Use Desired Replicas when scaling from zero

During some exploratory testing, I ran into an issue where
the gateway would attempt to scale a deployment from zero
replicas to min, despite there already being min replicas.

Why?

The scaling logic was looking for Available replicas when
it should have looked for Desired replicas. So when a
deployment had zero ready replicas due to readiness checks
failing, the gateway was attempting to scale from zero
to min.

This logic has been corrected and separated from the
a holding pattern where the gateway waits for a ready
replica.

Tested with KinD and an edited function which had a
readiness probe, which was failing and no ready
replicas. As desired, the gateway did not scale to min.

However, when setting desired replicas to zero, the
gateway did scale up as expected.

This change also modifies all print statements for
"seconds" and makes them use 4 decimal places instead of
the default which was a longer, more verbose string for
the logs.

Signed-off-by: Alex Ellis (OpenFaaS Ltd) <alexellis2@gmail.com>

gateway/Makefile

@@ -7,7 +7,7 @@ SERVER?=docker.io
 OWNER?=alexellis2
 NAME=gateway
 
-.PHONY: build-local
+.PHONY: local-docker
 build-local:
 	@echo $(SERVER)/$(OWNER)/$(NAME):$(TAG) \
 	&& docker buildx create --use --name=multiarch --node multiarch \
@@ -17,8 +17,8 @@ build-local:
 		--output "type=docker,push=false" \
 		--tag $(SERVER)/$(OWNER)/$(NAME):$(TAG) .
 
-.PHONY: build-push
-build-push:
+.PHONY: push-docker
+push-docker:
 	@echo $(SERVER)/$(OWNER)/$(NAME):$(TAG) \
 	&& docker buildx create --use --name=multiarch --node multiarch \
 	&& docker buildx build \

gateway/handlers/notifiers.go

@@ -82,6 +82,6 @@ type LoggingNotifier struct {
 // Notify the LoggingNotifier about a request
 func (LoggingNotifier) Notify(method string, URL string, originalURL string, statusCode int, event string, duration time.Duration) {
 	if event == "completed" {
-		log.Printf("Forwarded [%s] to %s - [%d] - %fs seconds", method, originalURL, statusCode, duration.Seconds())
+		log.Printf("Forwarded [%s] to %s - [%d] - %.4fs", method, originalURL, statusCode, duration.Seconds())
 	}
 }

gateway/handlers/scaling.go

@@ -48,6 +48,7 @@ func MakeScalingHandler(next http.HandlerFunc, scaler scaling.FunctionScaler, co
 			return
 		}
 
-		log.Printf("[Scale] function=%s.%s 0=>N timed-out after %fs\n", functionName, namespace, res.Duration.Seconds())
+		log.Printf("[Scale] function=%s.%s 0=>N timed-out after %.4fs\n",
+			functionName, namespace, res.Duration.Seconds())
 	}
 }

gateway/plugin/external.go

@@ -102,7 +102,7 @@ func (s ExternalServiceQuery) GetReplicas(serviceName, serviceNamespace string)
 		// log.Printf("GetReplicas [%s.%s] took: %fs", serviceName, serviceNamespace, time.Since(start).Seconds())
 
 	} else {
-		log.Printf("GetReplicas [%s.%s] took: %fs, code: %d\n", serviceName, serviceNamespace, time.Since(start).Seconds(), res.StatusCode)
+		log.Printf("GetReplicas [%s.%s] took: %.4fs, code: %d\n", serviceName, serviceNamespace, time.Since(start).Seconds(), res.StatusCode)
 		return emptyServiceQueryResponse, fmt.Errorf("server returned non-200 status code (%d) for function, %s, body: %s", res.StatusCode, serviceName, string(bytesOut))
 	}
 
@@ -176,7 +176,8 @@ func (s ExternalServiceQuery) SetReplicas(serviceName, serviceNamespace string,
 		err = fmt.Errorf("error scaling HTTP code %d, %s", res.StatusCode, urlPath)
 	}
 
-	log.Printf("SetReplicas [%s.%s] took: %fs", serviceName, serviceNamespace, time.Since(start).Seconds())
+	log.Printf("SetReplicas [%s.%s] took: %.4fs",
+		serviceName, serviceNamespace, time.Since(start).Seconds())
 
 	return err
 }

gateway/scaling/function_scaler.go

@@ -39,6 +39,8 @@ type FunctionScaleResult struct {
 func (f *FunctionScaler) Scale(functionName, namespace string) FunctionScaleResult {
 	start := time.Now()
 
