Files
LocalAI/core/services/nodes/inflight.go
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

279 lines
12 KiB
Go

package nodes
import (
"context"
"sync"
"time"
"github.com/mudler/LocalAI/pkg/grpc"
"github.com/mudler/LocalAI/pkg/grpc/grpcerrors"
pb "github.com/mudler/LocalAI/pkg/grpc/proto"
"github.com/mudler/xlog"
ggrpc "google.golang.org/grpc"
)
// InFlightTrackingClient wraps a grpc.Backend and tracks active inference requests
// in the NodeRegistry. This allows the router's eviction logic to know which models
// are actively serving and should not be unloaded.
//
// Per-replica: a single tracker instance is bound to (nodeID, modelName, replicaIndex).
// The router constructs one tracker per Route() result, so each in-flight tick lands
// on the correct row even when multiple replicas of the same model live on the same node.
//
// Embedding only grpc.ControlBackend (not the whole grpc.Backend) is what makes
// the in-flight accounting safe by construction: the control-plane methods pass
// through untracked, while every grpc.InferenceBackend method must be declared
// explicitly below to satisfy grpc.Backend. Adding an inference method to the
// interface therefore breaks this file's build (see the var assertion below)
// until it is wrapped with track() - so a new inference path can't be added
// without an in-flight accounting decision.
type InFlightTrackingClient struct {
grpc.ControlBackend // passthrough for control-plane / streaming-constructor methods
inner grpc.InferenceBackend // tracked inference methods delegate here
registry InFlightTracker
nodeID string
modelName string
replicaIndex int
firstOnce sync.Once // guards onFirstComplete
onFirstComplete func() // called once after the first tracked inference call completes
}
// Compile-time contract: *InFlightTrackingClient must implement the FULL backend
// surface. Because it embeds only ControlBackend, this fails to compile if any
// InferenceBackend method is left unwrapped.
var _ grpc.Backend = (*InFlightTrackingClient)(nil)
// NewInFlightTrackingClient wraps a gRPC backend client with in-flight tracking.
func NewInFlightTrackingClient(inner grpc.Backend, registry InFlightTracker, nodeID, modelName string, replicaIndex int) *InFlightTrackingClient {
return &InFlightTrackingClient{
ControlBackend: inner,
inner: inner,
registry: registry,
nodeID: nodeID,
modelName: modelName,
replicaIndex: replicaIndex,
}
}
// OnFirstComplete registers a callback that fires once after the first tracked
// inference call completes. This is used to release the initial in-flight
// reservation (set during model load) after the triggering request finishes,
// so that in-flight returns to 0 when the model is idle.
func (c *InFlightTrackingClient) OnFirstComplete(fn func()) {
c.onFirstComplete = fn
}
func (c *InFlightTrackingClient) track(ctx context.Context) func() {
if err := c.registry.IncrementInFlight(ctx, c.nodeID, c.modelName, c.replicaIndex); err != nil {
xlog.Warn("Failed to increment in-flight counter", "node", c.nodeID, "model", c.modelName, "replica", c.replicaIndex, "error", err)
return func() {}
}
return func() {
decCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
c.registry.DecrementInFlight(decCtx, c.nodeID, c.modelName, c.replicaIndex)
// Release the initial reservation after the first inference call completes
if c.onFirstComplete != nil {
c.firstOnce.Do(c.onFirstComplete)
}
}
}
// reconcile self-heals stale routing: when a backend reports that the model is
// no longer loaded (the process survived but the model was evicted, while the
// registry still lists it as loaded), it drops the replica row so the next
// request triggers a fresh load instead of routing back here. Without this the
// model stays unreachable until the controller restarts. The original error is
// returned unchanged.
