aiops/backend/controller/agent_install.go

package controller

import (
	"fmt"
	"log"
	"net"
	"net/http"
	"strconv"
	"strings"
	"time"

	"ezops/model"

	"github.com/gin-gonic/gin"
	"golang.org/x/crypto/ssh"
	"gorm.io/gorm"
)

// AgentInstallHandler Agent 远程安装处理器 // Logan.Li
type AgentInstallHandler struct {
	db *gorm.DB
}

func NewAgentInstallHandler(db *gorm.DB) *AgentInstallHandler {
	return &AgentInstallHandler{db: db}
}

// sshDial 复用 webshell.go 的 SSH 连接逻辑（TOFU HostKey 验证）
func sshDial(server model.Server) (*ssh.Client, error) {
	config := &ssh.ClientConfig{
		User:    server.Username,
		Timeout: 10 * time.Second,
		HostKeyCallback: func(hostname string, remote net.Addr, key ssh.PublicKey) error {
			// TOFU: 记录并信任首次连接的 HostKey
			fp := ssh.FingerprintSHA256(key)
			log.Printf("[AgentInstall] SSH TOFU: trusting %s (%s): %s", server.Hostname, hostname, fp)
			return nil
		},
	}

	switch server.AuthType {
	case "key":
		signer, err := ssh.ParsePrivateKey([]byte(server.SSHKey))
		if err != nil {
			config.Auth = []ssh.AuthMethod{ssh.Password(server.Password)}
		} else {
			config.Auth = []ssh.AuthMethod{ssh.PublicKeys(signer)}
		}
	default:
		config.Auth = []ssh.AuthMethod{ssh.Password(server.Password)}
	}

	addr := server.IP + ":" + strconv.Itoa(server.Port)
	return ssh.Dial("tcp", addr, config)
}

// runSSHCommand 在远程主机执行命令并返回输出
func runSSHCommand(client *ssh.Client, cmd string) (string, error) {
	session, err := client.NewSession()
	if err != nil {
		return "", err
	}
	defer session.Close()

	output, err := session.CombinedOutput(cmd)
	return string(output), err
}

// InstallAgent 远程安装 Agent 到指定服务器
// POST /servers/:id/install-agent
// 支持传入临时凭证：首次连接成功后自动更新服务器认证凭据
func (h *AgentInstallHandler) InstallAgent(c *gin.Context) {
	id, _ := strconv.Atoi(c.Param("id"))

	var req struct {
		Username string `json:"username"`
		Password string `json:"password"`
		SSHKey   string `json:"ssh_key"`
		AuthType string `json:"auth_type"`
		Token    string `json:"token"`
	}
	// 兼容旧调用方式：token 在 query，新方式在 body
	c.ShouldBindJSON(&req)
	if req.Token == "" {
		req.Token = c.Query("token")
	}
	if req.AuthType == "" {
		req.AuthType = "password"
	}

	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
	}

	if server.Status != "running" {
		// 允许安装，不强制 running 状态
	}

	// 如果请求中提供了凭据，使用请求凭据覆盖服务器存储的凭据进行 SSH
	effectiveServer := server
	if req.Username != "" {
		effectiveServer.Username = req.Username
	}
	if req.Password != "" {
		effectiveServer.Password = req.Password
	}
	if req.SSHKey != "" {
		effectiveServer.SSHKey = req.SSHKey
		effectiveServer.AuthType = "key"
	}
	if req.AuthType != "" && req.AuthType != effectiveServer.AuthType {
		effectiveServer.AuthType = req.AuthType
	}

	// 1. SSH 连接目标主机
	client, err := sshDial(effectiveServer)
	if err != nil {
		log.Printf("[AgentInstall] SSH连接失败 %s: %v", server.IP, err)
		c.JSON(http.StatusOK, model.Response{
			Code:    500,
			Message: "SSH 连接失败: " + formatSSHError(err),
		})
		return
	}
	defer client.Close()

	// 首次连接成功 → 如果服务器之前未配置凭据，自动保存使用的凭据
	if (server.Password == "" && server.SSHKey == "") || server.Username == "" {
		updates := map[string]interface{}{
			"username":  effectiveServer.Username,
			"auth_type": effectiveServer.AuthType,
		}
		if effectiveServer.Password != "" {
			updates["password"] = effectiveServer.Password
		}
		if effectiveServer.SSHKey != "" {
			updates["ssh_key"] = effectiveServer.SSHKey
		}
		if server.Username == "" {
			updates["username"] = effectiveServer.Username
		}
		h.db.Model(&server).Updates(updates)
		log.Printf("[AgentInstall] %s 首次连接成功，已自动关联凭证", server.Hostname)
	}

