Volcano

# 有2个socket，每个socket有10个核，每个核开超线程，总共2*10*2=40个逻辑处理器。每个socket划分到一个numa node，总共两个numa。
# lscpu 查看 cpu 拓扑
Architecture:          x86_64
CPU op-mode(s):        32-bit, 64-bit
Byte Order:            Little Endian
CPU(s):                40
On-line CPU(s) list:   0-39
Thread(s) per core:    2
Core(s) per socket:    10
Socket(s):             2
NUMA node(s):          2
Vendor ID:             GenuineIntel
CPU family:            6
Model:                 79
Model name:            Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz
Stepping:              1
CPU MHz:               2200.134
CPU max MHz:           2200.0000
CPU min MHz:           1200.0000
BogoMIPS:              4389.32
Virtualization:        VT-x
L1d cache:             32K
L1i cache:             32K
L2 cache:              256K
L3 cache:              25600K
NUMA node0 CPU(s):     0-9,20-29
NUMA node1 CPU(s):     10-19,30-39
Flags:                 fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch epb cat_l3 cdp_l3 intel_ppin intel_pt ssbd ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm rdt_a rdseed adx smap xsaveopt cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm arat pln pts spec_ctrl intel_stibp flush_l1d

// 新增了一个 pod
func (sc *SchedulerCache) addPod(pod *v1.Pod) error {
	pi, err := sc.NewTaskInfo(pod)
	if err != nil {
		klog.Errorf("generate taskInfo for pod(%s) failed: %v", pod.Name, err)
		sc.resyncTask(pi)
	}

	return sc.addTask(pi)
}

// https://github.com/volcano-sh/volcano/blob/8944bfd7bc48cb3e1d6b3c047ef2fa52fec4c276/pkg/scheduler/plugins/binpack/binpack.go
func BinPackingScore(task *api.TaskInfo, node *api.NodeInfo, weight priorityWeight) float64 {
	score := 0.0
	weightSum := 0
	requested := task.Resreq
	allocatable := node.Allocatable
	used := node.Used

	for _, resource := range requested.ResourceNames() {
		request := requested.Get(resource)
		if request == 0 {
			continue
		}
		allocate := allocatable.Get(resource)
		nodeUsed := used.Get(resource)

		resourceWeight, found := weight.BinPackingResources[resource]
		if !found {
			continue
		}

		// 计算得分
		resourceScore, err := ResourceBinPackingScore(request, allocate, nodeUsed, resourceWeight)
		if err != nil {
			klog.V(4).Infof("task %s/%s cannot binpack node %s: resource: %s is %s, need %f, used %f, allocatable %f",
				task.Namespace, task.Name, node.Name, resource, err.Error(), request, nodeUsed, allocate)
			return 0
		}
		klog.V(5).Infof("task %s/%s on node %s resource %s, need %f, used %f, allocatable %f, weight %d, score %f",
			task.Namespace, task.Name, node.Name, resource, request, nodeUsed, allocate, resourceWeight, resourceScore)

		score += resourceScore
		weightSum += resourceWeight
	}

	// mapping the result from [0, weightSum] to [0, 10(MaxPriority)]
	if weightSum > 0 {
		// 总得分
		score /= float64(weightSum)
	}
	score *= float64(k8sFramework.MaxNodeScore * int64(weight.BinPackingWeight))

	return score
}

// 计算公式  Resource.weight * (request + used) / allocatable
func ResourceBinPackingScore(requested, capacity, used float64, weight int) (float64, error) {
	if capacity == 0 || weight == 0 {
		return 0, nil
	}

	usedFinally := requested + used
	if usedFinally > capacity {
		return 0, fmt.Errorf("not enough")
	}

	score := usedFinally * float64(weight) / capacity
	return score, nil
}

# CPU 权重值越高，得分越高，节点资源使用量越满，得分越高。Memory、GPU 等资源原理类似
CPU.weight * (request + used) / allocatable

(⎈|kubeasz-test:monitoring)➜  ~ kubectl get numatopologies node5 -o yaml
apiVersion: nodeinfo.volcano.sh/v1alpha1
kind: Numatopology
metadata:
  creationTimestamp: "2025-10-26T08:34:49Z"
  generation: 1
  name: node5
  resourceVersion: "26517897"
  uid: b0fd84f3-9d97-4df5-b053-9e209f6e6e04
spec:
  # cpu的NUMANodeID、SocketID、CoreID的信息
  cpuDetail:
    "0": {}
    "1":
      core: 1
    "2":
      socket: 1
    "3":
      core: 1
      socket: 1
    "4":
      socket: 2
    "5":
      core: 1
      socket: 2
    "6":
      socket: 3
    "7":
      core: 1
      socket: 3
    "8":
      socket: 4
    "9":
      core: 1
      socket: 4
    "10":
      socket: 5
    "11":
      core: 1
      socket: 5
    "12":
      socket: 6
    "13":
      core: 1
      socket: 6
    "14":
      socket: 7
    "15":
      core: 1
      socket: 7
  # 资源的NUMA感知信息，包含可分配量和资源总量
  numares:
    cpu:
      capacity: 16
  # 包含cpuManager、topologyManager的策略配置。
  policies:
    CPUManagerPolicy: ""
    TopologyManagerPolicy: ""