//
// It also drops the row when the backend reports that it has a DIFFERENT model
// loaded. That happens when a worker recycles a stopped backend's gRPC port for
// another model's backend: the cached row still names that address, and a
// liveness-only probe cannot tell the stale row from a valid one, so the wrong
// backend answers (#10952). The two conditions differ in cause but not in cure.
func (c *InFlightTrackingClient) reconcile(err error) error {
mismatch := grpcerrors.IsModelMismatch(err)
if !grpcerrors.IsModelNotLoaded(err) && !mismatch {
return err
}
reason := "model not loaded"
if mismatch {
reason = "wrong model loaded"
}
rmCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
if rmErr := c.registry.RemoveNodeModel(rmCtx, c.nodeID, c.modelName, c.replicaIndex); rmErr != nil {
xlog.Warn("Failed to drop stale replica",
"node", c.nodeID, "model", c.modelName, "replica", c.replicaIndex, "reason", reason, "error", rmErr)
} else {
xlog.Warn("Dropped stale replica so the next request reloads",
"node", c.nodeID, "model", c.modelName, "replica", c.replicaIndex, "reason", reason)
}
return err
}
// --- Tracked inference methods ---
func (c *InFlightTrackingClient) Predict(ctx context.Context, in *pb.PredictOptions, opts ...ggrpc.CallOption) (*pb.Reply, error) {
defer c.track(ctx)()
reply, err := c.inner.Predict(ctx, in, opts...)
return reply, c.reconcile(err)
}
func (c *InFlightTrackingClient) PredictStream(ctx context.Context, in *pb.PredictOptions, f func(reply *pb.Reply), opts ...ggrpc.CallOption) error {
defer c.track(ctx)()
return c.reconcile(c.inner.PredictStream(ctx, in, f, opts...))
}
func (c *InFlightTrackingClient) Embeddings(ctx context.Context, in *pb.PredictOptions, opts ...ggrpc.CallOption) (*pb.EmbeddingResult, error) {
defer c.track(ctx)()
res, err := c.inner.Embeddings(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) GenerateImage(ctx context.Context, in *pb.GenerateImageRequest, opts ...ggrpc.CallOption) (*pb.Result, error) {
defer c.track(ctx)()
res, err := c.inner.GenerateImage(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) GenerateVideo(ctx context.Context, in *pb.GenerateVideoRequest, opts ...ggrpc.CallOption) (*pb.Result, error) {
defer c.track(ctx)()
res, err := c.inner.GenerateVideo(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) TTS(ctx context.Context, in *pb.TTSRequest, opts ...ggrpc.CallOption) (*pb.Result, error) {
defer c.track(ctx)()
res, err := c.inner.TTS(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) TTSStream(ctx context.Context, in *pb.TTSRequest, f func(reply *pb.Reply), opts ...ggrpc.CallOption) error {
defer c.track(ctx)()
return c.reconcile(c.inner.TTSStream(ctx, in, f, opts...))
}
func (c *InFlightTrackingClient) SoundGeneration(ctx context.Context, in *pb.SoundGenerationRequest, opts ...ggrpc.CallOption) (*pb.Result, error) {
defer c.track(ctx)()
res, err := c.inner.SoundGeneration(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) AudioTranscription(ctx context.Context, in *pb.TranscriptRequest, opts ...ggrpc.CallOption) (*pb.TranscriptResult, error) {
defer c.track(ctx)()
res, err := c.inner.AudioTranscription(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) AudioTranscriptionStream(ctx context.Context, in *pb.TranscriptRequest, f func(chunk *pb.TranscriptStreamResponse), opts ...ggrpc.CallOption) error {
defer c.track(ctx)()
return c.reconcile(c.inner.AudioTranscriptionStream(ctx, in, f, opts...))
