working version

Signed-off-by: odubajDT <[email protected]>

operator/api/v1alpha1/common/common.go

@@ -50,8 +50,10 @@ const PreDeploymentCheckType CheckType = "pre"
 const PostDeploymentCheckType CheckType = "post"
 
 type KeptnMeters struct {
-	DeploymentCount    syncint64.Counter
 	TaskCount          syncint64.Counter
 	TaskDuration       syncfloat64.Histogram
+	TaskActive         syncint64.UpDownCounter
+	DeploymentCount    syncint64.Counter
 	DeploymentDuration syncfloat64.Histogram
+	DeploymentActive   syncint64.UpDownCounter
 }

operator/api/v1alpha1/keptntask_types.go

@@ -20,6 +20,7 @@ import (
 	"time"
 
 	"github.com/keptn-sandbox/lifecycle-controller/operator/api/v1alpha1/common"
+	"go.opentelemetry.io/otel/attribute"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
 
@@ -96,3 +97,32 @@ func (i *KeptnTask) SetEndTime() {
 		i.Status.EndTime = metav1.NewTime(time.Now().UTC())
 	}
 }
+
+func (i *KeptnTask) IsStartTimeSet() bool {
+	return !i.Status.StartTime.IsZero()
+}
+
+func (i *KeptnTask) IsEndTimeSet() bool {
+	return !i.Status.EndTime.IsZero()
+}
+
+func (i KeptnTask) GetActiveMetricsAttributes() []attribute.KeyValue {
+	return []attribute.KeyValue{
+		attribute.Key("KeptnApp").String(i.Spec.AppName),
+		attribute.Key("KeptnWorkload").String(i.Spec.Workload),
+		attribute.Key("KeptnVersion").String(i.Spec.WorkloadVersion),
+		attribute.Key("TaskName").String(i.Name),
+		attribute.Key("Type").String(string(i.Spec.Type)),
+	}
+}
+
+func (i KeptnTask) GetMetricsAttributes() []attribute.KeyValue {
+	return []attribute.KeyValue{
+		attribute.Key("KeptnApp").String(i.Spec.AppName),
+		attribute.Key("KeptnWorkload").String(i.Spec.Workload),
+		attribute.Key("KeptnVersion").String(i.Spec.WorkloadVersion),
+		attribute.Key("TaskName").String(i.Name),
+		attribute.Key("Type").String(string(i.Spec.Type)),
+		attribute.Key("Status").String(string(i.Status.Status)),
+	}
+}

operator/api/v1alpha1/keptnworkloadinstance_types.go

@@ -20,6 +20,7 @@ import (
 	"time"
 
 	"github.com/keptn-sandbox/lifecycle-controller/operator/api/v1alpha1/common"
+	"go.opentelemetry.io/otel/attribute"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
 
@@ -109,6 +110,14 @@ func (i *KeptnWorkloadInstance) SetEndTime() {
 	}
 }
 
+func (i *KeptnWorkloadInstance) IsStartTimeSet() bool {
+	return !i.Status.StartTime.IsZero()
+}
+
+func (i *KeptnWorkloadInstance) IsEndTimeSet() bool {
+	return !i.Status.EndTime.IsZero()
+}
+
 func (i *WorkloadTaskStatus) SetStartTime() {
 	if i.StartTime.IsZero() {
 		i.StartTime = metav1.NewTime(time.Now().UTC())
@@ -120,3 +129,22 @@ func (i *WorkloadTaskStatus) SetEndTime() {
 		i.EndTime = metav1.NewTime(time.Now().UTC())
 	}
 }
+
+func (i KeptnWorkloadInstance) GetActiveMetricsAttributes() []attribute.KeyValue {
+	return []attribute.KeyValue{
+		attribute.Key("KeptnApp").String(i.Spec.AppName),
+		attribute.Key("KeptnWorkload").String(i.Spec.WorkloadName),
+		attribute.Key("KeptnVersion").String(i.Spec.Version),
+		attribute.Key("Namespace").String(i.Namespace),
+	}
+}
+
+func (i KeptnWorkloadInstance) GetMetricsAttributes() []attribute.KeyValue {
+	return []attribute.KeyValue{
+		attribute.Key("KeptnApp").String(i.Spec.AppName),
+		attribute.Key("KeptnWorkload").String(i.Spec.WorkloadName),
+		attribute.Key("KeptnVersion").String(i.Spec.Version),
+		attribute.Key("Namespace").String(i.Namespace),
+		attribute.Key("Status").String(string(i.Status.PostDeploymentStatus)),
+	}
+}

operator/config/manager/kustomization.yaml

@@ -11,4 +11,4 @@ kind: Kustomization
 images:
 - name: controller
   newName: docker.io/odubajdt/keptn-lifecycle-operator
-  newTag: metrics6
+  newTag: metrics10

operator/controllers/keptntask/controller.go

@@ -24,7 +24,6 @@ import (
 	"github.com/go-logr/logr"
 	klcv1alpha1 "github.com/keptn-sandbox/lifecycle-controller/operator/api/v1alpha1"
 	"github.com/keptn-sandbox/lifecycle-controller/operator/api/v1alpha1/common"
-	"go.opentelemetry.io/otel/attribute"
 	batchv1 "k8s.io/api/batch/v1"
 	"k8s.io/apimachinery/pkg/api/errors"
 	"k8s.io/apimachinery/pkg/runtime"
@@ -64,6 +63,10 @@ func (r *KeptnTaskReconciler) Reconcile(ctx context.Context, req ctrl.Request) (
 		return ctrl.Result{Requeue: true, RequeueAfter: 30 * time.Second}, nil
 	}
 