+	// First check the cache, if there are available replicas, then the
+	// request can be served.
 	if cachedResponse, hit := f.Cache.Get(functionName, namespace); hit &&
 		cachedResponse.AvailableReplicas > 0 {
 		return FunctionScaleResult{
@@ -48,8 +50,10 @@ func (f *FunctionScaler) Scale(functionName, namespace string) FunctionScaleResu
 			Duration:  time.Since(start),
 		}
 	}
-	getKey := fmt.Sprintf("GetReplicas-%s.%s", functionName, namespace)
 
+	// The wasn't a hit, or there were no available replicas found
+	// so query the live endpoint
+	getKey := fmt.Sprintf("GetReplicas-%s.%s", functionName, namespace)
 	res, err, _ := f.SingleFlight.Do(getKey, func() (interface{}, error) {
 		return f.Config.ServiceQuery.GetReplicas(functionName, namespace)
 	})
@@ -71,16 +75,30 @@ func (f *FunctionScaler) Scale(functionName, namespace string) FunctionScaleResu
 		}
 	}
 
-	queryResponse := res.(ServiceQueryResponse)
+	// Check if there are available replicas in the live data
+	if res.(ServiceQueryResponse).AvailableReplicas > 0 {
+		return FunctionScaleResult{
+			Error:     nil,
+			Available: true,
+			Found:     true,
+			Duration:  time.Since(start),
+		}
+	}
 
+	// Store the result of GetReplicas in the cache
+	queryResponse := res.(ServiceQueryResponse)
 	f.Cache.Set(functionName, namespace, queryResponse)
 
-	if queryResponse.AvailableReplicas == 0 {
+	// If the desired replica count is 0, then a scale up event
+	// is required.
+	if queryResponse.Replicas == 0 {
 		minReplicas := uint64(1)
 		if queryResponse.MinReplicas > 0 {
 			minReplicas = queryResponse.MinReplicas
 		}
 
+		// In a retry-loop, first query desired replicas, then
+		// set them if the value is still at 0.
 		scaleResult := types.Retry(func(attempt int) error {
 
 			res, err, _ := f.SingleFlight.Do(getKey, func() (interface{}, error) {
@@ -91,19 +109,23 @@ func (f *FunctionScaler) Scale(functionName, namespace string) FunctionScaleResu
 				return err
 			}
 
+			// Cache the response
 			queryResponse = res.(ServiceQueryResponse)
-
 			f.Cache.Set(functionName, namespace, queryResponse)
 
+			// The scale up is complete because the desired replica count
+			// has been set to 1 or more.
 			if queryResponse.Replicas > 0 {
 				return nil
 			}
 
+			// Request a scale up to the minimum amount of replicas
 			setKey := fmt.Sprintf("SetReplicas-%s.%s", functionName, namespace)
 
 			if _, err, _ := f.SingleFlight.Do(setKey, func() (interface{}, error) {
 
-				log.Printf("[Scale %d] function=%s 0 => %d requested", attempt, functionName, minReplicas)
+				log.Printf("[Scale %d/%d] function=%s 0 => %d requested",
+					attempt, int(f.Config.SetScaleRetries), functionName, minReplicas)
 
 				if err := f.Config.ServiceQuery.SetReplicas(functionName, namespace, minReplicas); err != nil {
 					return nil, fmt.Errorf("unable to scale function [%s], err: %s", functionName, err)
@@ -126,6 +148,9 @@ func (f *FunctionScaler) Scale(functionName, namespace string) FunctionScaleResu
 			}
 		}
 
+	}
+
+	// Holding pattern for at least one function replica to be available
 	for i := 0; i < int(f.Config.MaxPollCount); i++ {
 
 		res, err, _ := f.SingleFlight.Do(getKey, func() (interface{}, error) {
@@ -150,8 +175,7 @@ func (f *FunctionScaler) Scale(functionName, namespace string) FunctionScaleResu
 
 		if queryResponse.AvailableReplicas > 0 {
 
-				log.Printf("[Scale] function=%s 0 => %d successful - %fs",
-					functionName, queryResponse.AvailableReplicas, totalTime.Seconds())
+			log.Printf("[Ready] function=%s waited for - %.4fs", functionName, totalTime.Seconds())
 
 			return FunctionScaleResult{
 				Error:     nil,
@@ -163,7 +187,6 @@ func (f *FunctionScaler) Scale(functionName, namespace string) FunctionScaleResu
 
 		time.Sleep(f.Config.FunctionPollInterval)
 	}
-	}
 
 	return FunctionScaleResult{
 		Error:     nil,