fix(steer): revalidate the active-run backend on the cache-only path

The cache-only Steer path only special-cased backend == "gateway" and sent
every other value — including a missing, empty, or unknown backend — to the
cached local agent. Another runtime could own the run while guidance was
enqueued on a stale local agent.

Fail closed instead: the in-process worker now registers its active run with
the explicit local backend tag (WEBUI_LOCAL_CHAT_BACKEND, the same "legacy"
value webui_chat_backend_mode() returns), and cache-only Steer only enqueues
agent.steer() on that tag. Gateway keeps its gateway_steer_queued outcome; a
missing, empty, or foreign backend returns stream_dead without steering.

Extends the positive-ownership regression with backend cases and adds a
worker-side check that the local backend tag is actually registered.
This commit is contained in:
Anthony Ruiz committed 2026-09-20 17:07:57 +00:00
1 parent 20c16e9167
commit 26277f933b
3 files changed
+60 -4

No files matched your search

+4 -1
View File
@@ -153,6 +153,9 @@ _WEBUI_GATEWAY_BASE_URL_ENV = "HERMES_WEBUI_GATEWAY_BASE_URL"
_WEBUI_GATEWAY_API_KEY_ENV = "HERMES_WEBUI_GATEWAY_API_KEY"
_WEBUI_GATEWAY_USE_RUNS_API_ENV = "HERMES_WEBUI_GATEWAY_USE_RUNS_API"
_GATEWAY_CHAT_BACKENDS = {"gateway", "api_server", "api-server"}
# Backend tag of the in-process WebUI runtime. Local workers register their
# active run with it; cache-only Steer only enqueues on this explicit value.
WEBUI_LOCAL_CHAT_BACKEND = "legacy"
def _gateway_model_field(model: str | None) -> str:
@@ -272,7 +275,7 @@ def webui_chat_backend_mode(config_data=None, environ: dict[str, str] | None = N
).strip().lower()
if raw in _GATEWAY_CHAT_BACKENDS:
return "gateway"
return "legacy"
return WEBUI_LOCAL_CHAT_BACKEND
def webui_gateway_chat_enabled(config_data=None, environ: dict[str, str] | None = None) -> bool:
+10 -2
View File
@@ -67,6 +67,7 @@ from api.helpers import (
)
from api.compression_anchor import is_context_compression_marker, visible_messages_for_anchor
from api.compression_recovery import stamp_compression_exhausted_recovery
from api.gateway_chat import WEBUI_LOCAL_CHAT_BACKEND
from api.metering import meter
from api.run_journal import RunJournalWriter
from api.todo_state import attach_todo_state, emit_todo_state
@@ -9192,6 +9193,7 @@ def _run_agent_streaming(
model=model,
provider=model_provider,
ephemeral=bool(ephemeral),
backend=WEBUI_LOCAL_CHAT_BACKEND,
)
try:
run_journal = RunJournalWriter(session_id, stream_id)
@@ -13691,6 +13693,11 @@ def _handle_chat_steer(handler, body: dict) -> bool:
# stream owner AND the active-run session must equal the requesting
# session. Missing metadata is ambiguous and fails closed (stream_dead);
# an unfenced cache object is never steered on absent ownership.
#
# The active-run backend is revalidated the same way. With no registered
# worker, the backend tag is the only proof that an in-process runtime
# owns this run: Gateway resolves to its own outcome, the explicit local
# tag may enqueue, and a missing, empty, or foreign backend fails closed.
result = {"accepted": False, "fallback": "stream_dead", "stream_id": None}
with _cfg.STREAMS_LOCK:
if active_stream_id in _cfg.STREAMS:
@@ -13700,11 +13707,12 @@ def _handle_chat_steer(handler, body: dict) -> bool:
owned_stream = bool(owner) and owner == sid
owned_run = bool(run.get("session_id")) and run["session_id"] == sid
if owned_stream and owned_run and run.get("phase") != "cancelling":
if run.get("backend") == "gateway":
backend = run.get("backend")
if backend == "gateway":
# Gateway owns transport; a local cache object is never steered.
result = {"accepted": False, "fallback": "gateway_steer_queued",
