aiops/backend/controller/remote_exec.go

package controller

import (
	"encoding/json"
	"fmt"
	"net/http"
	"strconv"
	"strings"
	"time"

	"ezops/model"
	"ezops/pkg/remoteexec"

	"github.com/gin-gonic/gin"
	"gorm.io/gorm"
)

// RemoteExecHandler 远程执行处理器 — Go 原生 SSH 批量执行，替代 Ansible // Logan.Li
type RemoteExecHandler struct {
	db *gorm.DB
}

// NewRemoteExecHandler 创建远程执行处理器
func NewRemoteExecHandler(db *gorm.DB) *RemoteExecHandler {
	return &RemoteExecHandler{db: db}
}

// ==================== 远程命令执行 ====================

// BatchExecRequest 批量执行请求
type BatchExecRequest struct {
	ServerIDs []uint `json:"server_ids" binding:"required"` // 目标服务器 ID 列表
	Command   string `json:"command" binding:"required"`    // 要执行的命令
	Timeout   int    `json:"timeout"`                       // 超时（秒），默认 300
}

// BatchExec 批量远程执行命令 // Logan.Li
// POST /api/v1/remote-exec/execute
func (h *RemoteExecHandler) BatchExec(c *gin.Context) {
	var req BatchExecRequest
	if err := c.ShouldBindJSON(&req); err != nil {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "参数错误: " + err.Error()})
		return
	}

	// 查询目标服务器
	var servers []model.Server
	if err := h.db.Where("id IN ?", req.ServerIDs).Find(&servers).Error; err != nil {
		c.JSON(http.StatusInternalServerError, model.Response{Code: 500, Message: "查询服务器失败"})
		return
	}

	if len(servers) == 0 {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "未找到目标服务器"})
		return
	}

	// 构建 SSH 配置
	configs := make([]remoteexec.SSHConfig, 0, len(servers))
	for _, s := range servers {
		cfg, err := buildSSHConfig(h.db, s)
		if err != nil {
			c.JSON(http.StatusBadRequest, model.Response{
				Code: 400, Message: fmt.Sprintf("服务器 %s 认证配置错误: %v", s.Hostname, err),
			})
			return
		}
		configs = append(configs, cfg)
	}

	timeout := req.Timeout
	if timeout <= 0 {
		timeout = 300
	}

	// 并发执行
	results := remoteexec.BatchExec(configs, req.Command, time.Duration(timeout)*time.Second, 10)

	c.JSON(http.StatusOK, model.Response{Code: 200, Data: results})
}

// ==================== 文件批量上传 ====================

// BatchUploadRequest 批量上传请求
type BatchUploadRequest struct {
	ServerIDs  []uint `json:"server_ids" binding:"required"` // 目标服务器 ID 列表
	LocalPath  string `json:"local_path" binding:"required"` // AIOps 主机上的文件路径
	RemotePath string `json:"remote_path" binding:"required"` // 远程目标路径
}

// BatchUpload 批量上传文件到远程服务器 // Logan.Li
// POST /api/v1/remote-exec/upload
func (h *RemoteExecHandler) BatchUpload(c *gin.Context) {
	var req BatchUploadRequest
	if err := c.ShouldBindJSON(&req); err != nil {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "参数错误: " + err.Error()})
		return
	}

	if !strings.HasPrefix(req.LocalPath, "/") {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "local_path 必须是绝对路径"})
		return
	}

	// 查询目标服务器
	var servers []model.Server
	if err := h.db.Where("id IN ?", req.ServerIDs).Find(&servers).Error; err != nil {
		c.JSON(http.StatusInternalServerError, model.Response{Code: 500, Message: "查询服务器失败"})
		return
	}

	if len(servers) == 0 {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "未找到目标服务器"})
		return
	}

	// 构建 SSH 配置
	configs := make([]remoteexec.SSHConfig, 0, len(servers))
	for _, s := range servers {
		cfg, err := buildSSHConfig(h.db, s)
		if err != nil {
			c.JSON(http.StatusBadRequest, model.Response{
				Code: 400, Message: fmt.Sprintf("服务器 %s 认证配置错误: %v", s.Hostname, err),
			})
			return
		}
		configs = append(configs, cfg)
	}

	// 并发上传
	results := remoteexec.BatchUpload(configs, req.LocalPath, req.RemotePath, 5)

	c.JSON(http.StatusOK, model.Response{Code: 200, Data: results})
}

// ==================== 主机匹配 ====================

// MatchHosts 预览匹配的主机列表（从资产服务器表查询）// Logan.Li
// GET /api/v1/remote-exec/hosts?pattern=web*&match_type=prefix
func (h *RemoteExecHandler) MatchHosts(c *gin.Context) {
	pattern := c.Query("pattern")
	matchType := c.DefaultQuery("match_type", "contains")

	query := h.db.Model(&model.Server{})

	if pattern != "" {
		switch matchType {
		case "prefix":
			query = query.Where("hostname LIKE ? OR ip LIKE ?", pattern+"%", pattern+"%")
		case "regex":
			query = query.Where("hostname REGEXP ? OR ip REGEXP ?", pattern, pattern)
		default: // contains
			query = query.Where("hostname LIKE ? OR ip LIKE ? OR tags LIKE ?",
				"%"+pattern+"%", "%"+pattern+"%", "%"+pattern+"%")
		}
	}

	var total int64
	query.Count(&total)

	var items []model.Server
	query.Order("hostname").Find(&items)

