feat(gateway): GATEWAY_MULTIPLEX_PROFILES env override for multiplex flag (#60589)

The connector now depends on the single multiplexed gateway for per-profile
relay routing, so hosted deployments need to FORCE multiplexing on regardless
of the image's config.yaml. gateway.multiplex_profiles was config.yaml-only,
which a user could leave unset or flip off.

Add GATEWAY_MULTIPLEX_PROFILES as a standard operator override on top of the
existing config key — the same 'config.yaml is canonical, env is the operator
override' pattern the Telegram/Signal require_mention bridges use:

  env (recognized token) > config.yaml (top-level or nested gateway.*) > False

- gateway/config.py: _env_multiplex_profiles_override() resolves the env var
  tri-state — recognized truthy/falsy token → bool; unset/blank/unrecognized
  → None (fall through to config). Blank is deliberately None, not False, so a
  provisioned-but-unpopulated Fly secret ('') can't shadow a config.yaml opt-in
  (the empty-secret trap). Wired into GatewayConfig.from_dict so every consumer
  (run.py, session.py via self.config) sees the resolved value.
- hermes_cli/gateway.py: the named-profile-start guard
  (_guard_named_profile_under_multiplexer) reads config.yaml directly, so it
  gets the SAME env precedence — otherwise env-forced multiplex would leave the
  guard blind and someone could start a conflicting per-profile gateway that
  double-binds a bot token. Env-forced-on trips the guard even with no
  config.yaml key; env-forced-off disables it over a config opt-in.

Tests: full 3-tier precedence in test_config.py (incl. the discriminating
env-overrides-config cases + the empty/whitespace/unrecognized fall-through
trap + resolver tri-state), mutation-verified (flipping precedence fails
exactly the two env-wins tests); guard env cases in test_multiplex_lifecycle.py.

Force-on is safe on a single-profile instance: session keys stay byte-identical
(agent:main) and the _run_agent wrapper installs the per-turn secret scope, so
the fail-closed get_secret() path is satisfied.
This commit is contained in:
Ben Barclay 2026-07-08 10:34:34 +10:00 committed by GitHub
parent d9a4b5a5e5
commit 75de0057bc
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 214 additions and 10 deletions

View file

@ -37,6 +37,43 @@ def _coerce_bool(value: Any, default: bool = True) -> bool:
return is_truthy_value(value, default=default)
# Recognized truthy / falsy tokens for the GATEWAY_MULTIPLEX_PROFILES operator
# override. Anything not in either set — and a blank/whitespace value — is
# treated as "unset" so it falls through to config.yaml rather than silently
# forcing the flag off.
_MULTIPLEX_TRUTHY_STRINGS = frozenset({"1", "true", "yes", "on"})
_MULTIPLEX_FALSY_STRINGS = frozenset({"0", "false", "no", "off"})
def _env_multiplex_profiles_override() -> "bool | None":
"""Resolve the GATEWAY_MULTIPLEX_PROFILES operator override.
Returns ``True``/``False`` when the env var is set to a recognized truthy/
falsy token, or ``None`` when it is unset, blank, or unrecognized in which
case the caller keeps the config.yaml value (env > config > default). Blank
is deliberately ``None``, not ``False``: a provisioned-but-unpopulated Fly
secret arrives as ``""`` and must NOT shadow a config.yaml opt-in.
"""
raw = os.getenv("GATEWAY_MULTIPLEX_PROFILES")
if raw is None:
return None
token = raw.strip().lower()
if not token:
return None
if token in _MULTIPLEX_TRUTHY_STRINGS:
return True
if token in _MULTIPLEX_FALSY_STRINGS:
return False
logger.warning(
"Ignoring unrecognized GATEWAY_MULTIPLEX_PROFILES=%r "
"(expected one of %s or %s); falling back to config.yaml.",
raw,
sorted(_MULTIPLEX_TRUTHY_STRINGS),
sorted(_MULTIPLEX_FALSY_STRINGS),
)
return None
def _coerce_float(value: Any, default: float) -> float:
"""Coerce numeric config values, falling back on malformed input."""
if value is None:
@ -837,6 +874,18 @@ class GatewayConfig:
# Also honor gateway.multiplex_profiles written by
# ``hermes config set gateway.multiplex_profiles true``.
multiplex_profiles = nested_gateway.get("multiplex_profiles")
# Operator override: GATEWAY_MULTIPLEX_PROFILES wins over config.yaml when
# set to a recognized value. Hosted deployments (Nous Portal / Fly) stamp
# it on the container so the single multiplexed gateway — which the
# connector now depends on for per-profile relay routing — is forced on at
# every boot regardless of the image's config.yaml, while self-hosted
# users keep setting gateway.multiplex_profiles in config.yaml. A blank or
# unrecognized env value falls through to config (the empty-secret trap:
# a provisioned-but-unpopulated Fly secret must not shadow config), so
# this is a genuine 3-tier chain: env > config.yaml > default False.
env_multiplex = _env_multiplex_profiles_override()
if env_multiplex is not None:
multiplex_profiles = env_multiplex
if "max_concurrent_sessions" in data:
max_concurrent_raw = data.get("max_concurrent_sessions")
max_concurrent_key = "max_concurrent_sessions"