(⎈|kubeasz-test:volcano-system)➜  ~ kubectl api-resources| head -1;kubectl api-resources |grep volcano
NAME                                SHORTNAMES            APIVERSION                             NAMESPACED   KIND
jobs                                vcjob,vj              batch.volcano.sh/v1alpha1              true         Job
commands                                                  bus.volcano.sh/v1alpha1                true         Command
jobflows                            jf                    flow.volcano.sh/v1alpha1               true         JobFlow
jobtemplates                        jt                    flow.volcano.sh/v1alpha1               true         JobTemplate
numatopologies                      numatopo              nodeinfo.volcano.sh/v1alpha1           false        Numatopology
podgroups                           pg,podgroup-v1beta1   scheduling.volcano.sh/v1beta1          true         PodGroup
queues                              q,queue-v1beta1       scheduling.volcano.sh/v1beta1          false        Queue
hypernodes                          hn                    topology.volcano.sh/v1alpha1           false        HyperNode

# default queue
(⎈|kubeasz-test:volcano-system)➜  ~ kubectl get queue default -o yaml
apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  creationTimestamp: "2025-07-26T02:38:24Z"
  generation: 2
  name: default
  resourceVersion: "16927137"
  uid: b3816aae-d77e-4576-8699-eed4d0324848
spec:
  guarantee: {}
  parent: root
  reclaimable: true
  weight: 1
status:
  allocated:
    cpu: "3"
    memory: "0"
    pods: "3"
  reservation: {}
  state: Open

# root的queue，该queue为开启层级队列功能时使用，作为所有队列的根队列，default queue为root queue的子队列
(⎈|kubeasz-test:volcano-system)➜  ~ kubectl get queue root -o yaml
apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  creationTimestamp: "2025-07-26T02:38:24Z"
  generation: 786
  name: root
  resourceVersion: "17202849"
  uid: 8399648f-7082-49a1-98ec-7fb9c1138172
spec:
  deserved:
    attachable-volumes-csi-local.csi.openebs.io: 12884901882m
    cpu: 92400m
    ephemeral-storage: "381412259579"
    hugepages-2Mi: "0"
    kubernetes.io/batch-cpu: "46335"
    kubernetes.io/batch-memory: "38576847428"
    kubernetes.io/mid-cpu: "11076"
    kubernetes.io/mid-memory: "10040600338"
    memory: 96480436Ki
    pods: "660"
  guarantee:
    resource:
      cpu: "0"
      memory: "0"
  reclaimable: false
  weight: 1
status:
  allocated:
    cpu: "3"
    memory: "0"
    pods: "3"
  reservation: {}
  state: Open

// https://github.com/volcano-sh/volcano/blob/b27b4bbe7d19e225e75a11e424bee38ec29a4041/pkg/scheduler/actions/factory.go
func init() {
	// 注册 action
	framework.RegisterAction(reclaim.New())
	framework.RegisterAction(allocate.New())
	framework.RegisterAction(backfill.New())
	framework.RegisterAction(preempt.New())
	framework.RegisterAction(enqueue.New())
	framework.RegisterAction(shuffle.New())
}

type Action interface {
	// The unique name of Action.
	Name() string

	// Initialize initializes the allocator plugins.
	Initialize()

	// Execute allocates the cluster's resources into each queue.
	Execute(ssn *Session)

	// UnIntialize un-initializes the allocator plugins.
	UnInitialize()
}

func (pc *Scheduler) runOnce() {
	klog.V(4).Infof("Start scheduling ...")
	scheduleStartTime := time.Now()
	defer klog.V(4).Infof("End scheduling ...")

	pc.mutex.Lock()
	actions := pc.actions
	plugins := pc.plugins
	configurations := pc.configurations
	pc.mutex.Unlock()

	// Load ConfigMap to check which action is enabled.
	conf.EnabledActionMap = make(map[string]bool)
	for _, action := range actions {
		conf.EnabledActionMap[action.Name()] = true
	}

	ssn := framework.OpenSession(pc.cache, plugins, configurations)
	defer func() {
		framework.CloseSession(ssn)
		metrics.UpdateE2eDuration(metrics.Duration(scheduleStartTime))
	}()

	// 遍历 actions, 执行 action
	for _, action := range actions {
		actionStartTime := time.Now()
		action.Execute(ssn) // 传递了一个 ssn（*Session 类型）对象进去
		metrics.UpdateActionDuration(action.Name(), metrics.Duration(actionStartTime))
	}
}