CPU and GPU schedule type jobs

pkg/apis/ml/v1alpha1/batchjob.go

@@ -21,13 +21,9 @@
 package v1alpha1
 
 import (
-	"strings"
-
 	meta "k8s.io/apimachinery/pkg/apis/meta/v1"
 
 	"github.com/arangodb/kube-arangodb/pkg/apis/ml"
-	mlShared "github.com/arangodb/kube-arangodb/pkg/handlers/enterprise/ml/shared"
-	"github.com/arangodb/kube-arangodb/pkg/util/errors"
 )
 
 // +k8s:deepcopy-gen:interfaces=k8s.io/apimachinery/pkg/runtime.Object
@@ -71,57 +67,3 @@ func (a *ArangoMLBatchJob) GetStatus() ArangoMLBatchJobStatus {
 func (a *ArangoMLBatchJob) SetStatus(status ArangoMLBatchJobStatus) {
 	a.Status = status
 }
-
-func (a *ArangoMLBatchJob) GetJobType() string {
-	val, ok := a.Labels[mlShared.MLJobTypeLabel]
-	if !ok {
-		return ""
-	}
-	return strings.ToLower(val)
-}
-
-func (a *ArangoMLBatchJob) GetScheduleType() string {
-	val, ok := a.Labels[mlShared.MLJobScheduleLabel]
-	if !ok {
-		return ""
-	}
-	return strings.ToLower(val)
-}
-
-func (a *ArangoMLBatchJob) GetMLDeploymentName() string {
-	val, ok := a.Labels[mlShared.MLJobScheduleLabel]
-	if !ok {
-		return ""
-	}
-	return val
-}
-
-func (a *ArangoMLBatchJob) ValidateLabels() error {
-	depl, ok := a.Labels[mlShared.MLDeploymentLabel]
-	if !ok {
-		return errors.Newf("Job missing label: %s", mlShared.MLDeploymentLabel)
-	}
-	if depl == "" {
-		return errors.Newf("Job empty value for label: %s", mlShared.MLDeploymentLabel)
-	}
-
-	t, ok := a.Labels[mlShared.MLJobTypeLabel]
-	if !ok {
-		return errors.Newf("Job missing label: %s", mlShared.MLJobTypeLabel)
-	}
-	jobType := strings.ToLower(t)
-	if jobType != mlShared.MLJobTrainingType && jobType != mlShared.MLJobPredictionType {
-		return errors.Newf("Job label (%s) has unexpected value: %s", mlShared.MLJobTypeLabel, t)
-	}
-
-	s, ok := a.Labels[mlShared.MLJobScheduleLabel]
-	if !ok {
-		return errors.Newf("Job missing label: %s", mlShared.MLJobTypeLabel)
-	}
-	scheduleType := strings.ToLower(s)
-	if scheduleType != mlShared.MLJobScheduleCPU && scheduleType != mlShared.MLJobScheduleGPU {
-		return errors.Newf("Job label (%s) has unexpected value: %s", mlShared.MLJobScheduleLabel, s)
-	}
-
-	return nil
-}

pkg/apis/ml/v1alpha1/batchjob_spec.go

@@ -40,7 +40,7 @@ func (a *ArangoMLBatchJobSpec) Validate() error {
 
 	var err []error
 	if a.JobSpec == nil {
-		err = append(err, shared.PrefixResourceErrors("spec", errors.Newf("JobSpec is not defined")))
+		return shared.PrefixResourceErrors("spec", errors.Newf("JobSpec is not defined"))
 	}
 
 	if len(a.JobSpec.Template.Spec.Containers) != 1 {

pkg/apis/ml/v1alpha1/cronjob_spec.go

@@ -40,7 +40,7 @@ func (a *ArangoMLCronJobSpec) Validate() error {
 
 	var err []error
 	if a.CronJobSpec == nil {
-		err = append(err, shared.PrefixResourceErrors("spec", errors.Newf("CronJobSpec is not defined")))
+		return shared.PrefixResourceErrors("spec", errors.Newf("CronJobSpec is not defined"))
 	}
 
 	if len(a.CronJobSpec.JobTemplate.Spec.Template.Spec.Containers) != 1 {

pkg/apis/ml/v1alpha1/extension_conditions.go

@@ -23,6 +23,7 @@ package v1alpha1
 import api "github.com/arangodb/kube-arangodb/pkg/apis/deployment/v1"
 
 const (
+	MLExtensionFoundCondition              api.ConditionType = "MLExtensionFound"
 	ExtensionStorageFoundCondition         api.ConditionType = "StorageFound"
 	ExtensionDeploymentFoundCondition      api.ConditionType = "DeploymentFound"
 	ExtensionBootstrapCompletedCondition   api.ConditionType = "BootstrapCompleted"

pkg/apis/ml/v1alpha1/extension_spec_job.go

@@ -25,29 +25,78 @@ import (
 	sharedApi "github.com/arangodb/kube-arangodb/pkg/apis/shared/v1"
 )
 
+type JobType string
+
+const (
+	MLJobTrainingType   JobType = "training"
+	MLJobPredictionType JobType = "prediction"
+)
+
 type ArangoMLJobsTemplates struct {
 	// Prediction defines template for the prediction job
-	Prediction map[string]*ArangoMLExtensionTemplateSpec `json:"prediction,omitempty"`
+	Prediction *ArangoMLJobTemplates `json:"prediction,omitempty"`
 
 	// Training defines template for the training job
-	Training map[string]*ArangoMLExtensionTemplateSpec `json:"training,omitempty"`
+	Training *ArangoMLJobTemplates `json:"training,omitempty"`
+}
+
+func (a *ArangoMLJobsTemplates) GetJobTemplates(jobType JobType) *ArangoMLJobTemplates {
+	switch jobType {
+	case MLJobTrainingType:
+		return a.Prediction
+	case MLJobPredictionType:
+		return a.Training
+	default:
+		return nil
+	}
 }
 
-func (j *ArangoMLJobsTemplates) Validate() error {
-	if j == nil {
+func (a *ArangoMLJobsTemplates) Validate() error {
+	if a == nil {
 		return nil
 	}
 
-	var errs []error
-	for _, template := range j.Prediction {
-		errs = append(errs, shared.PrefixResourceErrors("prediction", template.Validate()))
+	return shared.WithErrors(
+		shared.PrefixResourceErrors("prediction", a.Prediction.Validate()),
+		shared.PrefixResourceErrors("training", a.Training.Validate()),
+	)
+}
+
+type JobScheduleType string
+
+const (
+	MLJobScheduleCPU JobScheduleType = "cpu"
+	MLJobScheduleGPU JobScheduleType = "gpu"
+)
+
+type ArangoMLJobTemplates struct {
+	// CPU defines templates for CPU jobs
+	CPU *ArangoMLExtensionTemplateSpec `json:"cpu,omitempty"`
+
+	// GPU defines templates for GPU jobs
+	GPU *ArangoMLExtensionTemplateSpec `json:"gpu,omitempty"`
+}
+
+func (a *ArangoMLJobTemplates) GetJobTemplateSpec(scheduleType JobScheduleType) *ArangoMLExtensionTemplateSpec {
+	switch scheduleType {
+	case MLJobScheduleCPU:
+		return a.CPU
+	case MLJobScheduleGPU:
+		return a.GPU
+	default:
+		return nil
 	}
+}
 
-	for _, template := range j.Training {
-		errs = append(errs, shared.PrefixResourceErrors("training", template.Validate()))
+func (a *ArangoMLJobTemplates) Validate() error {
+	if a == nil {
+		return nil
 	}
 
-	return shared.WithErrors(errs...)
+	return shared.WithErrors(
+		shared.PrefixResourceErrors("cpu", a.CPU.Validate()),
+		shared.PrefixResourceErrors("gpu", a.GPU.Validate()),
+	)
 }
 
 type ArangoMLExtensionTemplateSpec struct {