Take SubGroups into consideration when checking if job is ready for scheduling (#347)

Author: romanbaron

pkg/scheduler/api/cluster_info.go

@@ -24,6 +24,8 @@ import (
 
 	v1 "k8s.io/api/core/v1"
 
+	kueue "sigs.k8s.io/kueue/apis/kueue/v1alpha1"
+
 	"github.com/NVIDIA/KAI-scheduler/pkg/scheduler/api/bindrequest_info"
 	"github.com/NVIDIA/KAI-scheduler/pkg/scheduler/api/common_info"
 	"github.com/NVIDIA/KAI-scheduler/pkg/scheduler/api/configmap_info"
@@ -34,7 +36,6 @@ import (
 	"github.com/NVIDIA/KAI-scheduler/pkg/scheduler/api/storagecapacity_info"
 	"github.com/NVIDIA/KAI-scheduler/pkg/scheduler/api/storageclaim_info"
 	"github.com/NVIDIA/KAI-scheduler/pkg/scheduler/api/storageclass_info"
-	kueue "sigs.k8s.io/kueue/apis/kueue/v1alpha1"
 )
 
 // ClusterInfo is a snapshot of cluster by cache.
@@ -93,9 +94,14 @@ func (ci ClusterInfo) String() string {
 			str = str + fmt.Sprintf("\t Job(%s) name(%s) minAvailable(%v)\n",
 				job.UID, job.Name, job.MinAvailable)
 
+			for _, subGroup := range job.SubGroups {
+				str = str + fmt.Sprintf("\t\t subGroup(%s), minAvailable(%v)\n",
+					subGroup.Name, subGroup.MinAvailable)
+			}
+
 			i := 0
 			for _, task := range job.PodInfos {
-				str = str + fmt.Sprintf("\t\t %d: %v\n", i, task)
+				str = str + fmt.Sprintf("\t\t task %d: %v\n", i, task)
 				i++
 			}
 		}

pkg/scheduler/api/podgroup_info/job_info.go

@@ -351,7 +351,15 @@ func (pgi *PodGroupInfo) GetTasksActiveAllocatedReqResource() *resource_info.Res
 
 func (pgi *PodGroupInfo) IsReadyForScheduling() bool {
 	validTasks := pgi.GetNumAliveTasks() - pgi.GetNumGatedTasks()
-	return int32(validTasks) >= pgi.MinAvailable
+	if int32(validTasks) < pgi.MinAvailable {
+		return false
+	}
+	for _, subGroup := range pgi.SubGroups {
+		if !subGroup.IsReadyForScheduling() {
+			return false
+		}
+	}
+	return true
 }
 
 func (pgi *PodGroupInfo) IsElastic() bool {
@@ -432,9 +440,14 @@ func (pgi *PodGroupInfo) CloneWithTasks(tasks []*pod_info.PodInfo) *PodGroupInfo
 func (pgi *PodGroupInfo) String() string {
 	res := ""
 
+	for _, subGroup := range pgi.SubGroups {
+		res = res + fmt.Sprintf("\t\t subGroup %s: minAvailable(%v)\n",
+			subGroup.Name, subGroup.MinAvailable)
+	}
+
 	i := 0
 	for _, task := range pgi.PodInfos {
-		res = res + fmt.Sprintf("\n\t %d: %v", i, task)
+		res = res + fmt.Sprintf("\n\t task %d: %v", i, task)
 		i++
 	}