cloud, plz: add public detail and abort to events

api/v1alpha1/testruni.go

@@ -20,3 +20,14 @@ type TestRunI interface {
 	GetSpec() *TestRunSpec
 	NamespacedName() types.NamespacedName
 }
+
+// TestRunID is a tiny helper to get k6 Cloud test run ID.
+// PLZ test run will have test run ID as part of spec
+// while cloud output test run as part of status.
+func TestRunID(k6 TestRunI) string {
+	specId := k6.GetSpec().TestRunID
+	if len(specId) > 0 {
+		return specId
+	}
+	return k6.GetStatus().TestRunID
+}

controllers/common.go

@@ -46,7 +46,7 @@ func inspectTestRun(ctx context.Context, log logr.Logger, k6 v1alpha1.TestRunI,
 	}
 
 	// there should be only 1 initializer pod
-	if podList.Items[0].Status.Phase != "Succeeded" {
+	if podList.Items[0].Status.Phase != corev1.PodSucceeded && podList.Items[0].Status.Phase != corev1.PodFailed {
 		log.Info("Waiting for initializing pod to finish")
 		return
 	}

controllers/k6_create.go

@@ -40,14 +40,8 @@ func CreateJobs(ctx context.Context, log logr.Logger, k6 v1alpha1.TestRunI, r *T
 		token, tokenReady, err = loadToken(ctx, log, r.Client, k6.GetSpec().Token, sOpts)
 		if err != nil {
 			// An error here means a very likely mis-configuration of the token.
-			// Consider updating status to error to let a user know quicker?
+			// TODO: update status to error to let a user know quicker
 			log.Error(err, "A problem while getting token.")
-
-			if v1alpha1.IsTrue(k6, v1alpha1.CloudTestRun) {
-				events := cloud.ErrorEvent(cloud.K6OperatorStartError).WithDetail(fmt.Sprintf("Failed to retrieve token: %v", err))
-				cloud.SendTestRunEvents(r.k6CloudClient, k6.GetSpec().TestRunID, log, events)
-			}
-
 			return ctrl.Result{}, nil
 		}
 		if !tokenReady {
@@ -58,10 +52,11 @@ func CreateJobs(ctx context.Context, log logr.Logger, k6 v1alpha1.TestRunI, r *T
 	log.Info("Creating test jobs")
 
 	if res, err = createJobSpecs(ctx, log, k6, r, token); err != nil {
-
 		if v1alpha1.IsTrue(k6, v1alpha1.CloudTestRun) {
-			events := cloud.ErrorEvent(cloud.K6OperatorStartError).WithDetail(fmt.Sprintf("Failed to create runner jobs: %v", err))
-			cloud.SendTestRunEvents(r.k6CloudClient, k6.GetSpec().TestRunID, log, events)
+			events := cloud.ErrorEvent(cloud.K6OperatorStartError).
+				WithDetail(fmt.Sprintf("Failed to create runner jobs: %v", err)).
+				WithAbort()
+			cloud.SendTestRunEvents(r.k6CloudClient, v1alpha1.TestRunID(k6), log, events)
 		}
 
 		return res, err
@@ -86,7 +81,11 @@ func createJobSpecs(ctx context.Context, log logr.Logger, k6 v1alpha1.TestRunI,
 	}
 
 	if err := r.Get(ctx, namespacedName, found); err == nil || !errors.IsNotFound(err) {
-		log.Info("Could not start a new test, Make sure you've deleted your previous run.")
+		if err == nil {
+			err = fmt.Errorf("job with the name %s exists; make sure you've deleted your previous run", namespacedName.Name)
+		}
+
+		log.Info(err.Error())
 		return ctrl.Result{}, err
 	}
 

controllers/k6_finish.go

@@ -54,8 +54,10 @@ func FinishJobs(ctx context.Context, log logr.Logger, k6 v1alpha1.TestRunI, r *T
 	log.Info(msg)
 
 	if v1alpha1.IsTrue(k6, v1alpha1.CloudTestRun) && failed > 0 {
-		events := cloud.ErrorEvent(cloud.K6OperatorRunnerError).WithDetail(msg)
-		cloud.SendTestRunEvents(r.k6CloudClient, k6.GetSpec().TestRunID, log, events)
+		events := cloud.ErrorEvent(cloud.K6OperatorRunnerError).
+			WithDetail(msg).
+			WithAbort()
+		cloud.SendTestRunEvents(r.k6CloudClient, v1alpha1.TestRunID(k6), log, events)
 	}
 
 	if finished < k6.GetSpec().Parallelism {

controllers/k6_initialize.go

@@ -52,10 +52,14 @@ func RunValidations(ctx context.Context, log logr.Logger, k6 v1alpha1.TestRunI,
 
 	inspectOutput, inspectReady, err := inspectTestRun(ctx, log, k6, r.Client)
 	if err != nil {
+		// Cloud output test run is not created yet at this point, so sending
+		// events is possible only for PLZ test run.
 		if v1alpha1.IsTrue(k6, v1alpha1.CloudPLZTestRun) {
 			// This error won't allow to start a test so let k6 Cloud know of it
-			events := cloud.ErrorEvent(cloud.K6OperatorStartError).WithDetail(fmt.Sprintf("Failed to inspect the test script: %v", err))
-			cloud.SendTestRunEvents(r.k6CloudClient, k6.GetSpec().TestRunID, log, events)
+			events := cloud.ErrorEvent(cloud.K6OperatorStartError).
+				WithDetail(fmt.Sprintf("Failed to inspect the test script: %v", err)).
+				WithAbort()
+			cloud.SendTestRunEvents(r.k6CloudClient, v1alpha1.TestRunID(k6), log, events)
 		}
 
