validate studyJob when first reconcile it

pkg/controller/studyjob/katib_api_util.go

@@ -16,7 +16,6 @@ package studyjob
 
 import (
 	"context"
-	"fmt"
 	"log"
 
 	"github.com/kubeflow/katib/pkg"
@@ -26,13 +25,15 @@ import (
 )
 
 func initializeStudy(instance *katibv1alpha1.StudyJob, ns string) error {
-	if instance.Spec.SuggestionSpec == nil {
+	if validErr := validateStudy(instance, ns); validErr != nil {
 		instance.Status.Condition = katibv1alpha1.ConditionFailed
-		return fmt.Errorf("No Spec.SuggestionSpec specified.")
+		return validErr
 	}
+
 	if instance.Spec.SuggestionSpec.SuggestionAlgorithm == "" {
 		instance.Spec.SuggestionSpec.SuggestionAlgorithm = "random"
 	}
+
 	instance.Status.Condition = katibv1alpha1.ConditionRunning
 
 	conn, err := grpc.Dial(pkg.ManagerAddr, grpc.WithInsecure())

pkg/controller/studyjob/manifest_parser.go

@@ -25,6 +25,7 @@ import (
 	katibv1alpha1 "github.com/kubeflow/katib/pkg/api/operators/apis/studyjob/v1alpha1"
 	"github.com/kubeflow/katib/pkg/manager/studyjobclient"
 
+	"k8s.io/apimachinery/pkg/util/uuid"
 	k8syaml "k8s.io/apimachinery/pkg/util/yaml"
 )
 
@@ -103,7 +104,7 @@ func getWorkerManifest(c katibapi.ManagerClient, studyID string, trial *katibapi
 	}
 	var wid string
 	if dryrun {
-		wid = "validation"
+		wid = string(uuid.NewUUID())
 	} else {
 		cwreq := &katibapi.RegisterWorkerRequest{
 			Worker: &katibapi.Worker{

pkg/controller/studyjob/utils.go

@@ -12,11 +12,19 @@ limitations under the License.
 package studyjob
 
 import (
+	"fmt"
+	"log"
+
+	katibapi "github.com/kubeflow/katib/pkg/api"
+	katibv1alpha1 "github.com/kubeflow/katib/pkg/api/operators/apis/studyjob/v1alpha1"
 	pytorchjobv1beta1 "github.com/kubeflow/pytorch-operator/pkg/apis/pytorch/v1beta1"
 	tfjobv1beta1 "github.com/kubeflow/tf-operator/pkg/apis/tensorflow/v1beta1"
 
 	batchv1 "k8s.io/api/batch/v1"
+	batchv1beta "k8s.io/api/batch/v1beta1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/runtime"
+	k8syaml "k8s.io/apimachinery/pkg/util/yaml"
 )
 
 func createWorkerJobObj(kind string) runtime.Object {
@@ -30,3 +38,61 @@ func createWorkerJobObj(kind string) runtime.Object {
 	}
 	return nil
 }
+
+func validateStudy(instance *katibv1alpha1.StudyJob, namespace string) error {
+	if instance.Spec.SuggestionSpec == nil {
+		return fmt.Errorf("No Spec.SuggestionSpec specified.")
+	}
+	BUFSIZE := 1024
+	wkind, err := getWorkerKind(instance.Spec.WorkerSpec)
+	if err != nil {
+		log.Printf("getWorkerKind error %v", err)
+		return err
+	}
+
+	studyID := "studyID4Validation"
+	trialID := "trialID4Validation"
+	workerID, wm, err := getWorkerManifest(
+		nil,
+		studyID,
+		&katibapi.Trial{
+			TrialId:      trialID,
+			ParameterSet: []*katibapi.Parameter{},
+		},
+		instance.Spec.WorkerSpec,
+		wkind,
+		namespace,
+		true,
+	)
+	if err != nil {
+		return err
+	}
+
+	job := createWorkerJobObj(wkind)
+	if err := k8syaml.NewYAMLOrJSONDecoder(wm, BUFSIZE).Decode(job); err != nil {
+		log.Printf("Yaml decode error %v", err)
+		return err
+	}
+
+	metav1Job := job.(metav1.Object)
+	if metav1Job.GetNamespace() != namespace || metav1Job.GetName() != workerID {
+		return fmt.Errorf("Invalid worker template.")
+	}
+
+	var mcjob batchv1beta.CronJob
+	mcm, err := getMetricsCollectorManifest(studyID, trialID, workerID, wkind, namespace, instance.Spec.MetricsCollectorSpec)
+	if err != nil {
+		log.Printf("getMetricsCollectorManifest error %v", err)
+		return err
+	}
+
+	if err := k8syaml.NewYAMLOrJSONDecoder(mcm, BUFSIZE).Decode(&mcjob); err != nil {
+		log.Printf("MetricsCollector Yaml decode error %v", err)
+		return err
+	}
+
+	if mcjob.GetNamespace() != namespace || mcjob.GetName() != workerID {
+		return fmt.Errorf("Invalid metricsCollector template.")
+	}
+	return nil
+}