	// 2. 检测系统架构
	archOut, _ := runSSHCommand(client, "uname -m")
	arch := "amd64"
	if len(archOut) > 0 && (contains(archOut, "aarch64") || contains(archOut, "arm64")) {
		arch = "arm64"
	}

	// 3. 构造安装命令（使用 Agent API 服务地址）
	// 从请求头获取后端可访问地址
	serverURL := fmt.Sprintf("http://%s", c.Request.Host)
	token := c.Query("token")

	installCmd := fmt.Sprintf(
		`curl -fsSL %s/agent/install.sh | sudo bash -s -- --server %s --token %s`,
		serverURL, serverURL, token,
	)

	// 4. 执行安装
	installOut, err := runSSHCommand(client, installCmd)
	if err != nil {
		c.JSON(http.StatusOK, model.Response{
			Code:    500,
			Message: "安装命令执行失败（可能需要 sudo 权限或 curl 未安装）",
			Data:    installOut,
		})
		return
	}

	// 5. 验证安装是否成功
	verifyOut, _ := runSSHCommand(client, "systemctl is-active aiops-agent 2>/dev/null || echo 'not_found'")
	installed := contains(verifyOut, "active") || contains(verifyOut, "running")

	if installed {
		// 更新服务器 Agent 状态（版本号：安装成功后由 Agent 下一次上报自动更新）
		h.db.Model(&server).Updates(map[string]interface{}{
			"agent_status":  "online",
			"agent_version": "v0.1.0",
			"last_check_at": time.Now(),
		})
	}

	c.JSON(http.StatusOK, model.Response{
		Code:    200,
		Message: func() string {
			if installed {
				return "Agent 安装成功，已自动启动并设置开机自启"
			}
			return "安装命令已执行，但 Agent 服务未检测到运行（请查看日志排查）"
		}(),
		Data: gin.H{
			"installed": installed,
			"arch":      arch,
			"output":    installOut,
			"verify":    verifyOut,
		},
	})
}

// installOne 安装单个服务器的 Agent（内部方法）
func (h *AgentInstallHandler) installOne(server model.Server, serverURL, token string) (bool, string, string, error) {
	client, err := sshDial(server)
	if err != nil {
		return false, "", "", fmt.Errorf("SSH连接失败: %w", err)
	}
	defer client.Close()

	// 不强制 server.Status==running：只要 SSH 可达即可安装（DB 状态可能滞后于实际开机状态） // Logan.Li

	installCmd := fmt.Sprintf(
		`curl -fsSL %s/agent/install.sh | sudo bash -s -- --server %s --token %s`,
		serverURL, serverURL, token,
	)

	installOut, err := runSSHCommand(client, installCmd)
	if err != nil {
		return false, installOut, "", err
	}

	verifyOut, _ := runSSHCommand(client, "systemctl is-active aiops-agent 2>/dev/null || echo 'not_found'")
	installed := contains(verifyOut, "active") || contains(verifyOut, "running")

	if installed {
		h.db.Model(&server).Updates(map[string]interface{}{
			"agent_status":  "online",
			"agent_version": "v0.1.0",
			"last_check_at": time.Now(),
		})
	}

	return installed, installOut, verifyOut, nil
}

// BatchInstallAgent 批量安装 Agent（异步执行，立即返回）// Logan.Li
// POST /servers/install-agent
func (h *AgentInstallHandler) BatchInstallAgent(c *gin.Context) {
	var req struct {
		ServerIDs []uint `json:"server_ids"`
	}
	if err := c.ShouldBindJSON(&req); err != nil || len(req.ServerIDs) == 0 {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "请提供要安装的服务器 ID 列表"})
		return
	}

	serverURL := fmt.Sprintf("http://%s", c.Request.Host)
	token := c.Query("token")

	// 异步执行：立即返回，后台 goroutine 逐台安装
	go func() {
		successCount := 0
		for _, sid := range req.ServerIDs {
			var server model.Server
			if err := h.db.First(&server, sid).Error; err != nil {
				log.Printf("[BatchInstall] 服务器 %d 不存在，跳过", sid)
				continue
			}
			installed, _, _, err := h.installOne(server, serverURL, token)
			if err != nil {
				log.Printf("[BatchInstall] %s (%s) 安装失败: %v", server.Hostname, server.IP, err)
			} else {
				successCount++
				if installed {
					log.Printf("[BatchInstall] %s Agent 安装成功", server.Hostname)
				} else {
					log.Printf("[BatchInstall] %s Agent 安装命令已执行但服务未检测到运行", server.Hostname)
				}
			}
		}
		log.Printf("[BatchInstall] 批量安装完成: %d/%d 成功", successCount, len(req.ServerIDs))
	}()

	c.JSON(http.StatusOK, model.Response{
		Code:    200,
		Message: fmt.Sprintf("已提交异步安装任务，共 %d 台主机，正在后台执行…", len(req.ServerIDs)),
		Data:    gin.H{"total": len(req.ServerIDs), "async": true},
	})
}

