如何删除 TKE 节点¶

from tencentcloud.common import credential
from tencentcloud.tke.v20180525 import tke_client, models

# 初始化认证
cred = credential.Credential("SecretId", "SecretKey")
client = tke_client.TkeClient(cred, "ap-guangzhou")

# 方式1: 删除节点（默认销毁 CVM）
req = models.DeleteClusterInstancesRequest()
req.ClusterId = "cls-xxxxxxxx"
req.InstanceIds = ["ins-xxxxxxxx"]
req.InstanceDeleteMode = "terminate"  # 可选，默认值

resp = client.DeleteClusterInstances(req)
print(f"删除请求已提交，RequestId: {resp.RequestId}")

# 方式2: 删除节点但保留 CVM
req = models.DeleteClusterInstancesRequest()
req.ClusterId = "cls-xxxxxxxx"
req.InstanceIds = ["ins-xxxxxxxx"]
req.InstanceDeleteMode = "retain"

resp = client.DeleteClusterInstances(req)
print("节点已从集群移除，CVM 实例已保留")

# 方式3: 批量删除
req = models.DeleteClusterInstancesRequest()
req.ClusterId = "cls-xxxxxxxx"
req.InstanceIds = ["ins-xxx", "ins-yyy", "ins-zzz"]

resp = client.DeleteClusterInstances(req)
print(f"批量删除 {len(req.InstanceIds)} 个节点")

def batch_delete_nodes(client, cluster_id, instance_ids, batch_size=10):
    """
    批量删除节点（分批）

    Args:
        client: TKE 客户端
        cluster_id: 集群 ID
        instance_ids: 节点 ID 列表
        batch_size: 每批删除数量
    """
    for i in range(0, len(instance_ids), batch_size):
        batch = instance_ids[i:i+batch_size]

        req = models.DeleteClusterInstancesRequest()
        req.ClusterId = cluster_id
        req.InstanceIds = batch

        try:
            resp = client.DeleteClusterInstances(req)
            print(f"批次 {i//batch_size + 1}: 已提交删除 {len(batch)} 个节点")
            print(f"  RequestId: {resp.RequestId}")
        except Exception as e:
            print(f"批次 {i//batch_size + 1} 删除失败: {e}")

        # 等待一段时间再提交下一批
        time.sleep(5)

# 使用示例
nodes_to_delete = ["ins-xxx", "ins-yyy", "ins-zzz", "ins-aaa", "ins-bbb"]
batch_delete_nodes(client, "cls-xxxxxxxx", nodes_to_delete)

def safe_delete_node(client, cluster_id, instance_id):
    """
    安全删除节点（先驱逐 Pod）

    Args:
        client: TKE 客户端
        cluster_id: 集群 ID
        instance_id: 节点 ID
    """
    import subprocess

    # 1. 获取节点名称
    # 需要先查询节点列表获取节点名称

    # 2. 设置节点不可调度
    subprocess.run(["kubectl", "cordon", node_name], check=True)
    print(f"节点 {node_name} 已设置为不可调度")

    # 3. 驱逐 Pod
    subprocess.run([
        "kubectl", "drain", node_name,
        "--ignore-daemonsets",
        "--delete-emptydir-data",
        "--force",
        "--timeout=300s"
    ], check=True)
    print(f"节点 {node_name} 上的 Pod 已驱逐")

    # 4. 删除节点
    req = models.DeleteClusterInstancesRequest()
    req.ClusterId = cluster_id
    req.InstanceIds = [instance_id]

    resp = client.DeleteClusterInstances(req)
    print(f"节点删除请求已提交，RequestId: {resp.RequestId}")

def delete_nodes_by_label(client, cluster_id, label_selector):
    """
    根据标签删除节点

    Args:
        client: TKE 客户端
        cluster_id: 集群 ID
        label_selector: 标签选择器（如 "env=test"）
    """
    import subprocess
    import json

    # 1. 使用 kubectl 查询匹配标签的节点
    result = subprocess.run(
        ["kubectl", "get", "nodes", "-l", label_selector, "-o", "json"],
        capture_output=True,
        text=True,
        check=True
    )

    nodes_data = json.loads(result.stdout)

    # 2. 提取节点的 instance-id
    instance_ids = []
    for node in nodes_data.get("items", []):
        provider_id = node["spec"]["providerID"]  # 格式: qcloud:///ap-guangzhou/ins-xxx
        instance_id = provider_id.split("/")[-1]
        instance_ids.append(instance_id)

    print(f"找到 {len(instance_ids)} 个匹配标签 '{label_selector}' 的节点")

    # 3. 删除节点
    if instance_ids:
        req = models.DeleteClusterInstancesRequest()
        req.ClusterId = cluster_id
        req.InstanceIds = instance_ids

        resp = client.DeleteClusterInstances(req)
        print(f"删除请求已提交，RequestId: {resp.RequestId}")

# 使用示例
delete_nodes_by_label(client, "cls-xxxxxxxx", "env=test")