Karmada 多集群调度

什么是 Karmada

Karmada（Kubernetes Armada）是华为开源的多集群管理系统，提供跨集群的应用部署、调度和治理能力。

架构

Karmada 控制平面
├── Karmada API Server（兼容 K8s API）
├── Karmada Controller Manager
├── Karmada Scheduler（多集群调度）
└── etcd

成员集群
├── Cluster 1（北京）
├── Cluster 2（上海）
└── Cluster 3（广州）

安装

# 使用 karmadactl 安装
curl -s https://raw.githubusercontent.com/karmada-io/karmada/master/hack/install-cli.sh | sudo bash

karmadactl init \
  --kube-image-mirror-country=cn \
  --etcd-storage-mode=hostPath

# 注册成员集群
karmadactl join cluster1 \
  --kubeconfig=/root/.kube/karmada.config \
  --cluster-kubeconfig=/root/.kube/cluster1.config

# 查看集群
kubectl get clusters --kubeconfig=/root/.kube/karmada.config

传播策略

# 将 Deployment 部署到多个集群
apiVersion: policy.karmada.io/v1alpha1
kind: PropagationPolicy
metadata:
  name: my-app-policy
spec:
  resourceSelectors:
  - apiVersion: apps/v1
    kind: Deployment
    name: my-app
  - apiVersion: v1
    kind: Service
    name: my-app

  placement:
    # 集群亲和性
    clusterAffinity:
      clusterNames:
      - cluster1
      - cluster2
      - cluster3

    # 副本调度
    replicaScheduling:
      replicaSchedulingType: Divided
      replicaDivisionPreference: Weighted
      weightPreference:
        staticClusterWeight:
        - targetCluster:
            clusterNames: [cluster1]
          weight: 3
        - targetCluster:
            clusterNames: [cluster2]
          weight: 2
        - targetCluster:
            clusterNames: [cluster3]
          weight: 1

覆盖策略

# 不同集群使用不同配置
apiVersion: policy.karmada.io/v1alpha1
kind: OverridePolicy
metadata:
  name: my-app-override
spec:
  resourceSelectors:
  - apiVersion: apps/v1
    kind: Deployment
    name: my-app
  overrideRules:
  # 北京集群使用北京镜像仓库
  - targetCluster:
      clusterNames: [cluster1]
    overriders:
      imageOverrider:
      - component: Registry
        operator: replace
        value: registry.beijing.example.com
  # 上海集群使用上海镜像仓库
  - targetCluster:
      clusterNames: [cluster2]
    overriders:
      imageOverrider:
      - component: Registry
        operator: replace
        value: registry.shanghai.example.com

故障转移

# 集群故障时自动迁移
apiVersion: policy.karmada.io/v1alpha1
kind: PropagationPolicy
metadata:
  name: failover-policy
spec:
  resourceSelectors:
  - apiVersion: apps/v1
    kind: Deployment
    name: my-app
  placement:
    clusterAffinity:
      clusterNames: [cluster1, cluster2]
    replicaScheduling:
      replicaSchedulingType: Duplicated  # 每个集群都有完整副本
  failover:
    application:
      decisionConditions:
        tolerationSeconds: 60  # 集群不可用 60s 后触发迁移
      purgeMode: Graciously
      gracePeriodSeconds: 600