Skip to content

Commit df53a34

Browse files
committed
remove request control hooks
1 parent 932b5e0 commit df53a34

File tree

4 files changed

+9
-209
lines changed

4 files changed

+9
-209
lines changed

pkg/plugins/scorer/pd_slo_aware_router_hooks.go

Lines changed: 0 additions & 111 deletions
This file was deleted.

pkg/plugins/scorer/pd_slo_aware_scorer.go

Lines changed: 6 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -74,16 +74,11 @@ func PDSLOAwareScorerFactory(name string, rawConfig json.RawMessage, handle plug
7474
return nil, fmt.Errorf("failed to start latency predictor: %w", err)
7575
}
7676

77-
// Create the SLO aware router and inject P/D-aware request builder
78-
// This builder will populate the PodType field based on llm-d.ai/role labels
79-
baseRouter := predictedlatency.NewPredictedLatency(cfg, predictor).WithName(name)
80-
baseRouter.SetRequestBuilder(NewPDPredictionRequestBuilder())
77+
// Create the SLO aware router with P/D-aware request builder
78+
// The builder populates the PodType field based on llm-d.ai/role labels
79+
// to distinguish prefill vs decode pods in training data
80+
router := predictedlatency.NewPredictedLatency(cfg, predictor).WithName(name)
81+
router.SetRequestBuilder(NewPDPredictionRequestBuilder())
8182

82-
// Wrap with PDSLOAwareRouter to add P/D-specific hook logic
83-
// The wrapper delegates to the base router while adding P/D-specific header extraction
84-
pdRouter := &PDSLOAwareRouter{
85-
PredictedLatency: baseRouter,
86-
}
87-
88-
return pdRouter, nil
83+
return router, nil
8984
}

pkg/sidecar/proxy/connector_nixlv2.go

Lines changed: 3 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -174,19 +174,12 @@ func (s *Server) runNIXLProtocolV2(w http.ResponseWriter, r *http.Request, prefi
174174
dreq.Body = io.NopCloser(strings.NewReader(string(dbody)))
175175
dreq.ContentLength = int64(len(dbody))
176176

177-
// 2. Forward to local decoder with prefill timing headers
177+
// 2. Forward to local decoder
178178

179179
s.logger.V(5).Info("sending request to decoder", "body", string(dbody))
180180

181-
// Wrap response writer to inject prefill timing headers for EPP training
182-
timingWriter := &timingResponseWriter{
183-
ResponseWriter: w,
184-
prefillLatencyMs: float64(prefillLatency.Milliseconds()),
185-
prefillPodHost: prefillPodHostPort,
186-
}
187-
188-
if !s.forwardDataParallel || !s.dataParallelHandler(timingWriter, dreq) {
181+
if !s.forwardDataParallel || !s.dataParallelHandler(w, dreq) {
189182
s.logger.V(4).Info("sending request to decoder", "to", s.decoderURL.Host)
190-
s.decoderProxy.ServeHTTP(timingWriter, dreq)
183+
s.decoderProxy.ServeHTTP(w, dreq)
191184
}
192185
}

pkg/sidecar/proxy/timing_writer.go

Lines changed: 0 additions & 77 deletions
This file was deleted.

0 commit comments

Comments
 (0)