}
func (c *InFlightTrackingClient) Detect(ctx context.Context, in *pb.DetectOptions, opts ...ggrpc.CallOption) (*pb.DetectResponse, error) {
defer c.track(ctx)()
res, err := c.inner.Detect(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) Depth(ctx context.Context, in *pb.DepthRequest, opts ...ggrpc.CallOption) (*pb.DepthResponse, error) {
defer c.track(ctx)()
res, err := c.inner.Depth(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) Rerank(ctx context.Context, in *pb.RerankRequest, opts ...ggrpc.CallOption) (*pb.RerankResult, error) {
defer c.track(ctx)()
res, err := c.inner.Rerank(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) VAD(ctx context.Context, in *pb.VADRequest, opts ...ggrpc.CallOption) (*pb.VADResponse, error) {
defer c.track(ctx)()
res, err := c.inner.VAD(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) Diarize(ctx context.Context, in *pb.DiarizeRequest, opts ...ggrpc.CallOption) (*pb.DiarizeResponse, error) {
defer c.track(ctx)()
res, err := c.inner.Diarize(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) FaceVerify(ctx context.Context, in *pb.FaceVerifyRequest, opts ...ggrpc.CallOption) (*pb.FaceVerifyResponse, error) {
defer c.track(ctx)()
res, err := c.inner.FaceVerify(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) FaceAnalyze(ctx context.Context, in *pb.FaceAnalyzeRequest, opts ...ggrpc.CallOption) (*pb.FaceAnalyzeResponse, error) {
defer c.track(ctx)()
res, err := c.inner.FaceAnalyze(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) VoiceVerify(ctx context.Context, in *pb.VoiceVerifyRequest, opts ...ggrpc.CallOption) (*pb.VoiceVerifyResponse, error) {
defer c.track(ctx)()
res, err := c.inner.VoiceVerify(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) VoiceAnalyze(ctx context.Context, in *pb.VoiceAnalyzeRequest, opts ...ggrpc.CallOption) (*pb.VoiceAnalyzeResponse, error) {
defer c.track(ctx)()
res, err := c.inner.VoiceAnalyze(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) VoiceEmbed(ctx context.Context, in *pb.VoiceEmbedRequest, opts ...ggrpc.CallOption) (*pb.VoiceEmbedResponse, error) {
defer c.track(ctx)()
res, err := c.inner.VoiceEmbed(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) TokenClassify(ctx context.Context, in *pb.TokenClassifyRequest, opts ...ggrpc.CallOption) (*pb.TokenClassifyResponse, error) {
defer c.track(ctx)()
res, err := c.inner.TokenClassify(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) Score(ctx context.Context, in *pb.ScoreRequest, opts ...ggrpc.CallOption) (*pb.ScoreResponse, error) {
defer c.track(ctx)()
res, err := c.inner.Score(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) SoundDetection(ctx context.Context, in *pb.SoundDetectionRequest, opts ...ggrpc.CallOption) (*pb.SoundDetectionResponse, error) {
defer c.track(ctx)()
res, err := c.inner.SoundDetection(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) AudioEncode(ctx context.Context, in *pb.AudioEncodeRequest, opts ...ggrpc.CallOption) (*pb.AudioEncodeResult, error) {
defer c.track(ctx)()
res, err := c.inner.AudioEncode(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) AudioDecode(ctx context.Context, in *pb.AudioDecodeRequest, opts ...ggrpc.CallOption) (*pb.AudioDecodeResult, error) {
defer c.track(ctx)()
res, err := c.inner.AudioDecode(ctx, in, opts...)
return res, c.reconcile(err)
}
func (c *InFlightTrackingClient) AudioTransform(ctx context.Context, in *pb.AudioTransformRequest, opts ...ggrpc.CallOption) (*pb.AudioTransformResult, error) {
defer c.track(ctx)()
res, err := c.inner.AudioTransform(ctx, in, opts...)
return res, c.reconcile(err)
}
// AudioTransformStream, AudioToAudioStream and Forward live in grpc.ControlBackend
// and are passed through via the embedded field, NOT tracked: they return a stream
// client and the inference spans the stream's lifetime, not the constructor call.
// Wrapping the constructor with track() would increment and immediately decrement
// (and fire onFirstComplete) before any audio flows. Tracking those correctly needs
// the done() func tied to stream close, which the Backend interface doesn't surface
// here. If they ever need tracking, move them to grpc.InferenceBackend - the build
// will then force an explicit wrapper here.