	// 补充 HasAuth 字段
	for i := range items {
		items[i].HasAuth = items[i].Password != "" || items[i].SSHKey != ""
	}

	c.JSON(http.StatusOK, model.Response{Code: 200, Data: model.ListResponse{List: items, Total: total}})
}

// ==================== 单机命令执行 ====================

// ExecOnServer 在指定服务器上执行命令并返回结果
// POST /api/v1/remote-exec/server/:id/exec
func (h *RemoteExecHandler) ExecOnServer(c *gin.Context) {
	id, err := strconv.Atoi(c.Param("id"))
	if err != nil {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "无效的服务器 ID"})
		return
	}

	var req struct {
		Command string `json:"command" binding:"required"`
		Timeout int    `json:"timeout"`
	}
	if err := c.ShouldBindJSON(&req); err != nil {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "参数错误: " + err.Error()})
		return
	}

	var server model.Server
	if err := h.db.First(&server, id).Error; err != nil {
		c.JSON(http.StatusNotFound, model.Response{Code: 404, Message: "服务器不存在"})
		return
	}

	cfg, err := buildSSHConfig(h.db, server)
	if err != nil {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "认证配置错误: " + err.Error()})
		return
	}

	timeout := req.Timeout
	if timeout <= 0 {
		timeout = 300
	}

	result, err := remoteexec.Exec(cfg, req.Command, time.Duration(timeout)*time.Second)
	if err != nil {
		c.JSON(http.StatusInternalServerError, model.Response{Code: 500, Message: err.Error(), Data: result})
		return
	}

	c.JSON(http.StatusOK, model.Response{Code: 200, Data: result})
}

// ==================== 辅助函数 ====================

// buildSSHConfig 从 Server 模型构建 SSHConfig。
// 优先使用 server 自带的 Password/SSHKey 字段，
// 如果 server 关联了 credential_id，则从凭证表解密获取。
func buildSSHConfig(db *gorm.DB, server model.Server) (remoteexec.SSHConfig, error) {
	cfg := remoteexec.SSHConfig{
		Host: server.IP,
		Port: server.Port,
		User: server.Username,
	}
	if cfg.Port == 0 {
		cfg.Port = 22
	}
	if cfg.User == "" {
		cfg.User = "root"
	}

	// 优先使用 server 自带的认证信息
	if server.AuthType == "key" && server.SSHKey != "" {
		cfg.AuthType = remoteexec.AuthKey
		cfg.KeyData = server.SSHKey
		return cfg, nil
	}
	if server.Password != "" {
		cfg.AuthType = remoteexec.AuthPassword
		cfg.Password = server.Password
		return cfg, nil
	}

	// 从凭证表查找关联的 SSH 凭据
	// Server 模型目前没有 CredentialID 字段，通过 tags 或 name 匹配
	// 未来可扩展 server.credential_id FK

	return cfg, fmt.Errorf("服务器 %s 未配置 SSH 认证（密码或密钥）", server.Hostname)
}

// ==================== 定时任务远程执行（对接 CronJob target_type） ====================

// ExecuteCronOnTargets 根据 CronJob 的 target_type 在目标主机上执行脚本。
// 由 automation_cron.go 调用的公共函数。
func ExecuteCronOnTargets(db *gorm.DB, job model.CronJob, scriptType, content string, timeoutSec int) ([]remoteexec.ExecResult, error) {
	// 解析目标服务器
	var targetIDs []uint
	if job.Targets != "" {
		json.Unmarshal([]byte(job.Targets), &targetIDs)
	}

	if len(targetIDs) == 0 {
		return nil, fmt.Errorf("cron job %s has no targets", job.Name)
	}

	// 查询服务器
	var servers []model.Server
	switch job.TargetType {
	case "server":
		db.Where("id IN ?", targetIDs).Find(&servers)
	case "group":
		db.Where("id IN ?", targetIDs).Find(&servers)
	case "tag":
		// 标签匹配 — 通过 tags JSON 字段模糊查询
		tagStr := fmt.Sprintf("%v", targetIDs)
		db.Where("tags LIKE ?", "%"+tagStr+"%").Find(&servers)
	default:
		db.Where("id IN ?", targetIDs).Find(&servers)
	}

	if len(servers) == 0 {
		return nil, fmt.Errorf("no target servers found for cron job %s", job.Name)
	}

	// 构建 SSH 配置
	configs := make([]remoteexec.SSHConfig, 0, len(servers))
	for _, s := range servers {
		cfg, err := buildSSHConfig(db, s)
		if err != nil {
			continue // 跳过认证失败的主机
		}
		configs = append(configs, cfg)
	}

	if len(configs) == 0 {
		return nil, fmt.Errorf("all target servers have auth errors for cron job %s", job.Name)
	}

	// 并发执行
	if timeoutSec <= 0 {
		timeoutSec = 300
	}
	results := remoteexec.BatchExec(configs, content, time.Duration(timeoutSec)*time.Second, 5)
	return results, nil
}