feature: Add podgroups statistics

Signed-off-by: jessestutler <[email protected]>

go.mod

@@ -152,6 +152,7 @@ replace (
 	k8s.io/component-helpers => k8s.io/component-helpers v0.31.1
 	k8s.io/controller-manager => k8s.io/controller-manager v0.31.1
 	k8s.io/cri-api => k8s.io/cri-api v0.31.1
+	k8s.io/cri-client => k8s.io/cri-client v0.31.1
 	k8s.io/csi-translation-lib => k8s.io/csi-translation-lib v0.31.1
 	k8s.io/dynamic-resource-allocation => k8s.io/dynamic-resource-allocation v0.31.1
 	k8s.io/endpointslice => k8s.io/endpointslice v0.31.1

installer/helm/chart/volcano/templates/controllers.yaml

@@ -58,8 +58,11 @@ rules:
     resources: ["secrets"]
     verbs: ["get", "create", "delete", "update"]
   - apiGroups: ["scheduling.incubator.k8s.io", "scheduling.volcano.sh"]
-    resources: ["podgroups", "queues", "queues/status"]
+    resources: ["podgroups", "queues"]
     verbs: ["get", "list", "watch", "create", "delete", "update"]
+  - apiGroups: [ "scheduling.incubator.k8s.io", "scheduling.volcano.sh" ]
+    resources: [ "queues/status" ]
+    verbs: ["get", "watch", "patch"]
   - apiGroups: ["flow.volcano.sh"]
     resources: ["jobflows", "jobtemplates"]
     verbs: ["get", "list", "watch", "create", "delete", "update"]

installer/volcano-development.yaml

@@ -4318,8 +4318,11 @@ rules:
     resources: ["secrets"]
     verbs: ["get", "create", "delete", "update"]
   - apiGroups: ["scheduling.incubator.k8s.io", "scheduling.volcano.sh"]
-    resources: ["podgroups", "queues", "queues/status"]
+    resources: ["podgroups", "queues"]
     verbs: ["get", "list", "watch", "create", "delete", "update"]
+  - apiGroups: [ "scheduling.incubator.k8s.io", "scheduling.volcano.sh" ]
+    resources: [ "queues/status" ]
+    verbs: ["get", "watch", "patch"]
   - apiGroups: ["flow.volcano.sh"]
     resources: ["jobflows", "jobtemplates"]
     verbs: ["get", "list", "watch", "create", "delete", "update"]

pkg/cli/queue/get.go

@@ -63,21 +63,37 @@ func GetQueue(ctx context.Context) error {
 		return err
 	}
 
-	PrintQueue(queue, os.Stdout)
+	// Although the featuregate called CustomResourceFieldSelectors is enabled by default after v1.31, there are still
+	// users using k8s versions lower than v1.31. Therefore we can only get all the podgroups from kube-apiserver
+	// and then filtering them.
+	pgList, err := queueClient.SchedulingV1beta1().PodGroups("").List(ctx, metav1.ListOptions{})
+	if err != nil {
+		return fmt.Errorf("failed to list podgroup for queue %s with err: %v", getQueueFlags.Name, err)
+	}
+
+	pgStats := &PodGroupStatistics{}
+	for _, pg := range pgList.Items {
+		if pg.Spec.Queue == getQueueFlags.Name {
+			pgStats.StatPodGroupCountsForQueue(&pg)
+		}
+	}
+
+	PrintQueue(queue, pgStats, os.Stdout)
 
