feature: Add podgroups statistics

Signed-off-by: jessestutler <jesseincomparable@hotmail.com>

cmd/controller-manager/app/options/options.go

@@ -37,6 +37,7 @@ const (
 	defaultMaxRequeueNum       = 15
 	defaultSchedulerName       = "volcano"
 	defaultHealthzAddress      = ":11251"
+	defaultListenAddress       = ":8080"
 	defaultLockObjectNamespace = "volcano-system"
 	defaultPodGroupWorkers     = 5
 	defaultGCWorkers           = 1
@@ -70,6 +71,8 @@ type ServerOption struct {
 	// defaulting to 0.0.0.0:11251
 	HealthzBindAddress string
 	EnableHealthz      bool
+	EnableMetrics      bool
+	ListenAddress      string
 	// To determine whether inherit owner's annotations for pods when create podgroup
 	InheritOwnerAnnotations bool
 	// WorkerThreadsForPG is the number of threads syncing podgroup operations
@@ -114,6 +117,8 @@ func (s *ServerOption) AddFlags(fs *pflag.FlagSet, knownControllers []string) {
 	fs.IntVar(&s.MaxRequeueNum, "max-requeue-num", defaultMaxRequeueNum, "The number of times a job, queue or command will be requeued before it is dropped out of the queue")
 	fs.StringVar(&s.HealthzBindAddress, "healthz-address", defaultHealthzAddress, "The address to listen on for the health check server.")
 	fs.BoolVar(&s.EnableHealthz, "enable-healthz", false, "Enable the health check; it is false by default")
+	fs.BoolVar(&s.EnableMetrics, "enable-metrics", false, "Enable the metrics function; it is false by default")
+	fs.StringVar(&s.ListenAddress, "listen-address", defaultListenAddress, "The address to listen on for HTTP requests.")
 	fs.BoolVar(&s.InheritOwnerAnnotations, "inherit-owner-annotations", true, "Enable inherit owner annotations for pods when create podgroup; it is enabled by default")
 	fs.Uint32Var(&s.WorkerThreadsForPG, "worker-threads-for-podgroup", defaultPodGroupWorkers, "The number of threads syncing podgroup operations. The larger the number, the faster the podgroup processing, but requires more CPU load.")
 	fs.Uint32Var(&s.WorkerThreadsForGC, "worker-threads-for-gc", defaultGCWorkers, "The number of threads for recycling jobs. The larger the number, the faster the job recycling, but requires more CPU load.")

cmd/controller-manager/app/server.go

@@ -19,8 +19,12 @@ package app
 import (
 	"context"
 	"fmt"
+	"net/http"
 	"os"
 
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/collectors"
+	"github.com/prometheus/client_golang/prometheus/promhttp"
 	v1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/util/uuid"
 	"k8s.io/client-go/informers"
@@ -31,6 +35,7 @@ import (
 	"k8s.io/client-go/tools/leaderelection"
 	"k8s.io/client-go/tools/leaderelection/resourcelock"
 	"k8s.io/client-go/tools/record"
+	"k8s.io/component-base/metrics/legacyregistry"
 	"k8s.io/klog/v2"
 
 	"volcano.sh/apis/pkg/apis/helpers"
@@ -48,13 +53,19 @@ func Run(opt *options.ServerOption) error {
 	if err != nil {
 		return err
 	}
-
 	if opt.EnableHealthz {
 		if err := helpers.StartHealthz(opt.HealthzBindAddress, "volcano-controller", opt.CaCertData, opt.CertData, opt.KeyData); err != nil {
 			return err
 		}
 	}
 
+	if opt.EnableMetrics {
+		go func() {
+			http.Handle("/metrics", promHandler())
+			klog.Fatalf("Prometheus Http Server failed %s", http.ListenAndServe(opt.ListenAddress, nil))
+		}()
+	}
+
 	run := startControllers(config, opt)
 
 	ctx := signals.SetupSignalContext()
@@ -177,3 +188,10 @@ func isControllerEnabled(name string, controllers []string) bool {
 	// if we get here, there was no explicit inclusion or exclusion
 	return hasStar
 }
+
+func promHandler() http.Handler {
+	// Unregister go and process related collector because it's duplicated and `legacyregistry.DefaultGatherer` also has registered them.
+	prometheus.DefaultRegisterer.Unregister(collectors.NewProcessCollector(collectors.ProcessCollectorOpts{}))
+	prometheus.DefaultRegisterer.Unregister(collectors.NewGoCollector())
+	return promhttp.InstrumentMetricHandler(prometheus.DefaultRegisterer, promhttp.HandlerFor(prometheus.Gatherers{prometheus.DefaultGatherer, legacyregistry.DefaultGatherer}, promhttp.HandlerOpts{}))
+}

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

@@ -58,7 +58,7 @@ rules:
     verbs: ["get", "create", "delete", "update"]
   - apiGroups: ["scheduling.incubator.k8s.io", "scheduling.volcano.sh"]
     resources: ["podgroups", "queues", "queues/status"]
-    verbs: ["get", "list", "watch", "create", "delete", "update"]
+    verbs: ["get", "list", "watch", "create", "delete", "update", "patch"]
   - apiGroups: ["flow.volcano.sh"]
     resources: ["jobflows", "jobtemplates"]
     verbs: ["get", "list", "watch", "create", "delete", "update"]
@@ -155,6 +155,7 @@ spec:
             args:
               - --logtostderr
               - --enable-healthz=true
+              - --enable-metrics=true
               - --leader-elect={{ .Values.custom.leader_elect_enable }}
               {{- if .Values.custom.leader_elect_enable }}
               - --leader-elect-resource-namespace={{ .Release.Namespace }}
@@ -167,3 +168,28 @@ spec:
               {{- toYaml .Values.custom.controller_default_csc | nindent 14 }}
             {{- end }}
 {{- end }}
+---
+apiVersion: v1
+kind: Service
+metadata:
+  annotations:
+    prometheus.io/path: /metrics
+    prometheus.io/port: "8080"
+    prometheus.io/scrape: "true"
+  name: {{ .Release.Name }}-controllers-service
+  namespace: {{ .Release.Namespace }}
+  labels:
+    app: volcano-controller
+    {{- if .Values.custom.common_labels }}
+    {{- toYaml .Values.custom.common_labels | nindent 4 }}
+    {{- end }}
+spec:
+  ports:
+    - port: 8080
+      protocol: TCP
+      targetPort: 8080
+      name: "metrics"
+  selector:
+    app: volcano-controller
+  type: ClusterIP
+{{- end }}