// UninstallAgent 远程卸载 Agent
// POST /api/v1/agent/uninstall
func (h *AgentInstallHandler) UninstallAgent(c *gin.Context) {
	var req struct {
		Hostname string `json:"hostname"`
		ServerID *uint  `json:"server_id"`
	}
	if err := c.ShouldBindJSON(&req); err != nil {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "参数错误"})
		return
	}

	var server model.Server
	if req.ServerID != nil {
		if err := h.db.First(&server, *req.ServerID).Error; err != nil {
			c.JSON(http.StatusNotFound, model.Response{Code: 404, Message: "服务器不存在"})
			return
		}
	} else if req.Hostname != "" {
		if err := h.db.Where("hostname = ? OR ip = ?", req.Hostname, req.Hostname).First(&server).Error; err != nil {
			c.JSON(http.StatusNotFound, model.Response{Code: 404, Message: "未找到服务器: " + req.Hostname})
			return
		}
	} else {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "请提供 hostname 或 server_id"})
		return
	}

	// 无 SSH 凭据的主机（自动注册的）直接改本地状态；注意 SSH key 认证的主机 Password 为空但 SSHKey 非空，仍可远程卸载 // Logan.Li
	if server.Username == "" || (server.Password == "" && server.SSHKey == "") {
		h.db.Model(&server).Updates(map[string]interface{}{
			"agent_status": "offline", "agent_version": "",
		})
		c.JSON(http.StatusOK, model.Response{Code: 200, Message: "已标记为离线（该主机无SSH凭据，无法远程卸载，请手动执行 systemctl stop aiops-agent）"})
		return
	}

	client, err := sshDial(server)
	if err != nil {
		c.JSON(http.StatusOK, model.Response{Code: 500, Message: "SSH连接失败: " + formatSSHError(err)})
		return
	}
	defer client.Close()

	runSSHCommand(client, "systemctl stop aiops-agent 2>/dev/null; systemctl disable aiops-agent 2>/dev/null; rm -f /usr/local/bin/aiops-agent /etc/systemd/system/aiops-agent.service; systemctl daemon-reload 2>/dev/null")

	h.db.Model(&server).Updates(map[string]interface{}{
		"agent_status": "offline", "agent_version": "",
	})

	c.JSON(http.StatusOK, model.Response{Code: 200, Message: fmt.Sprintf("已卸载 %s (%s) 的 Agent", server.Hostname, server.IP)})
}

// BatchUninstallAgent 批量卸载 Agent（异步执行，立即返回）// Logan.Li
// POST /api/v1/agent/batch-uninstall
func (h *AgentInstallHandler) BatchUninstallAgent(c *gin.Context) {
	var req struct {
		ServerIDs []uint `json:"server_ids"`
	}
	if err := c.ShouldBindJSON(&req); err != nil || len(req.ServerIDs) == 0 {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "请提供服务器 ID 列表"})
		return
	}

	// 异步执行：立即返回，后台逐台卸载
	go func() {
		successCount := 0
		for _, id := range req.ServerIDs {
			var server model.Server
			if err := h.db.First(&server, id).Error; err != nil {
				log.Printf("[BatchUninstall] 服务器 %d 不存在，跳过", id)
				continue
			}
			if server.Username == "" || (server.Password == "" && server.SSHKey == "") {
				h.db.Model(&server).Updates(map[string]interface{}{"agent_status": "offline", "agent_version": ""})
				log.Printf("[BatchUninstall] %s 无SSH凭据，已标记离线", server.Hostname)
				successCount++
				continue
			}
			client, err := sshDial(server)
			if err != nil {
				log.Printf("[BatchUninstall] %s SSH失败: %v", server.Hostname, err)
				continue
			}
			runSSHCommand(client, "systemctl stop aiops-agent 2>/dev/null; systemctl disable aiops-agent 2>/dev/null; rm -f /usr/local/bin/aiops-agent /etc/systemd/system/aiops-agent.service; systemctl daemon-reload 2>/dev/null")
			client.Close()
			h.db.Model(&server).Updates(map[string]interface{}{"agent_status": "offline", "agent_version": ""})
			log.Printf("[BatchUninstall] %s Agent 已卸载", server.Hostname)
			successCount++
		}
		log.Printf("[BatchUninstall] 批量卸载完成: %d/%d 成功", successCount, len(req.ServerIDs))
	}()

	c.JSON(http.StatusOK, model.Response{
		Code:    200,
		Message: fmt.Sprintf("已提交异步卸载任务，共 %d 台主机，正在后台执行…", len(req.ServerIDs)),
		Data:    gin.H{"total": len(req.ServerIDs), "async": true},
	})
}

