版本：v1.13.0 (最新版)

NodeGroup 插件用户指南

简介

NodeGroup 插件用于通过为节点分配标签并在 Queue 上设置节点标签亲和性来实现资源隔离。

用法

为节点分配标签

为节点分配标签，标签键为 volcano.sh/nodegroup-name。

kubectl label nodes <nodename> volcano.sh/nodegroup-name=<groupname>

配置 Queue

创建 Queue 并将其绑定到 NodeGroup。

apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: default
  spec:
    reclaimable: true
    weight: 1
    affinity: # added field
      nodeGroupAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          - <groupname>
        preferredDuringSchedulingIgnoredDuringExecution:
          - <groupname>
      nodeGroupAntiAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          - <groupname>
        preferredDuringSchedulingIgnoredDuringExecution:
          - <groupname>

提交 Volcano Job

将名为 job-1 的 Volcano Job 提交到 default Queue。

$ cat <<EOF | kubectl apply -f -
apiVersion: batch.volcano.sh/v1alpha1
kind: Job
metadata:
  name: job-1
spec:
  minAvailable: 1
  schedulerName: volcano
  queue: default
  policies:
    - event: PodEvicted
      action: RestartJob
  tasks:
    - replicas: 1
      name: nginx
      policies:
      - event: TaskCompleted
        action: CompleteJob
      template:
        spec:
          containers:
            - command:
              - sleep
              - 10m
              image: nginx:latest
              name: nginx
              resources:
                requests:
                  cpu: 1
                limits:
                  cpu: 1
          restartPolicy: Never
EOF

验证 Queue 亲和性与反亲和规则是否生效

查询 Pod 信息，确认 Pod 是否已调度到正确节点。Pod 应调度到带有 nodeGroupAffinity.requiredDuringSchedulingIgnoredDuringExecution 或 nodeGroupAffinity.preferredDuringSchedulingIgnoredDuringExecution 标签的节点；否则，应调度到带有 nodeGroupAntiAffinity.preferredDuringSchedulingIgnoredDuringExecution 标签的节点。特别地，Pod 不得调度到带有 nodeGroupAntiAffinity.requiredDuringSchedulingIgnoredDuringExecution 标签的节点。

kubectl get po job-1-nginx-0 -o wide

NodeGroup 插件 strict 配置

strict 为布尔参数，用于控制未定义 NodeGroup 亲和性的 Queue 的调度行为。

当 strict: false 时，未配置亲和性的 Queue 中的任务可调度到未带有 volcano.sh/nodegroup-name 标签的节点。
当 strict: true（默认值）时，任务仅允许调度到带有 volcano.sh/nodegroup-name 标签的节点。

默认值为 true。

# scheduler configuration
actions: "allocate, backfill, preempt, reclaim"
tiers:
  - plugins:
      - name: priority
      - name: gang
      - name: conformance
  - plugins:
      - name: drf
      - name: predicates
      - name: proportion
      - name: nodegroup
        arguments:
          strict: false

层级 Queue 配置

启用层级支持

若要在 NodeGroup 插件中使用层级 Queue，请确保插件配置中包含 enableHierarchy: true：

# scheduler configuration
actions: "allocate, backfill, preempt, reclaim"
tiers:
  - plugins:
      - name: priority
      - name: gang
      - name: conformance
  - plugins:
      - name: drf
      - name: predicates
      - name: proportion
      - name: nodegroup
        enableHierarchy: true

创建 Queue 层级

基本层级设置

创建带 NodeGroup 亲和性的根 Queue：

apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: root
spec:
  weight: 1
  affinity:
    nodeGroupAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        - production

创建继承亲和性的子 Queue：

apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: engineering
spec:
  weight: 1
  parent: root
  # No affinity specified - inherits from root

---
apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: backend-team
spec:
  weight: 1
  parent: engineering
  # No affinity specified - inherits from root (through engineering)

创建带自定义亲和性的子 Queue：

apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: frontend-team
spec:
  weight: 1
  parent: engineering
  affinity:
    nodeGroupAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        - frontend
    # Overrides inherited affinity from root

继承行为

编号	场景	行为
1	子 Queue 未配置亲和性，父 Queue 已配置亲和性	继承父 Queue 的亲和性
2	子 Queue 与父 Queue 均未配置亲和性	继承最近祖先的亲和性
3	子 Queue 已配置亲和性	使用自身亲和性（不继承）
4	未指定父 Queue	视为根 Queue 的子 Queue

NodeGroup 插件工作原理

NodeGroup 设计文档对 NodeGroup 有最详细的说明。以下提示基于四节点集群及名为 job-1 的 Volcano Job，可帮助避免某些问题：

编号	节点	标签
1	node1	groupname1
2	node2	groupname2
3	node3	groupname3
4	node4	groupname4

apiVersion: batch.volcano.sh/v1alpha1
kind: Job
metadata:
  name: job-1
spec:
  minAvailable: 1
  schedulerName: volcano
  queue: default
  policies:
    - event: PodEvicted
      action: RestartJob
  tasks:
    - replicas: 1
      name: nginx
      policies:
        - event: TaskCompleted
          action: CompleteJob
      template:
        spec:
          containers:
            - command:
                - sleep
                - 10m
              image: nginx:latest
              name: nginx
              resources:
                requests:
                  cpu: 1
                limits:
                  cpu: 1
          restartPolicy: Never

