什么是GKE

GKE（Google Kubernetes Engine）是 Google Cloud 提供的容器管理服务，它基于 Kubernetes 构建，可以帮助您轻松部署、管理和扩展容器化的应用程序。您可以把它想象成一个强大的“容器编排大师”，让您的应用在云端高效、稳定地运行。GKE 简化了 Kubernetes 的复杂性，让您专注于应用开发，而无需花费大量精力在底层基础设施的管理上。它提供了自动伸缩、自动修复、滚动更新等功能，确保您的应用始终可用并保持最佳性能。简单来说，GKE 就是一个托管的 Kubernetes 服务。

一句话就是Google 在GCP上装好了k8s 让你使用

为什么选择GKE

既然是k8s, 那么不自己在gcp上的vm自己搭建？ GKE到底提供了什么inhouse build k8s 没有的功能？

1. 不再需要搭建k8s 平台软件（安装过的都知道有多烦）而无需手动配置和管理底层基础设施。包括后续的版本升级
2. 强大的可伸缩性和弹性：
   自动伸缩： GKE 可以根据应用程序的负载自动调整集群的大小，确保应用程序始终具有足够的资源。
   自动修复： GKE 会自动检测并修复集群中的故障，确保应用程序的持续可用性。
   区域集群： GKE 支持区域集群，可以将集群部署在多个可用区中，从而提高应用程序的可用性和容错能力。

3.与 Google Cloud Platform 的深度集成：
集成的身份验证和授权： GKE 与 Google Cloud IAM 集成，可以轻松管理集群的访问权限。
集成的日志记录和监控： GKE 与 Google Cloud Logging 和 Monitoring 集成，可以集中收集和分析集群的日志和指标。
集成的网络： GKE 与 Google Cloud VPC 集成，可以轻松创建安全的网络环境。

至于使用方法与inhouse build k8s一样， 用kubectl 可轻松管理。
而且gcp 提供一个基本的cluster /pod 管理ui

什么是GKE private cluster

GKE Private Cluster（私有集群）

GKE Private Cluster 是一种 Kubernetes 集群，它与公共互联网隔离。这意味着集群中的节点没有公共 IP 地址，并且只能通过内部网络访问。
总的来说，GKE Private Cluster 通过限制对公共互联网的访问来增强安全性，但同时也增加了配置和管理的复杂性

一些区别：

item	master	nodes vm
normal cluster	具有public ip endpoint	每个node 都有public ip
private cluster	两种都支持， 是否具有public endpoint 基于用户配置	没有public ip, 外部不能直接访问node vm

由于public cluster 每个node 都是创建在gce里的， 都分配public ip的话 需要一笔额外的cost， 本文 不考虑这个方案

至于 private cluster 也有两种：

一种是master 节点也没有public endpoint, 这样整个集群都在内网， 一般需要设置堡垒机才能访问master, 本文也不考虑这个方案

另一种是master 节点具有public endpiont, nodes是没有的， 本文关注的是这个方案

创建一个空的github terrform 项目

https://github.com/nvd11/terraform-gke-private-cluster2

之后准备好backend.tf 和provider.tf 两个关键配置
backend.tf

terraform {
  backend "gcs" {
    bucket  = "jason-hsbc"
    prefix  = "terraform/my-cluster2/state"
  }
}

provider.tf

terraform {
  required_providers {
    google = {
      source  = "hashicorp/google"
      version =  "~> 7.0.0"
    }
  }
}
provider "google" {
  project = var.project_id
  region  = var.region_id
  zone    = var.zone_id
}

再准备variables.tf

variable "project_id" {
  description = "The ID of the project"
  default     = "jason-hsbc" 
  type        = string
}

variable "region_id" {
  description = "The region of the project"
  default     = "europe-west2" 
  type        = string
}

variable "zone_id" {
  description = "The zone id of the project"
  default     = "europe-west2-c" 
  type        = string
}

//https://cloud.google.com/iam/docs/service-agents
variable "gcs_sa" {
  description = "built-in service acount of GCS"
  default     = "service-912156613264@gs-project-accounts.iam.gserviceaccount.com" 
  type        = string
}

//https://cloud.google.com/iam/docs/service-agents
variable "sts_sa" {
  description = "built-in service acount of Storage Transer service"
  default     = "project-912156613264@storage-transfer-service.iam.gserviceaccount.com" 
  type        = string
}

variable "vpc0" {
  description = "The name of the VPC network"
  default     = "tf-vpc0"
  type        = string
}

到这里就可以测试 terraform init 和 terraform plan了

创建1个vpc network 和vpc subnet

其中vpc network 我沿用之前创建的tf-vpc0, 这里不再重复创建
vpc network的tf 配置参考：
Google cloud 的VPC Network 虚拟局域网 介绍