 	return nil
 }
 
 // PrintQueue prints queue information.
-func PrintQueue(queue *v1beta1.Queue, writer io.Writer) {
-	_, err := fmt.Fprintf(writer, "%-25s%-8s%-8s%-8s%-8s%-8s%-8s\n",
-		Name, Weight, State, Inqueue, Pending, Running, Unknown)
+func PrintQueue(queue *v1beta1.Queue, pgStats *PodGroupStatistics, writer io.Writer) {
+	_, err := fmt.Fprintf(writer, "%-25s%-8s%-8s%-8s%-8s%-8s%-8s%-8s\n",
+		Name, Weight, State, Inqueue, Pending, Running, Unknown, Completed)
 	if err != nil {
 		fmt.Printf("Failed to print queue command result: %s.\n", err)
 	}
-	_, err = fmt.Fprintf(writer, "%-25s%-8d%-8s%-8d%-8d%-8d%-8d\n",
-		queue.Name, queue.Spec.Weight, queue.Status.State, queue.Status.Inqueue,
-		queue.Status.Pending, queue.Status.Running, queue.Status.Unknown)
+
+	_, err = fmt.Fprintf(writer, "%-25s%-8d%-8s%-8d%-8d%-8d%-8d%-8d\n",
+		queue.Name, queue.Spec.Weight, queue.Status.State, pgStats.Inqueue,
+		pgStats.Pending, pgStats.Running, pgStats.Unknown, pgStats.Completed)
 	if err != nil {
 		fmt.Printf("Failed to print queue command result: %s.\n", err)
 	}

pkg/cli/queue/list.go

@@ -53,6 +53,9 @@ const (
 	// Inqueue status of queue
 	Inqueue string = "Inqueue"
 
+	// Completed status of the queue
+	Completed string = "Completed"
+
 	// State is state of queue
 	State string = "State"
 )
@@ -81,22 +84,41 @@ func ListQueue(ctx context.Context) error {
 		fmt.Printf("No resources found\n")
 		return nil
 	}
-	PrintQueues(queues, os.Stdout)
+
+	// Although the featuregate called CustomResourceFieldSelectors is enabled by default after v1.31, there are still
+	// users using k8s versions lower than v1.31. Therefore we can only get all the podgroups from kube-apiserver
+	// and then filtering them.
+	pgList, err := jobClient.SchedulingV1beta1().PodGroups("").List(ctx, metav1.ListOptions{})
+	if err != nil {
+		return fmt.Errorf("failed to list podgroups with err: %v", err)
+	}
+
+	queueStats := make(map[string]*PodGroupStatistics, len(queues.Items))
+	for _, queue := range queues.Items {
+		queueStats[queue.Name] = &PodGroupStatistics{}
+	}
+
+	for _, pg := range pgList.Items {
+		queueStats[pg.Spec.Queue].StatPodGroupCountsForQueue(&pg)
+	}
+
+	PrintQueues(queues, queueStats, os.Stdout)
 
 	return nil
 }
 
 // PrintQueues prints queue information.
-func PrintQueues(queues *v1beta1.QueueList, writer io.Writer) {
-	_, err := fmt.Fprintf(writer, "%-25s%-8s%-8s%-8s%-8s%-8s%-8s\n",
-		Name, Weight, State, Inqueue, Pending, Running, Unknown)
+func PrintQueues(queues *v1beta1.QueueList, queueStats map[string]*PodGroupStatistics, writer io.Writer) {
+	_, err := fmt.Fprintf(writer, "%-25s%-8s%-8s%-8s%-8s%-8s%-8s%-8s\n",
+		Name, Weight, State, Inqueue, Pending, Running, Unknown, Completed)
 	if err != nil {
 		fmt.Printf("Failed to print queue command result: %s.\n", err)
 	}
+
 	for _, queue := range queues.Items {
-		_, err = fmt.Fprintf(writer, "%-25s%-8d%-8s%-8d%-8d%-8d%-8d\n",
-			queue.Name, queue.Spec.Weight, queue.Status.State, queue.Status.Inqueue,
-			queue.Status.Pending, queue.Status.Running, queue.Status.Unknown)
+		_, err = fmt.Fprintf(writer, "%-25s%-8d%-8s%-8d%-8d%-8d%-8d%-8d\n",
+			queue.Name, queue.Spec.Weight, queue.Status.State, queueStats[queue.Name].Inqueue, queueStats[queue.Name].Pending,
+			queueStats[queue.Name].Running, queueStats[queue.Name].Unknown, queueStats[queue.Name].Completed)
 		if err != nil {
 			fmt.Printf("Failed to print queue command result: %s.\n", err)
 		}

pkg/cli/queue/util.go

@@ -29,6 +29,7 @@ import (
 
 	busv1alpha1 "volcano.sh/apis/pkg/apis/bus/v1alpha1"
 	"volcano.sh/apis/pkg/apis/helpers"
+	"volcano.sh/apis/pkg/apis/scheduling/v1beta1"
 	"volcano.sh/apis/pkg/client/clientset/versioned"
 )
 
