app/vmauth: pick first backend to process request when all backends are unavailable (#10886)

The commit restores the previous behavior where the first backend is still selected and the request is sent to it. This behavior existed before commit 9c36f0931a, but was later changed to return no backends. Hence, vmauth would reject all requests for the next 3s if all backends are unavailable. In some rare cases, it leads to an increase in error responses. 

The commit restores the original behavior, adds comments explaining why it is important, and introduces tests covering the logic.

Fixes https://github.com/VictoriaMetrics/VictoriaMetrics/issues/10837
PR https://github.com/VictoriaMetrics/VictoriaMetrics/pull/10886

---------

Signed-off-by: JAYICE <1185430411@qq.com>
Signed-off-by: Max Kotliar <kotlyar.maksim@gmail.com>
Co-authored-by: Max Kotliar <mkotlyar@victoriametrics.com>
Co-authored-by: Hui Wang <haley@victoriametrics.com>

app/vmauth/auth_config.go

@@ -610,6 +610,7 @@ func areEqualBackendURLs(a, b []*backendURL) bool {
 }
 
 // getFirstAvailableBackendURL returns the first available backendURL, which isn't broken.
+// If all backendURLs are broken, then returns the first backendURL.
 //
 // backendURL.put() must be called on the returned backendURL after the request is complete.
 func getFirstAvailableBackendURL(bus []*backendURL) *backendURL {
@@ -628,21 +629,22 @@ func getFirstAvailableBackendURL(bus []*backendURL) *backendURL {
 			return bu
 		}
 	}
-	return nil
+
+	// All backend urls are unavailable, then returning a first one, it could help increase the success rate of the requests。
+	// See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/10837#issuecomment-4307050980.
+	bu.get()
+	return bu
 }
 
 // getLeastLoadedBackendURL returns a non-broken backendURL with the lowest number of concurrent requests.
+// If all backendURLs are broken, then returns the first backendURL.
 //
 // backendURL.put() must be called on the returned backendURL after the request is complete.
 func getLeastLoadedBackendURL(bus []*backendURL, atomicCounter *atomic.Uint32) *backendURL {
+	firstBu := bus[0]
 	if len(bus) == 1 {
-		// Fast path - return the only backend url.
-		bu := bus[0]
-		if bu.isBroken() {
-			return nil
-		}
-		bu.get()
-		return bu
+		firstBu.get()
+		return firstBu
 	}
 
 	// Slow path - select other backend urls.
@@ -680,7 +682,10 @@ func getLeastLoadedBackendURL(bus []*backendURL, atomicCounter *atomic.Uint32) *
 	}
 	buMin := bus[buMinIdx]
 	if buMin.isBroken() {
-		return nil
+		// If all backendURLs are broken, then returns the first backendURL.
+		// See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/10837#issuecomment-4307050980.
+		firstBu.get()
+		return firstBu
 	}
 	buMin.get()
 	atomicCounter.CompareAndSwap(n+1, buMinIdx+1)

app/vmauth/auth_config_test.go

@@ -1031,6 +1031,33 @@ func TestLogRequest(t *testing.T) {
 	f("foo", 404, 10*time.Millisecond, `access_log request_host="localhost:8080" request_uri="" status_code=404 remote_addr="" user_agent="" referer="" duration_ms=10 username="foo"`)
 }
 
+func TestGetFirstAvailableBackend(t *testing.T) {
+	f := func(broken []bool, expectedIdx int) {
+		t.Helper()
+		bus := make([]*backendURL, len(broken))
+		for i := range broken {
+			bus[i] = &backendURL{
+				url: &url.URL{Host: fmt.Sprintf("server-%d", i)},
+			}
+			bus[i].broken.Store(broken[i])
+		}
+		bu := getFirstAvailableBackendURL(bus)
+		if bu == nil {
+			t.Fatalf("unexpected nil backend")
+		}
+		if bu.url.Host != fmt.Sprintf("server-%d", expectedIdx) {
+			t.Fatalf("unexpected backend, expected server-%d, got %s", expectedIdx, bu.url.Host)
+		}
+	}
+
+	f([]bool{false, false, false}, 0)
+	f([]bool{true, true, false}, 2)
+	// all backend are broken, then return the first one.
+	f([]bool{true, true, true}, 0)
+	f([]bool{true}, 0)
+
+}
+
 func getRegexs(paths []string) []*Regex {
 	var sps []*Regex
 	for _, path := range paths {