但我这里会创建一个新的subnet 来for 这个cluster

vpc0-subnet3。tf

# create a subnet
# https://registry.terraform.io/providers/hashicorp/google/latest/docs/resources/compute_subnetwork
resource "google_compute_subnetwork" "tf-vpc0-subnet3" {
  project                  = var.project_id
  name                     = "tf-vpc0-subnet3"
  ip_cidr_range            = "192.168.5.0/24" # 192.168.4.1 ~ 192.168.04.255
  region                   = var.region_id
  # only PRIVATE could allow vm creation,  the PRIVATE item is displayed as "None" in GCP console subnet creation page
  # but we cannot set purpose to "None",  if we did , the subnet will still created as purpose = PRIVATE , and next terraform plan/apply will try to recreate the subnet!
  # as it detect changes for "PRIVATE" -> "NONE"
  # gcloud compute networks subnets describe tf-vpc0-subnet3 --region=europe-west2
  purpose                  = "PRIVATE" 
  role                     = "ACTIVE"
  private_ip_google_access = "true" # to eanble the vm to access gcp products via internal network but not internet, faster and less cost!
  network                  = "tf-vpc0"

  # Although the secondary\_ip\_range is not within the subnet's IP address range, 
  # they still belong to the same VPC network. GKE uses routing and firewall rules to ensure communication between Pods, Services, and VMs."

  secondary_ip_range {
    range_name    = "pods-range"      # 用于 Pods
    ip_cidr_range = "192.171.16.0/20"     # 选择一个不冲突的范围
  }

  secondary_ip_range {
    range_name    = "services-range"  # 用于 Services
    ip_cidr_range = "192.172.16.0/20"     # 选择一个不冲突的范围
  }
}

值得注意的是pod-range 和 seervices-range 的ip 都不能在这个subnet 本身的ip范围， 而且不能与任何vpc-network内定义过的subnet（or 其secondary ip range） 冲突。

也就是讲， 通常只有gfe 的node 节点会under 这个subnet本身定义的ip range内（192.168. 5.2～192.168.5.255）

创建cluster

cluster.tf

resource "google_container_cluster" "my-cluster2" {
  project  = var.project_id
  name     = "my-cluster2"
  location = var.region_id

  # use custom node pool but not default node pool
  remove_default_node_pool = true

   # initial_node_count - (Optional) The number of nodes to create in this cluster's default node pool. 
   # In regional or multi-zonal clusters, this is the number of nodes per zone. Must be set if node_pool is not set.
   #   If you're using google_container_node_pool objects with no default node pool, 
   # you'll need to set this to a value of at least 1, alongside setting remove_default_node_pool to true.
  initial_node_count       = 1

  deletion_protection = false
  # Gke master will has his own managed vpc
  #but gke will create nodes and svcs under below vpc and subnet
  # they will use vpc peering to connect each other
  network = var.vpc0
  subnetwork =  google_compute_subnetwork.tf-vpc0-subnet3.name

  # the desired configuration options for master authorized networks. 
  #Omit the nested cidr_blocks attribute to disallow external access (except the cluster node IPs, which GKE automatically whitelists)
  # we could just remove the whole block to allow all access 
  #master_authorized_networks_config {
 

  #}

  ip_allocation_policy {
    # tell where pods could get the ip
    cluster_secondary_range_name = "pods-range" # need pre-defined in tf-vpc0-SUbnet0

    #tell where svcs could get the ip
    services_secondary_range_name = "services-range" 
  }

    private_cluster_config {
        enable_private_nodes    = true #  nodes do not have public ip
        enable_private_endpoint = false # master have public ip， we need to set it to false ， or we need to configure master_authorized_networks_config to allow our ip
        
        master_global_access_config {
          enabled = true
        }

    }



    fleet {
        #Can't configure a value for "fleet.0.membership": its value will be decided automatically based on the result of applying this configuration.
        #membership = "projects/${var.project_id}/locations/global/memberships/${var.cluster_name}"
        project = var.project_id
    }
}

几个points：

什么是gke 的fleet

Fleet 配置在 GKE 集群中用于将集群注册到 Google Cloud Fleet Management。 Fleet Management 允许您将多个集群（无论它们位于何处，例如 Google Cloud、其他云提供商或本地）组织成一个逻辑单元，以便进行统一管理和策略应用。

fleet 在gcp 项目创建时就会自动创建， 通常不需要特别关注

private_cluster_config

这里就是private cluster 的核心配置
enable_private_endpoint = true 则代表master 没有任何public endpoint ，需要额外通过master_authorized_networks_config 来配置一些白名单（例如堡垒机）从集群外部访问Master。
这里配置enable_private_endpoint= false
python master_global_access_config { enabled = true }
这个配置则允许master 能从外网访问

创建node pool

由于我在 cluster 配置了不是用default node pool
则需要配置一个custom的

node-pool.tf

resource "google_container_node_pool" "my-cluster2-node-pool1" {
    count =1
    name ="node-pool1"
    #│ Because google_container_cluster.my-cluster1 has "count" set, its attributes must be accessed on
  # specific instances.
    cluster = google_container_cluster.my-cluster2.name
    location = google_container_cluster.my-cluster2.location

    #The number of nodes per instance group. This field can be used to update the number of nodes per instance group but should not be used alongsid
    node_count =1
 
    node_config {
      machine_type = "n2d-highmem-4"
      image_type = "COS_CONTAINERD"
      #grants the nodes in "my-node-pool1" full access to all Google Cloud Platform services.
      oauth_scopes = ["https://www.googleapis.com/auth/cloud-platform"] 
      service_account  = "vm-common@jason-hsbc.iam.gserviceaccount.com"
     