@@ -62,3 +63,26 @@ func createQueueCommand(ctx context.Context, config *rest.Config, action busv1al
 
 	return nil
 }
+
+type PodGroupStatistics struct {
+	Inqueue   int
+	Pending   int
+	Running   int
+	Unknown   int
+	Completed int
+}
+
+func (pgStats *PodGroupStatistics) StatPodGroupCountsForQueue(pg *v1beta1.PodGroup) {
+	switch pg.Status.Phase {
+	case v1beta1.PodGroupInqueue:
+		pgStats.Inqueue++
+	case v1beta1.PodGroupPending:
+		pgStats.Pending++
+	case v1beta1.PodGroupRunning:
+		pgStats.Running++
+	case v1beta1.PodGroupUnknown:
+		pgStats.Unknown++
+	case v1beta1.PodGroupCompleted:
+		pgStats.Completed++
+	}
+}

pkg/controllers/metrics/podgroup.go

@@ -0,0 +1,75 @@
+package metrics
+
+import (
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
+
+	"volcano.sh/volcano/pkg/scheduler/metrics"
+)
+
+var (
+	pgPendingPhaseNum = promauto.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Subsystem: metrics.VolcanoNamespace,
+			Name:      "podgroup_pending_phase_num",
+			Help:      "Number of podgroup at pending phase",
+		}, []string{"queue_name"},
+	)
+
+	pgRunningPhaseNum = promauto.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Subsystem: metrics.VolcanoNamespace,
+			Name:      "podgroup_running_phase_num",
+			Help:      "Number of podgroup at running phase",
+		}, []string{"queue_name"},
+	)
+
+	pgUnknownPhaseNum = promauto.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Subsystem: metrics.VolcanoNamespace,
+			Name:      "podgroup_unknown_phase_num",
+			Help:      "Number of podgroup at unknown phase",
+		}, []string{"queue_name"},
+	)
+
+	pgInqueuePhaseNum = promauto.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Subsystem: metrics.VolcanoNamespace,
+			Name:      "podgroup_inqueue_phase_num",
+			Help:      "Number of podgroup at inqueue phase",
+		}, []string{"queue_name"},
+	)
+
+	pgCompletedPhaseNum = promauto.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Subsystem: metrics.VolcanoNamespace,
+			Name:      "podgroup_completed_phase_num",
+			Help:      "Number of podgroup at completed phase",
+		}, []string{"queue_name"},
+	)
+)
+
+// UpdatePgPendingPhaseNum recored the num of podgroups at pending state in queue
+func UpdatePgPendingPhaseNum(queueName string, num float64) {
+	pgPendingPhaseNum.WithLabelValues(queueName).Set(num)
+}
+
+// UpdatePgRunningPhaseNum recored the num of podgroups at running state in queue
+func UpdatePgRunningPhaseNum(queueName string, num float64) {
+	pgRunningPhaseNum.WithLabelValues(queueName).Set(num)
+}
+
+// UpdatePgUnknownPhaseNum recored the num of podgroups at unknown state in queue
+func UpdatePgUnknownPhaseNum(queueName string, num float64) {
+	pgUnknownPhaseNum.WithLabelValues(queueName).Set(num)
+}
+
+// UpdatePgInqueuePhaseNum recored the num of podgroups at inqueue state in queue
+func UpdatePgInqueuePhaseNum(queueName string, num float64) {
+	pgInqueuePhaseNum.WithLabelValues(queueName).Set(num)
+}
+
+// UpdatePgCompletedPhaseNum recored the num of podgroups at completed state in queue
+func UpdatePgCompletedPhaseNum(queueName string, num float64) {
+	pgCompletedPhaseNum.WithLabelValues(queueName).Set(num)
+}

pkg/controllers/queue/queue_controller.go

@@ -160,8 +160,6 @@ func (c *queuecontroller) Initialize(opt *framework.ControllerOption) error {
 	}
 
 	queuestate.SyncQueue = c.syncQueue
-	queuestate.OpenQueue = c.openQueue
-	queuestate.CloseQueue = c.closeQueue
 
 	c.syncHandler = c.handleQueue
 	c.syncCommandHandler = c.handleCommand