cicd_yamls/cluster-setup/scripts/master.tf

# Clean up any existing k3s installation on the master node
resource "null_resource" "k3s_cleanup_master" {
  # SSH 登录你的 master server
  connection {
    type        = "ssh"
    host        = var.master_ip
    user        = "root"
    password    = var.master_password
  }

  # 远程执行清理命令
  provisioner "remote-exec" {
    inline = [
      "echo -e '\\033[32mCleanup start on master node\\033[0m'",
      "systemctl stop k3s 2>/dev/null || true",
      "pkill -f k3s 2>/dev/null || true",
      #"rm -rf /usr/local/bin/k3s",
      "rm -f /etc/systemd/system/k3s.service",
      "rm -rf /var/lib/rancher/k3s",
      "rm -rf /etc/rancher/k3s",
      "rm -rf /root/.kube",
      "echo -e '\\033[32mCleanup completed on master node\\033[0m'"
    ]
  }
}

# Install k3s server on the master node
resource "null_resource" "k3s_install_master" {
  depends_on = [null_resource.k3s_cleanup_master]
  
  # SSH 登录你的 master server
  connection {
    type        = "ssh"
    host        = var.master_ip
    user        = "root"
    password    = var.master_password
  }

  # 远程执行命令（你原来的所有逻辑，一模一样）
  provisioner "remote-exec" {
    inline = [
      "if [ ! -f /usr/local/bin/k3s ]; then",
        "wget -O /usr/local/bin/k3s http://download.9981.tech/k3s-v1.35.0%2Bk3s1",
        "chmod +x /usr/local/bin/k3s",
      "else",
        "echo -e '\\033[32m start installing master ... \\033[0m'",
        "chmod +x /usr/local/bin/k3s",
        "nohup /usr/local/bin/k3s server --disable=traefik --disable=servicelb --disable=metrics-server --token=my-secret-token --https-listen-port=6443 --pause-image=registry.cn-hangzhou.aliyuncs.com/google_containers/pause:3.6 --system-default-registry=registry.cn-hangzhou.aliyuncs.com > /root/k3s.log 2>&1 &",
        #"/usr/local/bin/k3s server &",
        "disown",
        "echo -e '\\033[32m Done server master！ \\033[0m'",
        #"systemctl start k3s",
        #"echo -e '\\033[32m Done installing master ... \\033[0m'",
      "fi"
    ]
  }
}

# Wait for k3s server to be ready on master node
resource "null_resource" "wait_for_k3s_ready" {
  depends_on = [null_resource.k3s_install_master]

  connection {
    type        = "ssh"
    host        = var.master_ip
    user        = "root"
    password    = var.master_password
  }

  provisioner "remote-exec" {
    inline = [
      "# 等待 k3s 服务完全启动并生成 token",
      "timeout 300 bash -c 'while [ ! -f /var/lib/rancher/k3s/server/node-token ]; do echo waiting for k3s token...; sleep 10; done' || exit 1",
      "  echo 'T----------------'",
    ]
  }
}