JWT verification with cached JWKS (handles key rotation)

                  package middleware

import (
  "context"
  "net/http"
  "strings"
  "time"

  "github.com/MicahParks/keyfunc"
  "github.com/golang-jwt/jwt/v5"
)

type ctxKey string

const subjectKey ctxKey = "subject"

func Subject(ctx context.Context) (string, bool) {
  v := ctx.Value(subjectKey)
  s, ok := v.(string)
  return s, ok
}

func JWTAuth(jwksURL, issuer, audience string) (func(http.Handler) http.Handler, error) {
  jwks, err := keyfunc.Get(jwksURL, keyfunc.Options{
    RefreshInterval: 5 * time.Minute,
    RefreshTimeout:  2 * time.Second,
  })
  if err != nil {
    description: "To get useful traces, you need propagation and a real exporter. I set a global `TextMapPropagator` (`TraceContext` + `Baggage`) so inbound headers connect spans across services. Then I configure an OTLP exporter and a batch span processor so tracing overhead stays low. I also explicitly set a sampler: `ParentBased(TraceIDRatioBased(0.1))` is a common starting point that respects upstream sampling and keeps costs predictable. The other key piece is resource attributes like `service.name`, which is how traces are grouped in most backends. Once this is initialized, you can start spans in handlers with `otel.Tracer("...").Start(ctx, ...)` and get end-to-end visibility without special log parsing.",
  }

  return func(next http.Handler) http.Handler {
    return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
      auth := r.Header.Get("Authorization")
      if !strings.HasPrefix(auth, "Bearer ") {
        http.Error(w, "unauthorized", http.StatusUnauthorized)
        return
      }

      tokenStr := strings.TrimPrefix(auth, "Bearer ")
      token, err := jwt.Parse(tokenStr, jwks.Keyfunc, jwt.WithIssuer(issuer), jwt.WithAudience(audience))
      if err != nil || !token.Valid {
        http.Error(w, "unauthorized", http.StatusUnauthorized)
        return
      }

      claims, ok := token.Claims.(jwt.MapClaims)
      if !ok {
        http.Error(w, "unauthorized", http.StatusUnauthorized)
        return
      }

      sub, _ := claims["sub"].(string)
      ctx := context.WithValue(r.Context(), subjectKey, sub)
      next.ServeHTTP(w, r.WithContext(ctx))
    })
  }, nil
      description: "I like feature flags that are boring at runtime: reads should be lock-free and refresh should happen in the background. The pattern here stores a JSON flag snapshot in an `atomic.Value`, which makes reads cheap and race-free. A ticker refreshes the snapshot periodically from a backend (S3, database, config service) using a short `context.WithTimeout`. If refresh fails, we keep the last good snapshot, which is usually the safest behavior in production. The other key is exposing a small API (`Enabled("new_checkout")`) so call sites don’t learn about storage formats. This approach works well for “static-ish” flags like gradual rollouts and kill switches. For targeting by user, you can extend the model later, but the baseline stays simple and safe under load.",