Merge pull request #982 from porter-dev/master

Kube HPA typo -> staging

docs/developing/test-autoscaling.md

@@ -0,0 +1,18 @@
+# Test Cluster and HPA Autoscaling
+
+Prerequisites: 
+- [`metrics-server`](https://artifacthub.io/packages/helm/bitnami/metrics-server) must be installed (installed by default on all Porter clusters). 
+- [`cluster-autoscaler`](https://github.com/kubernetes/autoscaler/tree/master/cluster-autoscaler) must be enabled (enabled by default on all Porter clusters). 
+- Have `kubectl` access to the cluster, as it is easiest to port-forward to the exposed service. 
+- Download [`hey`](https://github.com/rakyll/hey) or an equivalent. 
+
+# Steps
+
+1. Launch a Docker template with image URL `k8s.gcr.io/hpa-example`. Enable autoscaling under the **Resources** tab -- suggested params are 100 Mi RAM, 100m CPU, 1-100 replicas, 50% CPU & RAM util. Do not **Expose to external traffic** to run load tests.
+
+2. Confirm that the horizontal pod autoscaler was created via kubectl (`kubectl get hpa`) and that the metrics server is reporting metrics via `kubectl top pods`. Current utilization may initially show up as `<undefined>`, but this information should load after a minute or so.
+
+3. Port-forward to the service via `kubectl port-forward svc/<svc-name> 10000:80`. 
+4. In a different terminal window, run `hey http://localhost:10000`. Vary the `hey` parameters to test autoscaling under different loads (`hey -h` for options). 
+
+Check that replicas scale when passing the threshold from the metrics tab. Use kubectl to probe the nodes and confirm cluster-level upscaling has also occurred. Delete the chart and confirm node downscaling when done.

internal/kubernetes/prometheus/metrics.go

@@ -374,7 +374,7 @@ type promRawValuesQuery struct {
 }
 
 // getKubeHPAMetricName performs a "best guess" for the name of the kube HPA metric,
-// which was renamed to kube_horizontal_pod_autoscaler... in later versions of kube-state-metrics.
+// which was renamed to kube_horizontalpodautoscaler... in later versions of kube-state-metrics.
 // we query Prometheus for a list of metric names to see if any match the new query
 // value, otherwise we return the deprecated name.
 func getKubeHPAMetricName(
@@ -384,7 +384,7 @@ func getKubeHPAMetricName(
 	suffix string,
 ) string {
 	queryParams := map[string]string{
-		"match[]": fmt.Sprintf("kube_horizontal_pod_autoscaler_%s", suffix),
+		"match[]": fmt.Sprintf("kube_horizontalpodautoscaler_%s", suffix),
 		"start":   fmt.Sprintf("%d", opts.StartRange),
 		"end":     fmt.Sprintf("%d", opts.EndRange),
 	}
@@ -408,7 +408,7 @@ func getKubeHPAMetricName(
 	json.Unmarshal(rawQuery, rawQueryObj)
 
 	if rawQueryObj.Status == "success" && len(rawQueryObj.Data) == 1 {
-		return fmt.Sprintf("kube_horizontal_pod_autoscaler_%s", suffix)
+		return fmt.Sprintf("kube_horizontalpodautoscaler_%s", suffix)
 	}
 
 	return fmt.Sprintf("kube_hpa_%s", suffix)