feat: 🎉 Vectors na Vectors, the begining

Translate 1536 <-> 768 , 3072 <-> 2048

pkg/metrics/middleware.go

@@ -0,0 +1,62 @@
+package metrics
+
+import (
+	"fmt"
+	"net/http"
+
+	"github.com/uptrace/bunrouter"
+)
+
+// Middleware returns a bunrouter middleware that records per-request Prometheus metrics.
+// It reads timing from the RequestTrace stored in the context (set by server/trace.go).
+// The trace target/url labels are optional; pass empty strings if not applicable.
+func (r *Registry) Middleware(getTrace func(req bunrouter.Request) TraceSnapshot) bunrouter.MiddlewareFunc {
+	return func(next bunrouter.HandlerFunc) bunrouter.HandlerFunc {
+		return func(w http.ResponseWriter, req bunrouter.Request) error {
+			rw := &statusWriter{ResponseWriter: w, status: http.StatusOK}
+			err := next(rw, req)
+
+			snap := getTrace(req)
+			endpoint := req.URL.Path
+			status := fmt.Sprintf("%d", rw.status)
+
+			r.RequestsTotal.WithLabelValues(endpoint, status).Inc()
+			r.RequestDuration.WithLabelValues(endpoint).Observe(snap.TotalSeconds)
+			if snap.ForwardTarget != "" {
+				r.ForwardDuration.WithLabelValues(snap.ForwardTarget, snap.ForwardURL).Observe(snap.ForwardSeconds)
+			}
+			if snap.AdapterType != "" {
+				r.TranslateDuration.WithLabelValues(snap.AdapterType).Observe(snap.TranslateSeconds)
+			}
+			if snap.PromptTokens > 0 || snap.TotalTokens > 0 {
+				r.AddTokens(snap.ForwardTarget, snap.ForwardModel, snap.PromptTokens, snap.TotalTokens)
+			}
+
+			return err
+		}
+	}
+}
+
+// TraceSnapshot carries the timing and usage values the metrics middleware needs.
+type TraceSnapshot struct {
+	TotalSeconds     float64
+	ForwardSeconds   float64
+	TranslateSeconds float64
+	ForwardTarget    string
+	ForwardURL       string
+	ForwardModel     string
+	AdapterType      string
+	PromptTokens     int
+	TotalTokens      int
+}
+
+// statusWriter wraps http.ResponseWriter to capture the written status code.
+type statusWriter struct {
+	http.ResponseWriter
+	status int
+}
+
+func (sw *statusWriter) WriteHeader(code int) {
+	sw.status = code
+	sw.ResponseWriter.WriteHeader(code)
+}