View file

@ -4488,16 +4488,29 @@ def _guard_named_profile_under_multiplexer(force: bool = False) -> None:
if not pid or not _pid_exists(pid):
return
# (c) default config has multiplexing on
cfg_path = default_root / "config.yaml"
if not cfg_path.exists():
return
with open(cfg_path, encoding="utf-8") as f:
cfg = _yaml.safe_load(f) or {}
multiplex = bool(
cfg.get("multiplex_profiles")
or (cfg.get("gateway", {}) or {}).get("multiplex_profiles")
)
# (c) multiplexing is on for the default gateway. Precedence mirrors
# gateway.config: the GATEWAY_MULTIPLEX_PROFILES env override wins over
# config.yaml when set to a recognized value, so a hosted gateway that
# forces multiplex on via env (with no multiplex_profiles in config.yaml)
# still trips this guard. A blank/unrecognized env value falls through
# to config.yaml.
from gateway.config import _env_multiplex_profiles_override
env_multiplex = _env_multiplex_profiles_override()
if env_multiplex is False:
return # explicitly forced OFF by the operator env override
if env_multiplex is True:
multiplex = True
else:
cfg_path = default_root / "config.yaml"
if not cfg_path.exists():
return
with open(cfg_path, encoding="utf-8") as f:
cfg = _yaml.safe_load(f) or {}
multiplex = bool(
cfg.get("multiplex_profiles")
or (cfg.get("gateway", {}) or {}).get("multiplex_profiles")
)
if not multiplex:
return
except Exception:

View file

@ -1315,3 +1315,96 @@ class TestHomeChannelEnvOverrides:
home = config.platforms[platform].home_channel
assert home is not None, f"{platform.value}: home_channel should not be None"
assert (home.chat_id, home.name) == expected, platform.value
class TestMultiplexProfilesEnvOverride:
"""GATEWAY_MULTIPLEX_PROFILES env override — the 3-tier precedence chain.
env (recognized token) > config.yaml (top-level or nested gateway.*) >
default False. A blank / unrecognized env value is treated as UNSET and
falls through to config (the empty-secret trap: a provisioned-but-empty Fly
secret arrives as "" and must not shadow a config.yaml opt-in).
"""
def _load(self, tmp_path, monkeypatch, config_text=None):
hermes_home = tmp_path / ".hermes"
hermes_home.mkdir(exist_ok=True)
if config_text is not None:
(hermes_home / "config.yaml").write_text(config_text, encoding="utf-8")
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
return load_gateway_config()
# ── Tier 1: env wins ──────────────────────────────────────────────────
def test_env_true_forces_on_with_no_config(self, tmp_path, monkeypatch):
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", "true")
config = self._load(tmp_path, monkeypatch, config_text=None)
assert config.multiplex_profiles is True
def test_env_true_overrides_config_false(self, tmp_path, monkeypatch):
# THE discriminating test: env-set wins over an explicit config value.
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", "1")
config = self._load(
tmp_path, monkeypatch, config_text="multiplex_profiles: false\n"
)
assert config.multiplex_profiles is True
def test_env_false_overrides_config_true(self, tmp_path, monkeypatch):
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", "off")
config = self._load(
tmp_path, monkeypatch, config_text="multiplex_profiles: true\n"
)
assert config.multiplex_profiles is False
# ── Tier 2: config.yaml when env unset ────────────────────────────────
def test_config_true_when_env_unset(self, tmp_path, monkeypatch):
monkeypatch.delenv("GATEWAY_MULTIPLEX_PROFILES", raising=False)
config = self._load(
tmp_path, monkeypatch, config_text="multiplex_profiles: true\n"
)
assert config.multiplex_profiles is True
# ── The empty / unrecognized env trap: fall through, don't force off ──
def test_empty_env_does_not_shadow_config_true(self, tmp_path, monkeypatch):
# Provisioned-but-unpopulated Fly secret arrives as "". It must NOT
# turn OFF a config.yaml opt-in.
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", "")
config = self._load(
tmp_path, monkeypatch, config_text="multiplex_profiles: true\n"
)
assert config.multiplex_profiles is True
def test_whitespace_env_does_not_shadow_config_true(self, tmp_path, monkeypatch):
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", " ")
config = self._load(
tmp_path, monkeypatch, config_text="multiplex_profiles: true\n"
)
assert config.multiplex_profiles is True
def test_unrecognized_env_falls_through_to_config(self, tmp_path, monkeypatch):
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", "maybe")
config = self._load(
tmp_path, monkeypatch, config_text="multiplex_profiles: true\n"
)
assert config.multiplex_profiles is True
# ── Tier 3: default False ─────────────────────────────────────────────
def test_default_false_when_neither_set(self, tmp_path, monkeypatch):
monkeypatch.delenv("GATEWAY_MULTIPLEX_PROFILES", raising=False)
config = self._load(tmp_path, monkeypatch, config_text=None)
assert config.multiplex_profiles is False
# ── The resolver in isolation ─────────────────────────────────────────
def test_resolver_tristate(self, monkeypatch):
from gateway.config import _env_multiplex_profiles_override
monkeypatch.delenv("GATEWAY_MULTIPLEX_PROFILES", raising=False)
assert _env_multiplex_profiles_override() is None
for truthy in ("1", "true", "TRUE", "yes", "on"):
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", truthy)
assert _env_multiplex_profiles_override() is True, truthy
for falsy in ("0", "false", "FALSE", "no", "off"):
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", falsy)
assert _env_multiplex_profiles_override() is False, falsy
for noise in ("", " ", "maybe", "2"):
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", noise)
assert _env_multiplex_profiles_override() is None, repr(noise)