 		// inspectTestRun made a log message already so just return without requeue

controllers/k6_start.go

@@ -71,8 +71,10 @@ func StartJobs(ctx context.Context, log logr.Logger, k6 v1alpha1.TestRunI, r *Te
 			// let's try this approach
 			if time.Since(t).Minutes() > 5 {
 				if v1alpha1.IsTrue(k6, v1alpha1.CloudTestRun) {
-					events := cloud.ErrorEvent(cloud.K6OperatorStartError).WithDetail("Creation of runner pods takes too long: perhaps, something is off with your configuration. Check if runner jobs and pods were created successfully.")
-					cloud.SendTestRunEvents(r.k6CloudClient, k6.GetSpec().TestRunID, log, events)
+					events := cloud.ErrorEvent(cloud.K6OperatorStartError).
+						WithDetail("Creation of runner pods takes too long: your configuration might be off. Check if runner jobs and pods were created successfully.").
+						WithAbort()
+					cloud.SendTestRunEvents(r.k6CloudClient, v1alpha1.TestRunID(k6), log, events)
 				}
 			}
 		}

controllers/testrun_controller.go

@@ -82,7 +82,7 @@ func (r *TestRunReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ct
 
 func (r *TestRunReconciler) reconcile(ctx context.Context, req ctrl.Request, log logr.Logger, k6 v1alpha1.TestRunI) (ctrl.Result, error) {
 	var err error
-	if v1alpha1.IsTrue(k6, v1alpha1.CloudPLZTestRun) {
+	if v1alpha1.IsTrue(k6, v1alpha1.CloudTestRun) {
 		// bootstrap the client
 		found, err := r.createClient(ctx, k6, log)
 		if err != nil {

pkg/cloud/test_runs.go

@@ -110,7 +110,7 @@ func GetTestRunData(client *cloudapi.Client, refID string) (*TestRunData, error)
 	return getTestRun(client, url)
 }
 
-// called by K6 controller
+// called by TestRun controller
 func GetTestRunState(client *cloudapi.Client, refID string, log logr.Logger) (TestRunStatus, error) {
 	url := fmt.Sprintf("%s/loadtests/v4/test_runs(%s)?$select=id,run_status", ApiURL(client.BaseURL()), refID)
 	trData, err := getTestRun(client, url)
@@ -121,7 +121,7 @@ func GetTestRunState(client *cloudapi.Client, refID string, log logr.Logger) (Te
 	return TestRunStatus(trData.RunStatus), nil
 }
 
-// called by K6 controller
+// called by TestRun controller
 // If there's an error, it'll be logged.
 func SendTestRunEvents(client *cloudapi.Client, refID string, log logr.Logger, events *Events) {
 	if len(*events) == 0 {

pkg/cloud/types.go

@@ -76,10 +76,14 @@ type EventPayload struct {
 }
 
 type Event struct {
-	ErrorCode `json:"error_code,omitempty"`
-	Detail    string `json:"error_detail,omitempty"`
 	Origin    `json:"origin,omitempty"`
-	Reason    string `json:"reason,omitempty"`
+	ErrorCode `json:"error_code,omitempty"`
+
+	// reason is used for abort events,
+	// while details are for any non-abort event
+	Reason       string `json:"reason,omitempty"`
+	Detail       string `json:"error_detail,omitempty"`
+	PublicDetail string `json:"error_detail_public,omitempty"`
 }
 
 type EventType string
@@ -111,7 +115,8 @@ var (
 	OriginK6   = Origin("k6")
 )
 
-// WithDetail sets detail only for the 1st event
+// WithDetail sets detail only for the 1st event.
+// If it's abort, WithDetail sets reason field.
 func (e *Events) WithDetail(s string) *Events {
 	if len(*e) == 0 {
 		return e
@@ -121,32 +126,31 @@ func (e *Events) WithDetail(s string) *Events {
 		(*e)[0].Reason = s
 	} else {
 		(*e)[0].Detail = s
+		(*e)[0].PublicDetail = s
 	}
 	return e
 }
 
-// WithAbort adds abortEvent if errorEvent already exists
-// func (e *Events) WithAbort(s string) *Events {
-// 	if len(*e) == 0 {
-// 		return
-// 	}
-
-// 	if (*e)[0].EventType == errorEvent {
-// 		ae := EventPayload{
-// 			EventType: abort,
-// 		}
-// 	}
-//	return e
-// }
+// WithAbort adds abortEvent to errorEvent if it already exists.
+func (e *Events) WithAbort() *Events {
+	if len(*e) == 0 {
+		return e
+	}
 
-func AbortEvent(o Origin) *Events {
-	e := Events([]*EventPayload{{
+	if (*e)[0].EventType == errorEvent {
+		*e = append(*e, AbortEvent(OriginUser))
+	}
+	return e
+}
+
+func AbortEvent(o Origin) *EventPayload {
+	e := &EventPayload{
 		EventType: abortEvent,
 		Event: Event{
 			Origin: o,
 		},
-	}})
-	return &e
+	}
+	return e
 }
 
 func ErrorEvent(ec ErrorCode) *Events {