Files
hermes-agent/tests/hermes_cli/test_cross_profile_kill_refusal.py
Teknium ca16cafee4 test(guard): block spawning a real gateway runtime from tests
Tests that exercise the dashboard's gateway-restart path can end up
spawning a REAL `python -m hermes_cli.main gateway restart` child when
the spawn seam is not intercepted. `_spawn_hermes_action` launches it
with start_new_session=True, so it outlives the pytest worker; the
child inherits the pytest-tmp HERMES_HOME, which is not a profile and
hashes to no service suffix, so `get_service_name()` resolves the
DEVELOPER's `hermes-gateway` unit, `systemd_restart` restarts the live
gateway, and without systemd the fallback runs `run_gateway()`
in-process forever and squats the webhook port.

Live repro on this machine (origin/main): an unintercepted spawn of
["gateway", "restart"] from a test restarted the production gateway
(MainPID 136820 -> 1689090, NRestarts=1). On 2026-09-03 a sibling
refactor moved `_spawn_hermes_action` from the `hermes_cli.web_server`
facade to `hermes_cli.web_server_gateway` ~10 minutes before the tests
were repointed; runs in that window patched a name production never
read and left 39 orphans alive for six days.

The live-system guard now rejects any subprocess primitive whose
command line the canonical matcher (`gateway.status.
_gateway_command_subcommand`) classifies as `gateway run|start|restart`.
Argv substrings are never consulted, so `gateway status`, `gateway
--help`, `hermes_cli.main serve`, etc. pass through. Three files that
deliberately spawn and reap a stub child with a gateway-shaped argv
(flock holders, sleep sleepers with an argv tail) opt out with the new
`spawns_gateway_lookalike` marker, which lifts only this check and keeps
os.kill guarded. Two canary tests pin the block and the pass-through.
2026-09-09 09:19:54 -07:00