软约束是硬约束的子集，包括亲和与反亲和。例如，Queue 配置如下：

apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: default
  spec:
    reclaimable: true
    weight: 1
    affinity: # added field
      nodeGroupAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          - groupname1
          - groupname2
        preferredDuringSchedulingIgnoredDuringExecution:
          - groupname1
      nodeGroupAntiAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          - groupname3
          - groupname4
        preferredDuringSchedulingIgnoredDuringExecution:
          - groupname3

这意味着 default Queue 中的任务将调度到 groupname1 与 groupname2，并优先在 groupname1 上运行。任务不得运行在 groupname3 与 groupname4 上。但若其他 NodeGroup 资源不足，任务可运行在 groupname3 上。

若软约束不是硬约束的子集，则 Queue 配置不正确，会导致任务运行在 groupname2 上：

apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: default
  spec:
    reclaimable: true
    weight: 1
    affinity: # added field
      nodeGroupAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          - groupname2
        preferredDuringSchedulingIgnoredDuringExecution:
          - groupname1
      nodeGroupAntiAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          - groupname4
        preferredDuringSchedulingIgnoredDuringExecution:
          - groupname3

若 nodeGroupAffinity 与 nodeGroupAntiAffinity 冲突，nodeGroupAntiAffinity 优先级更高。

apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: default
  spec:
    reclaimable: true
    weight: 1
    affinity: # added field
      nodeGroupAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          - groupname1
          - groupname2
        preferredDuringSchedulingIgnoredDuringExecution:
          - groupname1
      nodeGroupAntiAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          - groupname1
        preferredDuringSchedulingIgnoredDuringExecution:
          - groupname2

这意味着 default Queue 中的任务只能在 groupname2 上运行。

通常任务会优先运行在 groupname1 上，因为它是软约束。但评分函数由多个插件组成，因此任务有时也可能运行在 groupname2 上。

apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: default
  spec:
    reclaimable: true
    weight: 1
    affinity: # added field
      nodeGroupAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          - groupname1
          - groupname2
        preferredDuringSchedulingIgnoredDuringExecution:
          - groupname1
      nodeGroupAntiAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          - groupname3
          - groupname4
        preferredDuringSchedulingIgnoredDuringExecution:
          - groupname3

层级 Queue 故障排查

问题：节点可用但任务仍未调度

现象：

NodeGroup 中虽有可用节点，任务仍处于 Pending
出现与 Queue 亲和性要求相关的错误信息

可能原因：

子 Queue 继承了父 Queue 的严格亲和性
NodeGroup 标签与继承的亲和规则不匹配
Queue 存在循环依赖

解决方案：

检查继承链：

kubectl get queue <queue-name> -o yaml
# Verify parent hierarchy and affinity inheritance

覆盖过于严格的继承：

apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: flexible-team
spec:
  parent: restrictive-parent
  affinity:
    nodeGroupAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        - flexible-nodes

问题：亲和行为不符合预期

现象：

任务被调度到意外的 NodeGroup
继承未按预期生效

排查步骤：

启用详细日志：

# Check scheduler logs with higher verbosity
kubectl logs -n volcano-system volcano-scheduler-xxx --tail=100 | grep -i nodegroup

验证 Queue 配置：

# List all queues with their hierarchy
kubectl get queues -o custom-columns=NAME:.metadata.name,PARENT:.spec.parent,WEIGHT:.spec.weight

问题：Queue 循环依赖

现象：

Queue 无法正确构建祖先关系
出现与环检测相关的警告信息

解决方案：

检查 Queue 层级：

# Check for circular references
kubectl get queues -o custom-columns=NAME:.metadata.name,PARENT:.spec.parent

通过更新父引用修复循环依赖：

# Remove or correct circular parent relationships
apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: problematic-queue
spec:
  parent: "" # Remove circular reference
  weight: 1

简介​

用法​

为节点分配标签​

配置 Queue​

提交 Volcano Job​

验证 Queue 亲和性与反亲和规则是否生效​

NodeGroup 插件 strict 配置​

层级 Queue 配置​

启用层级支持​

创建 Queue 层级​

继承行为​

NodeGroup 插件工作原理​

层级 Queue 故障排查​

问题：节点可用但任务仍未调度​

问题：亲和行为不符合预期​

问题：Queue 循环依赖​

简介

用法

为节点分配标签

配置 Queue

提交 Volcano Job

验证 Queue 亲和性与反亲和规则是否生效

NodeGroup 插件 strict 配置

层级 Queue 配置

启用层级支持

创建 Queue 层级

继承行为

NodeGroup 插件工作原理

层级 Queue 故障排查

问题：节点可用但任务仍未调度

问题：亲和行为不符合预期

问题：Queue 循环依赖