Add metric that is emitted after a successful single run of static

autoscaler.

This will allow to detect situations where loop starts and not ends (because of a health check for example).

Note that existing loop start metric is not enough because if we have
autoscaler killed another autoscaler will take over role of the leader
and again emit the metric of the main starting.

Author: jbtk

cluster-autoscaler/loop/run.go

@@ -37,7 +37,9 @@ func RunAutoscalerOnce(autoscaler autoscaler, healthCheck *metrics.HealthCheck,
 	if err != nil && err.Type() != errors.TransientError {
 		metrics.RegisterError(err)
 	} else {
-		healthCheck.UpdateLastSuccessfulRun(time.Now())
+		var successTime = time.Now()
+		healthCheck.UpdateLastSuccessfulRun(successTime)
+		metrics.UpdateLastTime(metrics.MainSuccessful, successTime)
 	}
 
 	metrics.UpdateDurationFromStart(metrics.Main, loopStart)

cluster-autoscaler/metrics/metrics.go

@@ -110,6 +110,7 @@ const (
 	FilterOutSchedulable       FunctionLabel = "filterOutSchedulable"
 	CloudProviderRefresh       FunctionLabel = "cloudProviderRefresh"
 	Main                       FunctionLabel = "main"
+	MainSuccessful             FunctionLabel = "mainSuccessful"
 	Poll                       FunctionLabel = "poll"
 	Reconfigure                FunctionLabel = "reconfigure"
 	Autoscaling                FunctionLabel = "autoscaling"