cache: add negative cache; router: skip race for cached 404s

Signed-off-by: NotAShelf <raf@notashelf.dev>
Change-Id: Ibeb44b313850395898bb20f2d947b0b76a6a6964

internal/router/router.go

@@ -41,7 +41,7 @@ type Router struct {
 	negativeTTL  time.Duration
 	client       *http.Client
 	mu           sync.RWMutex
-	upstreamKeys map[string]string // upstream URL → Nix public key string
+	upstreamKeys map[string]string // upstream URL -> Nix public key string
 	group        singleflight.Group
 }
 
@@ -73,6 +73,12 @@ func (r *Router) SetUpstreamKey(url, pubKeyStr string) error {
 // Returns the best upstream for the given store hash.
 // Checks the route cache first; on miss races the provided candidates.
 func (r *Router) Resolve(storeHash string, candidates []string) (*Result, error) {
+	// Fast path: negative cache.
+	if neg, err := r.db.IsNegative(storeHash); err == nil && neg {
+		return nil, ErrNotFound
+	}
+
+	// Fast path: route cache hit.
 	entry, err := r.db.GetRoute(storeHash)
 	if err == nil && entry != nil && entry.IsValid() {
 		h := r.prober.GetHealth(entry.UpstreamURL)
@@ -88,7 +94,14 @@ func (r *Router) Resolve(storeHash string, candidates []string) (*Result, error)
 	metrics.NarinfoCacheMisses.Inc()
 
 	v, raceErr, _ := r.group.Do(storeHash, func() (interface{}, error) {
-		return r.race(storeHash, candidates)
+		result, err := r.race(storeHash, candidates)
+		if errors.Is(err, ErrNotFound) {
+			_ = r.db.SetNegative(storeHash, r.negativeTTL)
+		}
+		if err != nil {
+			return nil, err
+		}
+		return result, nil
 	})
 	if raceErr != nil {
 		return nil, raceErr

internal/router/router_test.go

@@ -177,6 +177,40 @@ func TestResolveWithDownUpstream(t *testing.T) {
 	}
 }
 
+func TestNegativeCaching(t *testing.T) {
+	var raceCount int32
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		atomic.AddInt32(&raceCount, 1)
+		w.WriteHeader(http.StatusNotFound)
+	}))
+	defer ts.Close()
+
+	db, err := cache.Open(":memory:", 1000)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer db.Close()
+	p := prober.New(0.3)
+	p.InitUpstreams([]config.UpstreamConfig{{URL: ts.URL}})
+	r := router.New(db, p, time.Hour, 5*time.Second, 10*time.Minute)
+
+	_, err = r.Resolve("not-on-any-upstream", []string{ts.URL})
+	if !errors.Is(err, router.ErrNotFound) {
+		t.Fatalf("first resolve: expected ErrNotFound, got %v", err)
+	}
+	count1 := atomic.LoadInt32(&raceCount)
+
+	_, err = r.Resolve("not-on-any-upstream", []string{ts.URL})
+	if !errors.Is(err, router.ErrNotFound) {
+		t.Fatalf("second resolve: expected ErrNotFound, got %v", err)
+	}
+	count2 := atomic.LoadInt32(&raceCount)
+
+	if count2 != count1 {
+		t.Errorf("second resolve hit upstream %d extra times, want 0 (should be negatively cached)", count2-count1)
+	}
+}
+
 func TestSingleflightDedup(t *testing.T) {
 	var headCount int32
 	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {