Enable job suspend for Kueue

Signed-off-by: ted chang <[email protected]>

api/lmes/v1alpha1/lmevaljob_types.go

@@ -28,7 +28,7 @@ import (
 // NOTE: json tags are required.  Any new fields you add must have json tags for the fields to be serialized.
 
 // Represent a job's status
-// +kubebuilder:validation:Enum=New;Scheduled;Running;Complete;Cancelled
+// +kubebuilder:validation:Enum=New;Scheduled;Running;Complete;Cancelled;Suspended
 type JobState string
 
 const (
@@ -42,6 +42,8 @@ const (
 	CompleteJobState JobState = "Complete"
 	// The job is cancelled
 	CancelledJobState JobState = "Cancelled"
+	// The job is suspended
+	SuspendedJobState JobState = "Suspended"
 )
 
 // +kubebuilder:validation:Enum=NoReason;Succeeded;Failed;Cancelled
@@ -236,6 +238,9 @@ type LMEvalJobSpec struct {
 	// Specify extra information for the lm-eval job's pod
 	// +optional
 	Pod *LMEvalPodSpec `json:"pod,omitempty"`
+	// Suspend keeps the job but without pods. This is intended to be used by the Kueue integration
+	// +optional
+	Suspend bool `json:"suspend,omitempty"`
 }
 
 // LMEvalJobStatus defines the observed state of LMEvalJob

config/crd/bases/trustyai.opendatahub.io_trustyaiservices.yaml

@@ -3,8 +3,7 @@ apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
   annotations:
-    controller-gen.kubebuilder.io/version: v0.11.1
-  creationTimestamp: null
+    controller-gen.kubebuilder.io/version: v0.16.3
   name: trustyaiservices.trustyai.opendatahub.io
 spec:
   group: trustyai.opendatahub.io
@@ -21,14 +20,19 @@ spec:
         description: TrustyAIService is the Schema for the trustyaiservices API
         properties:
           apiVersion:
-            description: 'APIVersion defines the versioned schema of this representation
-              of an object. Servers should convert recognized schemas to the latest
-              internal value, and may reject unrecognized values. More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#resources'
+            description: |-
+              APIVersion defines the versioned schema of this representation of an object.
+              Servers should convert recognized schemas to the latest internal value, and
+              may reject unrecognized values.
+              More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#resources
             type: string
           kind:
-            description: 'Kind is a string value representing the REST resource this
-              object represents. Servers may infer this from the endpoint the client
-              submits requests to. Cannot be updated. In CamelCase. More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#types-kinds'
+            description: |-
+              Kind is a string value representing the REST resource this object represents.
+              Servers may infer this from the endpoint the client submits requests to.
+              Cannot be updated.
+              In CamelCase.
+              More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#types-kinds
             type: string
           metadata:
             type: object

config/rbac/role.yaml

@@ -2,13 +2,16 @@
 apiVersion: rbac.authorization.k8s.io/v1
 kind: ClusterRole
 metadata:
-  creationTimestamp: null
   name: manager-role
 rules:
 - apiGroups:
   - ""
   resources:
   - configmaps
+  - persistentvolumeclaims
+  - pods
+  - secrets
+  - services
   verbs:
   - create
   - delete
@@ -28,14 +31,10 @@ rules:
 - apiGroups:
   - ""
   resources:
-  - pods
+  - persistentvolumes
   verbs:
-  - create
-  - delete
   - get
   - list
-  - patch
-  - update
   - watch
 - apiGroups:
   - ""
@@ -47,18 +46,6 @@ rules:
   - get
   - list
   - watch
-- apiGroups:
-  - ""
-  resources:
-  - secrets
-  verbs:
-  - create
-  - delete
-  - get
-  - list
-  - patch
-  - update
-  - watch
 - apiGroups:
   - ""
   resources:
@@ -104,38 +91,6 @@ rules:
   - create
   - get
   - update
-- apiGroups:
-  - ""
-  resources:
-  - persistentvolumeclaims
-  verbs:
-  - create
-  - delete
-  - get
-  - list
-  - patch
-  - update
-  - watch
-- apiGroups:
-  - ""
-  resources:
-  - persistentvolumes
-  verbs:
-  - get
-  - list
-  - watch
-- apiGroups:
-  - ""
-  resources:
-  - services
-  verbs:
-  - create
-  - delete
-  - get
-  - list
-  - patch
-  - update
-  - watch
 - apiGroups:
   - monitoring.coreos.com
   resources:
@@ -212,31 +167,6 @@ rules:
   - trustyai.opendatahub.io
   resources:
   - lmevaljobs
-  verbs:
-  - create
-  - delete
-  - get
-  - list
-  - patch
-  - update
-  - watch
-- apiGroups:
-  - trustyai.opendatahub.io
-  resources:
-  - lmevaljobs/finalizers
-  verbs:
-  - update
-- apiGroups:
-  - trustyai.opendatahub.io
-  resources:
-  - lmevaljobs/status
-  verbs:
-  - get
-  - patch
-  - update
-- apiGroups:
-  - trustyai.opendatahub.io
-  resources:
   - trustyaiservices
   verbs:
   - create
@@ -249,12 +179,14 @@ rules:
 - apiGroups:
   - trustyai.opendatahub.io
   resources:
+  - lmevaljobs/finalizers
   - trustyaiservices/finalizers
   verbs:
   - update
 - apiGroups:
   - trustyai.opendatahub.io
   resources:
+  - lmevaljobs/status
   - trustyaiservices/status
   verbs:
   - get

controllers/lmes/lmevaljob_controller.go

@@ -181,6 +181,10 @@ func (r *LMEvalJobReconciler) Reconcile(ctx context.Context, req ctrl.Request) (
 		job.Status.State = lmesv1alpha1.NewJobState
 	}
 
+	if job.Spec.Suspend {
+		r.handleSuspend(ctx, log, job)
+	}
+
 	// Handle the job based on its state
 	switch job.Status.State {
 	case lmesv1alpha1.NewJobState:
@@ -198,6 +202,10 @@ func (r *LMEvalJobReconciler) Reconcile(ctx context.Context, req ctrl.Request) (
 		return r.handleComplete(ctx, log, job)
 	case lmesv1alpha1.CancelledJobState:
 		return r.handleCancel(ctx, log, job)
+	case lmesv1alpha1.SuspendedJobState:
+		if !job.Spec.Suspend {
+			return r.handleResume(ctx, log, job)
+		}
 	}
 
 	return ctrl.Result{}, nil
@@ -634,6 +642,41 @@ func (r *LMEvalJobReconciler) handleCancel(ctx context.Context, log logr.Logger,
 	return ctrl.Result{}, err
 }
 
+func (r *LMEvalJobReconciler) handleSuspend(ctx context.Context, log logr.Logger, job *lmesv1alpha1.LMEvalJob) (ctrl.Result, error) {
+	defer r.pullingJobs.remove(string(job.GetUID()))
+	if job.Status.State != lmesv1alpha1.NewJobState {
+		log.Info("Suspend job")
+		if err := r.deleteJobPod(ctx, job); err != nil && client.IgnoreNotFound(err) != nil {
+			log.Error(err, "failed to delete pod for suspended job")
+			return r.pullingJobs.addOrUpdate(string(job.GetUID()), r.options.PodCheckingInterval), nil
+		}
+	} else {
+		log.Info("Create job in suspend state.")
+	}
+	job.Status.State = lmesv1alpha1.SuspendedJobState
+	err := r.Status().Update(ctx, job)
+	if err != nil {
+		log.Error(err, "failed to update job status to suspended")
+	}
+
+	return ctrl.Result{}, err
+}
+
+func (r *LMEvalJobReconciler) handleResume(ctx context.Context, log logr.Logger, job *lmesv1alpha1.LMEvalJob) (ctrl.Result, error) {
+	log.Info("Resume job")
+	pod := r.createPod(job, log)
+	if err := r.Create(ctx, pod); err != nil {
+		log.Error(err, "failed to create pod to resume job")
+		return r.pullingJobs.addOrUpdate(string(job.GetUID()), r.options.PodCheckingInterval), nil
+	}
+	job.Status.State = lmesv1alpha1.ScheduledJobState
+	err := r.Status().Update(ctx, job)
+	if err != nil {
+		log.Error(err, "failed to update job status to scheduled")
+	}
+	return ctrl.Result{}, err
+}
+
 func (r *LMEvalJobReconciler) validateCustomCard(job *lmesv1alpha1.LMEvalJob, log logr.Logger) error {
 	if job.Spec.TaskList.TaskRecipes == nil {
 		return nil