222 lines
8.4 KiB
Python

"""Cross-profile kill refusal regression tests (#89315).
A poisoned/contaminated ``gateway.pid`` inside one profile's HERMES_HOME can
truthfully name ANOTHER profile's live gateway (its ``hermes_home`` stamp
records the real owner). ``gateway stop`` / the restart force-kill escalation
/ ``profile delete`` must refuse to signal such a PID instead of starting the
mutual cross-profile SIGTERM restart loop from the issue report.
These tests exercise the REAL code paths against real PID files, a real
flock-held gateway lock, and a real dummy child process — no mocks of the
code under test.
"""
import json
import os
import subprocess
import sys
import time
from pathlib import Path
import pytest
from gateway.status import recorded_gateway_home_conflicts
def _spawn_gateway_lookalike(bin_dir: Path, lock_path: Path) -> subprocess.Popen:
"""Real child process whose argv matches the gateway runtime matcher."""
bin_dir.mkdir(parents=True, exist_ok=True)
lock_path.parent.mkdir(parents=True, exist_ok=True)
script = bin_dir / "hermes"
if sys.platform == "win32":
body = "import time\ntime.sleep(120)\n"
else:
body = (
"import fcntl, time\n"
f"fh = open({str(lock_path)!r}, 'a+')\n"
"fcntl.flock(fh, fcntl.LOCK_EX | fcntl.LOCK_NB)\n"
"time.sleep(120)\n"
)
script.write_text(f"#!{sys.executable}\n{body}", encoding="utf-8")
if sys.platform != "win32":
script.chmod(0o755)
cmd = [str(script), "gateway", "run"]
else:
cmd = [sys.executable, str(script), "gateway", "run"]
proc = subprocess.Popen(
cmd, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL
)
deadline = time.monotonic() + 10.0
while time.monotonic() < deadline and not lock_path.exists():
if proc.poll() is not None:
raise RuntimeError("gateway lookalike died at startup")
time.sleep(0.05)
return proc
def _pid_record(proc: subprocess.Popen, script: Path, owner_home: Path) -> dict:
from gateway.status import get_process_start_time
return {
"pid": proc.pid,
"kind": "hermes-gateway",
"argv": [str(script), "gateway", "run"],
"start_time": get_process_start_time(proc.pid),
"hermes_home": str(owner_home),
}
class TestRecordedGatewayHomeConflicts:
def test_conflicting_home_detected(self, tmp_path, monkeypatch):
monkeypatch.setenv("HERMES_HOME", str(tmp_path / "profiles" / "tim"))
record = {"pid": 1, "hermes_home": str(tmp_path)}
assert recorded_gateway_home_conflicts(record) is True
def test_same_home_accepted(self, tmp_path, monkeypatch):
home = tmp_path / "profiles" / "tim"
monkeypatch.setenv("HERMES_HOME", str(home))
record = {"pid": 1, "hermes_home": str(home)}
assert recorded_gateway_home_conflicts(record) is False
def test_legacy_record_without_home_proves_nothing(self, tmp_path, monkeypatch):
monkeypatch.setenv("HERMES_HOME", str(tmp_path))
assert recorded_gateway_home_conflicts({"pid": 1}) is False
assert recorded_gateway_home_conflicts(None) is False
assert recorded_gateway_home_conflicts({"pid": 1, "hermes_home": " "}) is False
def test_expected_home_override(self, tmp_path):
target = tmp_path / "profiles" / "tim"
record = {"pid": 1, "hermes_home": str(tmp_path)}
assert (
recorded_gateway_home_conflicts(record, expected_home=target) is True
)
assert (
recorded_gateway_home_conflicts(record, expected_home=tmp_path) is False
)
# The lookalike's argv ("<stub> gateway run") is what the guard's real-gateway spawn check matches;
# the child is a sleep stub the test kills in ``finally``, never a runtime.
@pytest.mark.spawns_gateway_lookalike
@pytest.mark.skipif(sys.platform == "win32", reason="POSIX flock harness")
class TestCrossProfileStopRefusal:
def test_stop_profile_gateway_refuses_other_profiles_pid(
self, tmp_path, monkeypatch
):
"""Profile B's ``gateway stop`` must not SIGTERM profile A's gateway.
On main this path is already safe upstream of any guard:
``get_running_pid()`` filters a pid record owned by another profile
(and unlinks the poisoned pid file) before ``stop_profile_gateway``
ever sees a pid — so the contract here is "returns False, other
profile's process untouched, poisoned pid file gone", not a printed
refusal.
"""
root_home = tmp_path / "root-home"
tim_home = tmp_path / "root-home" / "profiles" / "tim"
tim_home.mkdir(parents=True)
monkeypatch.setenv("HERMES_HOME", str(tim_home))
proc = _spawn_gateway_lookalike(
tmp_path / "bin", tim_home / "gateway.lock"
)
try:
record = _pid_record(proc, tmp_path / "bin" / "hermes", root_home)
(tim_home / "gateway.pid").write_text(json.dumps(record))
from hermes_cli import gateway as gateway_cli
assert gateway_cli.stop_profile_gateway() is False
assert not (tim_home / "gateway.pid").exists(), (
"poisoned cross-profile pid file should have been unlinked"
)
time.sleep(0.5)
assert proc.poll() is None, (
"cross-profile SIGTERM fired: profile A's gateway was killed"
)
finally:
proc.kill()
proc.wait(timeout=10)
def test_stop_profile_gateway_still_stops_own_gateway(
self, tmp_path, monkeypatch
):
"""Same-home records keep stopping normally (no false refusal)."""
tim_home = tmp_path / "profiles" / "tim"
tim_home.mkdir(parents=True)
monkeypatch.setenv("HERMES_HOME", str(tim_home))
proc = _spawn_gateway_lookalike(
tmp_path / "bin", tim_home / "gateway.lock"
)
try:
record = _pid_record(proc, tmp_path / "bin" / "hermes", tim_home)
(tim_home / "gateway.pid").write_text(json.dumps(record))
from hermes_cli import gateway as gateway_cli
assert gateway_cli.stop_profile_gateway() is True
deadline = time.monotonic() + 15.0
while time.monotonic() < deadline and proc.poll() is None:
time.sleep(0.1)
assert proc.poll() is not None, "own gateway was not stopped"
finally:
if proc.poll() is None:
proc.kill()
proc.wait(timeout=10)
@pytest.mark.spawns_gateway_lookalike
@pytest.mark.skipif(sys.platform == "win32", reason="POSIX flock harness")
class TestProfileDeleteStopRefusal:
def test_stop_gateway_process_refuses_other_profiles_pid(
self, tmp_path, capsys
):
"""``profile delete`` must not kill a gateway owned by another home."""
root_home = tmp_path / "root-home"
tim_home = root_home / "profiles" / "tim"
tim_home.mkdir(parents=True)
proc = _spawn_gateway_lookalike(
tmp_path / "bin", tim_home / "gateway.lock"
)
try:
record = _pid_record(proc, tmp_path / "bin" / "hermes", root_home)
(tim_home / "gateway.pid").write_text(json.dumps(record))
from hermes_cli.profiles import _stop_gateway_process
_stop_gateway_process(tim_home)
out = capsys.readouterr().out
assert "Refusing to stop" in out
time.sleep(0.5)
assert proc.poll() is None, (
"profile delete killed another profile's gateway"
)
finally:
proc.kill()
proc.wait(timeout=10)
def test_stop_gateway_process_still_stops_own_gateway(self, tmp_path):
tim_home = tmp_path / "profiles" / "tim"
tim_home.mkdir(parents=True)
proc = _spawn_gateway_lookalike(
tmp_path / "bin", tim_home / "gateway.lock"
)
try:
record = _pid_record(proc, tmp_path / "bin" / "hermes", tim_home)
(tim_home / "gateway.pid").write_text(json.dumps(record))
from hermes_cli.profiles import _stop_gateway_process
_stop_gateway_process(tim_home)
deadline = time.monotonic() + 15.0
while time.monotonic() < deadline and proc.poll() is None:
time.sleep(0.1)
assert proc.poll() is not None, "own gateway was not stopped"
finally:
if proc.poll() is None:
proc.kill()
proc.wait(timeout=10)