    }
}

注意的是node_count ， gke 会在当前选择的region( europe-west2) 中的每个zone（europe-west2-a,europe-west2-b,europe-west2-c） 中分别创建1个instance group

这个node_count 是配置每个mig内 有多少个node（gce-vm）
如果配置1, 就代表有3个nodes, 够用了

terraform apply

没什么好说的， 等个5分钟左右就创建好了

检查node pool 配置

首先是mig

gateman@MoreFine-S500: terraform-gke-private-cluster2$ gcloud compute instance-groups managed list
NAME                                        LOCATION        SCOPE  BASE_INSTANCE_NAME                      SIZE  TARGET_SIZE  INSTANCE_TEMPLATE                       AUTOSCALED
gke-my-cluster1-my-node-pool1-5cad8c5c-grp  europe-west2-a  zone   gke-my-cluster1-my-node-pool1-5cad8c5c  2     2            gke-my-cluster1-my-node-pool1-11210656  no
gke-my-cluster2-node-pool1-01eff82c-grp     europe-west2-a  zone   gke-my-cluster2-node-pool1-01eff82c     1     1            gke-my-cluster2-node-pool1-01eff82c     no
gke-my-cluster1-my-node-pool1-f7d2eb2b-grp  europe-west2-b  zone   gke-my-cluster1-my-node-pool1-f7d2eb2b  2     2            gke-my-cluster1-my-node-pool1-13492c03  no
gke-my-cluster2-node-pool1-6a83612b-grp     europe-west2-b  zone   gke-my-cluster2-node-pool1-6a83612b     1     1            gke-my-cluster2-node-pool1-6a83612b     no
gke-my-cluster1-my-node-pool1-8902d932-grp  europe-west2-c  zone   gke-my-cluster1-my-node-pool1-8902d932  2     2            gke-my-cluster1-my-node-pool1-ccef768c  no
gke-my-cluster2-node-pool1-8bb426f2-grp     europe-west2-c  zone   gke-my-cluster2-node-pool1-8bb426f2     1     1            gke-my-cluster2-node-pool1-8bb426f2     no

可见创建出来了3个 mig 分别在不同的zone

然后再看gce vm

gateman@MoreFine-S500: terraform-gke-private-cluster2$ gcloud compute instances list| grep -i my-cluster2
gke-my-cluster2-node-pool1-01eff82c-1r5b     europe-west2-a  n2d-highmem-4                192.168.5.18                            RUNNING
gke-my-cluster2-node-pool1-6a83612b-mj40     europe-west2-b  n2d-highmem-4                192.168.5.16                            RUNNING
gke-my-cluster2-node-pool1-8bb426f2-3nqn     europe-west2-c  n2d-highmem-4                192.168.5.17                            RUNNING

注意3个node 都没有分配外网ip 而且内网ip都在tf-vpc0-subnet3 内

使用kubectl 连接集群

虽然nodes是内网的， 但是我们已经配置了让master 可以从外网访问， 所以可以很方便地从cloudshell连接

用下面命令enable kubectl 配置

gcloud container clusters get-credentials my-cluster2 --region europe-west2 --project jason-hsbc

接下来就可以使用kubectl管理集群了

如何查看master 节点信息， master 到底部署在哪

其实上从gce vm 的list 里是找不到master node的。

我们先查看 cluster的相信信息

gateman@MoreFine-S500: envoy-config$ gcloud container clusters describe my-cluster2 --region=europe-west2
addonsConfig:
  gcePersistentDiskCsiDriverConfig:
    enabled: true
  kubernetesDashboard:
    disabled: true
  networkPolicyConfig:
    disabled: true
anonymousAuthenticationConfig:
  mode: ENABLED
autopilot: {}
autoscaling:
  autoscalingProfile: BALANCED
binaryAuthorization: {}
clusterIpv4Cidr: 192.171.16.0/20
controlPlaneEndpointsConfig:
  dnsEndpointConfig:
    allowExternalTraffic: false
    endpoint: gke-059344205081454eb228f72a1d7a92706645-912156613264.europe-west2.gke.goog
  ipEndpointsConfig:
    authorizedNetworksConfig:
      privateEndpointEnforcementEnabled: true
    enablePublicEndpoint: true
    enabled: true
    globalAccess: true
    privateEndpoint: 192.168.5.9
    publicEndpoint: 34.147.241.202
createTime: '2025-10-02T15:57:21+00:00'
currentMasterVersion: 1.33.4-gke.1172000
currentNodeCount: 3
currentNodeVersion: 1.33.4-gke.1172000

...

信息很长， 我们看重点
ipEndpointsConfig:
authorizedNetworksConfig:
privateEndpointEnforcementEnabled: true
enablePublicEndpoint: true
enabled: true
globalAccess: true
privateEndpoint: 192.168.5.9
publicEndpoint: 34.147.241.202

这里就是master 的外网ip和内网ip

至于master 装在哪？ 其实gke 是把master纳入control plane管理， 并没有显示部署master节点