+	if !task.IsStartTimeSet() {
+		// metrics: increment active task counter
+		r.Meters.TaskActive.Add(ctx, 1, task.GetActiveMetricsAttributes()...)
+	}
 	task.SetStartTime()
 
 	err := r.Client.Status().Update(ctx, task)
@@ -97,21 +100,18 @@ func (r *KeptnTaskReconciler) Reconcile(ctx context.Context, req ctrl.Request) (
 
 	// WorkloadInstance is completed at this place
 
+	if !task.IsEndTimeSet() {
+		// metrics: decrement active task counter
+		r.Meters.TaskActive.Add(ctx, -1, task.GetActiveMetricsAttributes()...)
+	}
 	task.SetEndTime()
 
 	err = r.Client.Status().Update(ctx, task)
 	if err != nil {
 		return ctrl.Result{Requeue: true}, err
 	}
 
-	attrs := []attribute.KeyValue{
-		attribute.Key("KeptnApp").String(task.Spec.AppName),
-		attribute.Key("KeptnWorkload").String(task.Spec.Workload),
-		attribute.Key("KeptnVersion").String(task.Spec.WorkloadVersion),
-		attribute.Key("TaskName").String(task.Name),
-		attribute.Key("Type").String(string(task.Spec.Type)),
-		attribute.Key("Status").String(string(task.Status.Status)),
-	}
+	attrs := task.GetMetricsAttributes()
 
 	r.Log.Info("Increasing task count")
 

operator/controllers/keptnworkloadinstance/controller.go

@@ -24,7 +24,6 @@ import (
 
 	"github.com/go-logr/logr"
 	"github.com/google/uuid"
-	"go.opentelemetry.io/otel/attribute"
 	"k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
@@ -93,6 +92,10 @@ func (r *KeptnWorkloadInstanceReconciler) Reconcile(ctx context.Context, req ctr
 		return reconcile.Result{}, fmt.Errorf("could not fetch KeptnWorkloadInstance: %+v", err)
 	}
 
+	if !workloadInstance.IsStartTimeSet() {
+		// metrics: increment active deployment counter
+		r.Meters.DeploymentActive.Add(ctx, 1, workloadInstance.GetActiveMetricsAttributes()...)
+	}
 	workloadInstance.SetStartTime()
 
 	if !workloadInstance.IsPreDeploymentCompleted() {
@@ -127,20 +130,18 @@ func (r *KeptnWorkloadInstanceReconciler) Reconcile(ctx context.Context, req ctr
 
 	// WorkloadInstance is completed at this place
 
+	if !workloadInstance.IsEndTimeSet() {
+		// metrics: decrement active deployment counter
+		r.Meters.DeploymentActive.Add(ctx, -1, workloadInstance.GetActiveMetricsAttributes()...)
+	}
 	workloadInstance.SetEndTime()
 
 	err = r.Client.Status().Update(ctx, workloadInstance)
 	if err != nil {
 		return ctrl.Result{Requeue: true}, err
 	}
 
-	attrs := []attribute.KeyValue{
-		attribute.Key("KeptnApp").String(workloadInstance.Spec.AppName),
-		attribute.Key("KeptnWorkload").String(workloadInstance.Spec.WorkloadName),
-		attribute.Key("KeptnVersion").String(workloadInstance.Spec.Version),
-		attribute.Key("Namespace").String(workloadInstance.Namespace),
-		attribute.Key("Status").String(string(workloadInstance.Status.PostDeploymentStatus)),
-	}
+	attrs := workloadInstance.GetMetricsAttributes()
 
 	r.Log.Info("Increasing deployment count")
 

operator/main.go

@@ -92,6 +92,10 @@ func main() {
 	if err != nil {
 		setupLog.Error(err, "unable to start OTel")
 	}
+	deploymentActive, err := meter.SyncInt64().UpDownCounter("keptn.deployment.active", instrument.WithDescription("a simple counter of active deployments for Keptn deployment"))
+	if err != nil {
+		setupLog.Error(err, "unable to start OTel")
+	}
 	taskCount, err := meter.SyncInt64().Counter("keptn.task.count", instrument.WithDescription("a simple counter for Keptn tasks"))
 	if err != nil {
 		setupLog.Error(err, "unable to start OTel")
@@ -100,12 +104,18 @@ func main() {
 	if err != nil {
 		setupLog.Error(err, "unable to start OTel")
 	}
+	taskActive, err := meter.SyncInt64().UpDownCounter("keptn.task.active", instrument.WithDescription("a simple counter of active tasks for Keptn tasks"))
+	if err != nil {
+		setupLog.Error(err, "unable to start OTel")
+	}
 
 	meters := common.KeptnMeters{
-		DeploymentCount:    deploymentCount,
 		TaskCount:          taskCount,
 		TaskDuration:       taskDuration,
+		TaskActive:         taskActive,
+		DeploymentCount:    deploymentCount,
 		DeploymentDuration: deploymentDuration,
+		DeploymentActive:   deploymentActive,
 	}
 
 	// Start the prometheus HTTP server and pass the exporter Collector to it