"stream_id": active_stream_id}
else:
elif backend == WEBUI_LOCAL_CHAT_BACKEND:
try:
accepted = bool(agent.steer(text))
except Exception as exc:
+46 -1
View File
@@ -121,6 +121,8 @@ def test_cache_only_mismatch_does_not_close_an_agent_owned_by_another_run(scene)
@pytest.mark.parametrize("ownership", [
"both", "missing-owner", "missing-run-session", "empty-run-session",
"missing-both", "missing-run-entry",
"local-backend", "missing-backend", "empty-backend", "foreign-backend",
"gateway-backend",
])
def test_cache_only_steer_requires_positive_stream_and_run_ownership(scene, ownership):
"""Cache-only Steer must prove ownership, not merely fail to refute it.
@@ -129,6 +131,11 @@ def test_cache_only_steer_requires_positive_stream_and_run_ownership(scene, owne
missing stream owner or a missing active-run session is ambiguous and must
fail closed with ``stream_dead`` without calling ``agent.steer()``. Only
both identities present and equal to the requesting session may enqueue.
The active-run backend is revalidated the same way: only the explicit
local backend tag registered by ``_run_agent_streaming`` may enqueue on the
cached local agent. Gateway keeps its own terminal outcome; a missing,
empty, or foreign backend is ambiguous and fails closed.
"""
agent, other, _ = scene
agent.session_id = "original"
@@ -141,12 +148,25 @@ def test_cache_only_steer_requires_positive_stream_and_run_ownership(scene, owne
config.ACTIVE_RUNS["run"]["session_id"] = ""
elif ownership == "missing-run-entry":
config.ACTIVE_RUNS.pop("run")
elif ownership == "local-backend":
config.ACTIVE_RUNS["run"]["backend"] = streaming.WEBUI_LOCAL_CHAT_BACKEND
elif ownership == "missing-backend":
config.ACTIVE_RUNS["run"].pop("backend")
elif ownership == "empty-backend":
config.ACTIVE_RUNS["run"]["backend"] = ""
elif ownership == "foreign-backend":
config.ACTIVE_RUNS["run"]["backend"] = "foreign"
elif ownership == "gateway-backend":
config.ACTIVE_RUNS["run"]["backend"] = "gateway"
before = dict(config.SESSION_AGENT_CACHE)
assert "run" in config.STREAMS
result = steer()
if ownership == "both":
if ownership in ("both", "local-backend"):
assert result == {"accepted": True, "fallback": None, "stream_id": "run"}
agent.steer.assert_called_once_with("updated guidance")
elif ownership == "gateway-backend":
assert result == {"accepted": False, "fallback": "gateway_steer_queued", "stream_id": "run"}
agent.steer.assert_not_called()
else:
assert result == {"accepted": False, "fallback": "stream_dead", "stream_id": None}
agent.steer.assert_not_called()
@@ -157,6 +177,31 @@ def test_cache_only_steer_requires_positive_stream_and_run_ownership(scene, owne
assert "run" in config.STREAMS
def test_local_worker_registers_the_backend_cache_only_steer_requires(monkeypatch):
"""The in-process worker must tag its active run with the local backend.
Cache-only Steer only enqueues on that explicit tag, so a worker that
registered without it would make every cache-only Steer fail closed.
"""
captured = {}
class _Registered(Exception):
pass
def register(stream_id, **metadata):
captured[stream_id] = metadata
raise _Registered()
monkeypatch.setattr(streaming, "peek_stream", lambda stream_id: queue.Queue())
monkeypatch.setattr(streaming, "register_active_run", register)
with pytest.raises(_Registered):
streaming._run_agent_streaming("original", "hi", "m", "/tmp", "run")
assert captured["run"]["session_id"] == "original"
assert captured["run"]["backend"] == streaming.WEBUI_LOCAL_CHAT_BACKEND
assert streaming.WEBUI_LOCAL_CHAT_BACKEND
assert streaming.WEBUI_LOCAL_CHAT_BACKEND != "gateway"
def test_http_response_is_written_after_stream_lock_release(scene, monkeypatch):
from api import helpers