View file

@ -53,3 +53,52 @@ class TestNamedProfileMultiplexerGuard:
)
# No gateway.pid in tmp_path => no running default gateway => no raise.
gw._guard_named_profile_under_multiplexer(force=False)
def _fake_running_default_gateway(self, monkeypatch, tmp_path):
"""Make the guard believe a live default gateway exists at tmp_path."""
from hermes_cli import gateway as gw
import gateway.status as status
monkeypatch.setattr(gw, "_profile_suffix", lambda: "coder")
monkeypatch.setattr(
"hermes_constants.get_default_hermes_root", lambda: tmp_path
)
(tmp_path / "gateway.pid").write_text("12345", encoding="utf-8")
monkeypatch.setattr(status, "_read_pid_record", lambda p: {"pid": 12345})
monkeypatch.setattr(status, "_pid_from_record", lambda rec: 12345)
monkeypatch.setattr(status, "_pid_exists", lambda pid: True)
def test_env_forces_guard_even_without_config(self, monkeypatch, tmp_path):
"""GATEWAY_MULTIPLEX_PROFILES=true must trip the guard even when the
default profile's config.yaml has no multiplex_profiles key — the hosted
case where multiplex is forced purely by the env stamp."""
from hermes_cli import gateway as gw
self._fake_running_default_gateway(monkeypatch, tmp_path)
# No config.yaml written → the only signal is the env override.
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", "true")
with pytest.raises(SystemExit):
gw._guard_named_profile_under_multiplexer(force=False)
def test_env_false_disables_guard_over_config_true(self, monkeypatch, tmp_path):
"""GATEWAY_MULTIPLEX_PROFILES=false wins over a config.yaml opt-in, so
the guard stays inert (symmetric with the config precedence)."""
from hermes_cli import gateway as gw
self._fake_running_default_gateway(monkeypatch, tmp_path)
(tmp_path / "config.yaml").write_text(
"multiplex_profiles: true\n", encoding="utf-8"
)
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", "false")
# Env forces OFF → guard must NOT raise.
gw._guard_named_profile_under_multiplexer(force=False)
def test_blank_env_falls_through_to_config_and_raises(self, monkeypatch, tmp_path):
"""A blank env value must not shadow a config.yaml opt-in: the guard
still trips on the config value."""
from hermes_cli import gateway as gw
self._fake_running_default_gateway(monkeypatch, tmp_path)
(tmp_path / "config.yaml").write_text(
"multiplex_profiles: true\n", encoding="utf-8"
)
monkeypatch.setenv("GATEWAY_MULTIPLEX_PROFILES", "")
with pytest.raises(SystemExit):
gw._guard_named_profile_under_multiplexer(force=False)