From 8b45a8b0d42b34eeca1e9fbf3b1da41a5808aee5 Mon Sep 17 00:00:00 2001
From: teknium1 <127238744+teknium1@users.noreply.github.com>
Date: Fri, 24 Jul 2026 12:53:15 -0700
Subject: [PATCH] fix(gateway): deregister transient reload label after
 one-shot job + test

Follow-up on the #69500 salvage: 'launchctl submit' jobs remain
registered with launchd after they exit, so every plist reload leaked
one dead '<label>.reload.<pid>.<ts>' label. The helper script now ends
with 'launchctl remove' of its own transient label, and the recovery
test asserts the self-removal is present.
---
 hermes_cli/gateway.py                    | 11 +++++++++--
 tests/hermes_cli/test_gateway_service.py |  4 ++++
 2 files changed, 13 insertions(+), 2 deletions(-)

diff --git a/hermes_cli/gateway.py b/hermes_cli/gateway.py
index abac5fd0758a..bfb6ada9d596 100644
--- a/hermes_cli/gateway.py
+++ b/hermes_cli/gateway.py
@@ -4134,6 +4134,9 @@ def refresh_launchd_plist_if_needed() -> bool:
         # gateway is still draining (default agent.restart_drain_timeout=180s),
         # so a fixed ~10s window is too short — bound by that budget instead.
         _reload_budget = int(max(30.0, _get_restart_drain_timeout()))
+        # Label for the transient one-shot job (see `launchctl submit` below).
+        # Unique per reload so concurrent/repeated reloads never collide.
+        submit_label = f"{label}.reload.{os.getpid()}.{int(time.time())}"
         reload_script = (
             f"sleep 2; "
             f"launchctl bootout {shlex.quote(target)} 2>/dev/null; "
@@ -4148,7 +4151,12 @@ def refresh_launchd_plist_if_needed() -> bool:
             f"done; "
             f"if ! launchctl list {shlex.quote(label)} >/dev/null 2>&1; then "
             f"  echo \"[$(date '+%Y-%m-%d %H:%M:%S %z')] FAILED launchd reload for {shlex.quote(target)} — service NOT registered after {_reload_budget}s of retries\" >> {shlex.quote(str(reload_log_path))}; "
-            f"fi"
+            f"fi; "
+            # Submitted jobs stay registered with launchd after the script
+            # exits; without this, every reload leaks one dead label. Removing
+            # our own label is the documented way to end a one-shot submit job
+            # (it SIGTERMs the job, but this is the final statement anyway).
+            f"launchctl remove {shlex.quote(submit_label)} 2>/dev/null"
         )
         try:
             # Spawn the reload helper via `launchctl submit` (a transient
@@ -4162,7 +4170,6 @@ def refresh_launchd_plist_if_needed() -> bool:
             # `launchctl submit` creates a wholly independent transient launchd
             # job that launchd manages separately from the gateway, so bootout
             # of the gateway job cannot reach the helper.
-            submit_label = f"{label}.reload.{os.getpid()}.{int(time.time())}"
             subprocess.Popen(
                 [
                     "launchctl", "submit",
diff --git a/tests/hermes_cli/test_gateway_service.py b/tests/hermes_cli/test_gateway_service.py
index 803c6a8793f1..7a7911687fd0 100644
--- a/tests/hermes_cli/test_gateway_service.py
+++ b/tests/hermes_cli/test_gateway_service.py
@@ -752,6 +752,10 @@ class TestLaunchdServiceRecovery:
         script = cmd[bash_idx + 2]
         assert "bootout" in script and "bootstrap" in script
         assert str(plist_path) in script
+        # The one-shot job must deregister its own transient label at the end,
+        # otherwise every reload leaks a dead label in launchd.
+        submit_label = cmd[cmd.index("-l") + 1]
+        assert f"launchctl remove {submit_label}" in script
 
     def test_refresh_uses_direct_reload_when_not_inside_gateway_tree(self, tmp_path, monkeypatch):
         """Normal CLI-initiated refresh (outside the service tree) keeps the