Files
LocalAI/backend/python/common/grpc_auth.py
mudler's LocalAI [bot] 465d488c90 fix(distributed): reject wrong-model requests at the backend (#10970)
fix(distributed): reject wrong-model requests at the backend (#10952)

In distributed mode the controller caches a NodeModel row naming a backend's
host:port. A worker can recycle a stopped backend's gRPC port for a different
model's backend, and probeHealth verifies liveness rather than identity, so the
probe succeeds against whatever now occupies the port and the request is
dispatched to the wrong backend. The caller gets a silent wrong-model answer.

Nothing in the request could catch this: PredictOptions had no model field, so
model identity crossed the wire only in ModelOptions.Model at LoadModel time,
and the cached-hit path issues no LoadModel. Every backend's "model not loaded"
guard checks a nil handle, which a process holding a different model passes, so
the stale row was never dropped either.

Add PredictOptions.ModelIdentity and enforce it at the point of use:

  - The controller populates it in gRPCPredictOpts from ModelConfig.Model, the
    same expression ModelOptions feeds to model.WithModel and therefore the
    same value the backend received as ModelOptions.Model. Both are read from
    one config value in one function, so they are equal by construction and the
    comparison cannot false-reject.
  - Backends compare it against what they loaded and return NOT_FOUND with a
    fixed sentinel. Enforced in pkg/grpc/server.go (27 Go backends), an
    interceptor in backend/python/common (all 36 Python backends, no
    per-backend change), and the llama-cpp / ik-llama-cpp / ds4 C++ servers.
    That is every backend with real exposure: kokoros answers all four RPCs
    with unimplemented and privacy-filter implements none of them.
  - The router's reconcile drops the stale replica row on a mismatch, so the
    next request reloads somewhere correct.

Empty means "skip the check" on both sides: a controller that predates the
field sends nothing, a backend loaded by such a controller has nothing to
compare, and the C++ server synthesizes PredictOptions internally for ASR. That
keeps upgrades working in both directions.

Scoped to the four PredictOptions RPCs. TTSRequest.model and
SoundGenerationRequest.model are deliberately NOT validated: FileStagingClient
already rewrites them to worker-local absolute paths, so in distributed mode
they already differ from the load-time value and comparing them would reject
valid requests.

IsModelMismatch requires both the NOT_FOUND code and the sentinel, unlike the
neighbouring helpers which accept either. insightface's Embedding returns
NOT_FOUND "no face detected" on a PredictOptions RPC, and a code-only check
would drop a healthy replica row on every faceless image.


Assisted-by: Claude Code:claude-opus-4-8[1m] [Read] [Edit] [Bash]

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
2026-07-20 13:05:47 +02:00

99 lines
3.8 KiB
Python

"""Shared gRPC bearer token authentication interceptor for LocalAI Python backends.
When the environment variable LOCALAI_GRPC_AUTH_TOKEN is set, requests without
a valid Bearer token in the 'authorization' metadata header are rejected with
UNAUTHENTICATED. When the variable is empty or unset, no authentication is
performed (backward compatible).
"""
import hmac
import os
import grpc
from model_identity import AsyncModelIdentityInterceptor, ModelIdentityInterceptor
from parent_watch import start_parent_death_watcher
class _AbortHandler(grpc.RpcMethodHandler):
"""A method handler that immediately aborts with UNAUTHENTICATED."""
def __init__(self):
self.request_streaming = False
self.response_streaming = False
self.request_deserializer = None
self.response_serializer = None
self.unary_unary = self._abort
self.unary_stream = None
self.stream_unary = None
self.stream_stream = None
@staticmethod
def _abort(request, context):
context.abort(grpc.StatusCode.UNAUTHENTICATED, "invalid token")
class TokenAuthInterceptor(grpc.ServerInterceptor):
"""Sync gRPC server interceptor that validates a bearer token."""
def __init__(self, token: str):
self._token = token
self._abort_handler = _AbortHandler()
def intercept_service(self, continuation, handler_call_details):
metadata = dict(handler_call_details.invocation_metadata)
auth = metadata.get("authorization", "")
expected = "Bearer " + self._token
if not hmac.compare_digest(auth, expected):
return self._abort_handler
return continuation(handler_call_details)
class AsyncTokenAuthInterceptor(grpc.aio.ServerInterceptor):
"""Async gRPC server interceptor that validates a bearer token."""
def __init__(self, token: str):
self._token = token
async def intercept_service(self, continuation, handler_call_details):
metadata = dict(handler_call_details.invocation_metadata)
auth = metadata.get("authorization", "")
expected = "Bearer " + self._token
if not hmac.compare_digest(auth, expected):
return _AbortHandler()
return await continuation(handler_call_details)
def get_auth_interceptors(*, aio: bool = False):
"""Return the gRPC server interceptors every LocalAI Python backend installs.
Always includes model-identity enforcement (model_identity.py), which is
unrelated to authentication. Bearer token auth is added on top only when
LOCALAI_GRPC_AUTH_TOKEN is set.
Args:
aio: If True, return async-compatible interceptors for grpc.aio.server().
If False (default), return sync interceptors for grpc.server().
"""
# Arm the best-effort parent-death backstop here: this is the single helper
# every LocalAI Python backend invokes exactly once while building its gRPC
# server (mirroring how the Go watcher arms in pkg/grpc's shared serve path).
# start_parent_death_watcher() is idempotent and a no-op when disabled or on
# unsupported platforms — see parent_watch.py.
start_parent_death_watcher()
# Model-identity enforcement is independent of authentication and must be
# installed BEFORE the token check returns. gRPC auth is off by default, so
# an identity interceptor added below the early return would never be
# installed on any Python backend, and nothing would report it.
interceptors = [AsyncModelIdentityInterceptor()] if aio else [ModelIdentityInterceptor()]
token = os.environ.get("LOCALAI_GRPC_AUTH_TOKEN", "")
if not token:
return interceptors
if aio:
interceptors.append(AsyncTokenAuthInterceptor(token))
else:
interceptors.append(TokenAuthInterceptor(token))
return interceptors