not None:
+                if _parent_start_markers_match(actual_marker, expected_start_marker):
+                    return False
+                if _parent_start_marker_mismatch_is_conclusive(actual_marker, expected_start_marker):
+                    return True
+                # Inconclusive marker: degrade to PID liveness instead of exiting.
 
         if pid_exists is None:
             from gateway.status import _pid_exists
diff --git a/tests/hermes_cli/test_serve_parent_watchdog.py b/tests/hermes_cli/test_serve_parent_watchdog.py
index 0b146a2a80..1f23d02b7d 100644
--- a/tests/hermes_cli/test_serve_parent_watchdog.py
+++ b/tests/hermes_cli/test_serve_parent_watchdog.py
@@ -175,3 +175,66 @@ def test_parent_watchdog_warns_when_disarmed_by_unusable_marker(monkeypatch, cap
 class _NoThread:
     def start(self):
         raise AssertionError("watchdog thread must not start")
+
+
+def test_parent_watchdog_detects_dead_parent_when_start_marker_probe_raises_oserror():
+    """#80204: when marker probe fails with OSError on a dead parent, watchdog
+    must degrade to PID liveness check and reap the orphan, not return False."""
+    def broken_marker_probe(_pid: int) -> str:
+        raise OSError("ps could not inspect PID 4242: ps: 4242: No such process")
+
+    # Parent is dead: pid_exists returns False -> must return True (orphaned)
+    assert (
+        _is_serve_orphaned(
+            4242,
+            "ps:Thu Aug 20 22:33:11 2026",
+            pid_exists=lambda _pid: False,
+            process_start_marker=broken_marker_probe,
+        )
+        is True
+    )
+
+    # Parent is still alive: pid_exists returns True -> fail-safe False
+    assert (
+        _is_serve_orphaned(
+            4242,
+            "ps:Thu Aug 20 22:33:11 2026",
+            pid_exists=lambda _pid: True,
+            process_start_marker=broken_marker_probe,
+        )
+        is False
+    )
+
+
+def test_parent_watchdog_detects_dead_parent_on_process_lookup_error():
+    """#80204: ProcessLookupError from probe immediately signals parent is gone."""
+    def lookup_error_probe(pid: int) -> str:
+        raise ProcessLookupError(pid)
+
+    assert (
+        _is_serve_orphaned(
+            4242,
+            "ps:Thu Aug 20 22:33:11 2026",
+            pid_exists=lambda _pid: True,
+            process_start_marker=lookup_error_probe,
+        )
+        is True
+    )
+
+
+def test_ps_process_start_marker_raises_process_lookup_error_on_missing_process(monkeypatch):
+    """#80204: _process_start_marker raises ProcessLookupError when ps exits non-zero with no marker."""
+    import subprocess
+    import pytest
+    from hermes_cli import web_server_lifecycle
+
+    def fake_run(*args, **kwargs):
+        return subprocess.CompletedProcess(args=args, returncode=1, stdout="", stderr="ps: 4242: No such process")
+
+    monkeypatch.setattr(subprocess, "run", fake_run)
+    monkeypatch.setattr(web_server_lifecycle.sys, "platform", "darwin")
+    monkeypatch.setattr(web_server_lifecycle.os, "name", "posix")
+
+    with pytest.raises(ProcessLookupError):
+        web_server_lifecycle._process_start_marker(4242)
+