pkg/cli/queue/get.go

@@ -63,21 +63,38 @@ func GetQueue(ctx context.Context) error {
 		return err
 	}
 
-	PrintQueue(queue, os.Stdout)
+	// Currently CRD does not support filtering directly from kube-apiserver using fieldSelector,
+	// The featuregate called CustomResourceFieldSelectors is turned off by default in kube-apiserver
+	// and is still in alpha in v1.30. 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 +84,42 @@ func ListQueue(ctx context.Context) error {
 		fmt.Printf("No resources found\n")
 		return nil
 	}
-	PrintQueues(queues, os.Stdout)
+
+	// Currently CRD does not support filtering directly from kube-apiserver using fieldSelector,
+	// The featuregate called CustomResourceFieldSelectors is turned off by default in kube-apiserver
+	// and is still in alpha in v1.30. 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 +1,73 @@
+package metrics
+
+import (
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
+)
+
+var (
+	pgPendingPhaseNum = promauto.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Subsystem: VolcanoNamespace,
+			Name:      "podgroup_pending_phase_num",
+			Help:      "Number of podgroup at pending phase",
+		}, []string{"queue_name"},
+	)
+
+	pgRunningPhaseNum = promauto.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Subsystem: VolcanoNamespace,
+			Name:      "podgroup_running_phase_num",
+			Help:      "Number of podgroup at running phase",
+		}, []string{"queue_name"},
+	)
+
+	pgUnknownPhaseNum = promauto.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Subsystem: VolcanoNamespace,
+			Name:      "podgroup_unknown_phase_num",
+			Help:      "Number of podgroup at unknown phase",
+		}, []string{"queue_name"},
+	)
+
+	pgInqueuePhaseNum = promauto.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Subsystem: VolcanoNamespace,
+			Name:      "podgroup_inqueue_phase_num",
+			Help:      "Number of podgroup at inqueue phase",
+		}, []string{"queue_name"},
+	)
+
+	pgCompletedPhaseNum = promauto.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Subsystem: 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/metrics/util.go

@@ -0,0 +1,6 @@
+package metrics
+
+const (
+	// VolcanoNamespace - namespace in prometheus used by volcano
+	VolcanoNamespace = "volcano"
+)

pkg/controllers/queue/queue_controller_action.go

@@ -28,6 +28,8 @@ import (
 
 	"volcano.sh/apis/pkg/apis/bus/v1alpha1"
 	schedulingv1beta1 "volcano.sh/apis/pkg/apis/scheduling/v1beta1"
+	v1beta1apply "volcano.sh/apis/pkg/client/applyconfiguration/scheduling/v1beta1"
+	"volcano.sh/volcano/pkg/controllers/metrics"
 	"volcano.sh/volcano/pkg/controllers/queue/state"
 )
 
@@ -57,9 +59,18 @@ func (c *queuecontroller) syncQueue(queue *schedulingv1beta1.Queue, updateStateF
 			queueStatus.Unknown++
 		case schedulingv1beta1.PodGroupInqueue:
 			queueStatus.Inqueue++
+		case schedulingv1beta1.PodGroupCompleted:
+			queueStatus.Completed++
 		}
 	}
 
+	// Update the metrics
+	metrics.UpdatePgPendingPhaseNum(queue.Name, float64(queueStatus.Pending))
+	metrics.UpdatePgRunningPhaseNum(queue.Name, float64(queueStatus.Running))
+	metrics.UpdatePgUnknownPhaseNum(queue.Name, float64(queueStatus.Unknown))
+	metrics.UpdatePgInqueuePhaseNum(queue.Name, float64(queueStatus.Inqueue))
+	metrics.UpdatePgCompletedPhaseNum(queue.Name, float64(queueStatus.Completed))
+
 	if updateStateFn != nil {
 		updateStateFn(&queueStatus, podGroups)
 	} else {
@@ -79,10 +90,10 @@ func (c *queuecontroller) syncQueue(queue *schedulingv1beta1.Queue, updateStateF
 		return nil
 	}
 
-	newQueue := queue.DeepCopy()
-	newQueue.Status = queueStatus
-	if _, err := c.vcClient.SchedulingV1beta1().Queues().UpdateStatus(context.TODO(), newQueue, metav1.UpdateOptions{}); err != nil {
-		klog.Errorf("Failed to update status of Queue %s: %v.", newQueue.Name, err)
+	queueStatusApply := v1beta1apply.QueueStatus().WithState(queueStatus.State).WithAllocated(queueStatus.Allocated)
+	queueApply := v1beta1apply.Queue(queue.Name).WithStatus(queueStatusApply)
+	if _, err := c.vcClient.SchedulingV1beta1().Queues().ApplyStatus(context.TODO(), queueApply, metav1.ApplyOptions{FieldManager: controllerName}); err != nil {
+		klog.Errorf("Failed to apply status of Queue %s: %v.", queue.Name, err)
 		return err
 	}