aiops/backend/pkg/agentcli/tools/kubernetes_action.go

// kubernetes_action.go - Kubernetes 集群变更操作（高风险，需用户确认） // Logan.Li
package tools

import (
	"context"
	"fmt"
	"strings"

	k8sclient "ezops/pkg/kubernetesclient"
)

// KubernetesActionTool 执行 Kubernetes 变更动作 // Logan.Li
type KubernetesActionTool struct {
	client    *k8sclient.Client
	namespace string
}

// NewKubernetesActionTool 创建变更工具 // Logan.Li
func NewKubernetesActionTool(client *k8sclient.Client, namespace string) *KubernetesActionTool {
	return &KubernetesActionTool{client: client, namespace: namespace}
}

// Name 工具名 // Logan.Li
func (t *KubernetesActionTool) Name() string { return "kubernetes_action" }

// Description 给 LLM 的说明 // Logan.Li
func (t *KubernetesActionTool) Description() string {
	return "对 Kubernetes 资源执行变更操作（高风险，需用户确认）。" +
		"action 支持: restart_deployment, scale_deployment, update_image, restart_statefulset, delete_pod, suspend_cronjob。" +
		"变更前请先用 kubernetes_get 确认当前状态。"
}

// Parameters JSON Schema // Logan.Li
func (t *KubernetesActionTool) Parameters() map[string]interface{} {
	return map[string]interface{}{
		"type": "object",
		"properties": map[string]interface{}{
			"action": map[string]interface{}{
				"type":        "string",
				"description": "操作类型: restart_deployment|scale_deployment|update_image|restart_statefulset|delete_pod|suspend_cronjob",
			},
			"namespace": map[string]interface{}{
				"type":        "string",
				"description": "命名空间，默认会话绑定命名空间",
			},
			"name": map[string]interface{}{
				"type":        "string",
				"description": "资源名称（Deployment/StatefulSet/CronJob/Pod）",
			},
			"replicas": map[string]interface{}{
				"type":        "integer",
				"description": "scale_deployment 目标副本数",
			},
			"image": map[string]interface{}{
				"type":        "string",
				"description": "update_image 新镜像地址",
			},
			"container": map[string]interface{}{
				"type":        "string",
				"description": "update_image 目标容器名，可省略（默认第一个）",
			},
			"suspend": map[string]interface{}{
				"type":        "boolean",
				"description": "suspend_cronjob: true=暂停, false=恢复",
			},
		},
		"required": []string{"action", "name"},
	}
}

// RiskLevel 静态高风险 // Logan.Li
func (t *KubernetesActionTool) RiskLevel() RiskLevel { return RiskHigh }

// Execute 执行变更 // Logan.Li
func (t *KubernetesActionTool) Execute(ctx context.Context, args map[string]interface{}) (*Result, error) {
	action := strings.ToLower(strings.TrimSpace(StringParam(args, "action")))
	name := strings.TrimSpace(StringParam(args, "name"))
	ns := strings.TrimSpace(StringParam(args, "namespace"))
	if ns == "" {
		ns = t.namespace
	}
	if action == "" || name == "" {
		return &Result{Success: false, Error: "action 和 name 不能为空"}, nil
	}

	var err error
	var msg string
	switch action {
	case "restart_deployment":
		err = t.client.RestartDeployment(ns, name)
		msg = fmt.Sprintf("已触发 Deployment %s/%s 滚动重启", ns, name)
	case "scale_deployment":
		replicas := IntParam(args, "replicas")
		if replicas < 0 {
			return &Result{Success: false, Error: "replicas 不能为负数"}, nil
		}
		err = t.client.ScaleDeployment(ns, name, replicas)
		msg = fmt.Sprintf("已将 Deployment %s/%s 扩缩容至 %d 副本", ns, name, replicas)
	case "update_image":
		image := strings.TrimSpace(StringParam(args, "image"))
		if image == "" {
			return &Result{Success: false, Error: "update_image 需要 image 参数"}, nil
		}
		container := strings.TrimSpace(StringParam(args, "container"))
		err = t.client.UpdateDeploymentImage(ns, name, container, image)
		msg = fmt.Sprintf("已更新 Deployment %s/%s 镜像为 %s", ns, name, image)
	case "restart_statefulset":
		err = t.client.RestartStatefulSet(ns, name)
		msg = fmt.Sprintf("已触发 StatefulSet %s/%s 滚动重启", ns, name)
	case "delete_pod":
		err = t.client.DeletePod(ns, name)
		msg = fmt.Sprintf("已删除 Pod %s/%s（控制器将视情况重建）", ns, name)
	case "suspend_cronjob":
		suspend := true
		if v, ok := args["suspend"].(bool); ok {
			suspend = v
		}
		err = t.client.SuspendCronJob(ns, name, suspend)
		if suspend {
			msg = fmt.Sprintf("已暂停 CronJob %s/%s", ns, name)
		} else {
			msg = fmt.Sprintf("已恢复 CronJob %s/%s", ns, name)
		}
	default:
		return &Result{Success: false, Error: fmt.Sprintf("不支持的 action: %s", action)}, nil
	}
	if err != nil {
		return &Result{Success: false, Error: fmt.Sprintf("操作失败: %v", err)}, nil
	}
	return &Result{Success: true, Output: msg}, nil
}