// BatchUpgradeAgent 批量升级 Agent 到最新版本（SSH下载新二进制 + 重启服务）
// POST /api/v1/agent/batch-upgrade
func (h *AgentInstallHandler) BatchUpgradeAgent(c *gin.Context) {
	var req struct {
		ServerIDs []uint `json:"server_ids"`
	}
	if err := c.ShouldBindJSON(&req); err != nil || len(req.ServerIDs) == 0 {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "请提供要升级的服务器 ID 列表"})
		return
	}

	serverURL := fmt.Sprintf("http://%s", c.Request.Host)

	// 异步执行：立即返回，后台逐台升级
	go func() {
		successCount := 0
		for _, sid := range req.ServerIDs {
			var server model.Server
			if err := h.db.First(&server, sid).Error; err != nil {
				log.Printf("[BatchUpgrade] 服务器 %d 不存在，跳过", sid)
				continue
			}
			if server.Username == "" || (server.Password == "" && server.SSHKey == "") {
				log.Printf("[BatchUpgrade] %s 无SSH凭据，跳过", server.Hostname)
				continue
			}

			// 检测架构
			client, err := sshDial(server)
			if err != nil {
				log.Printf("[BatchUpgrade] %s SSH连接失败: %v", server.Hostname, err)
				continue
			}

			archOut, _ := runSSHCommand(client, "uname -m")
			binaryFile := "aiops-agent-linux-amd64"
			if len(archOut) > 0 && (contains(archOut, "aarch64") || contains(archOut, "arm64")) {
				binaryFile = "aiops-agent-linux-arm64"
			}

			// 下载新二进制 + 获取版本号 + 重启服务
			upgradeScript := fmt.Sprintf(
				`curl -fsSL -o /tmp/aiops-agent %s/api/agent/bin/%s && chmod +x /tmp/aiops-agent && systemctl stop aiops-agent 2>/dev/null; mv /tmp/aiops-agent /usr/local/bin/aiops-agent && systemctl start aiops-agent; ver=$(/usr/local/bin/aiops-agent output 2>/dev/null | grep 'agent.version' | awk -F'[|[|]' '{print $2}' | tr -d ' '); echo "UPGRADE_OK ver=$ver"`,
				serverURL, binaryFile,
			)

			out, err := runSSHCommand(client, upgradeScript)
			client.Close()

			if err != nil || !contains(out, "UPGRADE_OK") {
				log.Printf("[BatchUpgrade] %s 升级失败: output=%s err=%v", server.Hostname, out, err)
				continue
			}

			// 从输出提取版本号
			agentVersion := "unknown"
			parts := strings.Split(out, "ver=")
			if len(parts) > 1 {
				agentVersion = strings.TrimSpace(parts[1])
			}

			h.db.Model(&server).Updates(map[string]interface{}{
				"agent_version": agentVersion,
				"last_check_at": time.Now(),
			})
			log.Printf("[BatchUpgrade] %s Agent 升级成功", server.Hostname)
			successCount++
		}
		log.Printf("[BatchUpgrade] 批量升级完成: %d/%d 成功", successCount, len(req.ServerIDs))
	}()

	c.JSON(http.StatusOK, model.Response{
		Code:    200,
		Message: fmt.Sprintf("已提交异步升级任务，共 %d 台主机，正在后台执行…", len(req.ServerIDs)),
		Data:    gin.H{"total": len(req.ServerIDs), "async": true},
	})
}

// RemoteExec 直连 SSH 执行命令（即时返回，无轮询）
// POST /api/v1/agent/exec
func (h *AgentInstallHandler) RemoteExec(c *gin.Context) {
	var req struct {
		Hostname string `json:"hostname"`
		Command  string `json:"command"`
		Timeout  int    `json:"timeout"`
	}
	if err := c.ShouldBindJSON(&req); err != nil || req.Hostname == "" || req.Command == "" {
		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "请提供 hostname 和 command"})
		return
	}
	if req.Timeout <= 0 {
		req.Timeout = 30
	}

	var server model.Server
	if err := h.db.Where("hostname = ? OR ip = ?", req.Hostname, req.Hostname).First(&server).Error; err != nil {
		c.JSON(http.StatusNotFound, model.Response{Code: 404, Message: "未找到服务器: " + req.Hostname})
		return
	}

	client, err := sshDial(server)
	if err != nil {
		c.JSON(http.StatusOK, model.Response{Code: 500, Message: "SSH连接失败: " + err.Error()})
		return
	}
	defer client.Close()

	type result struct {
		out string
		err error
	}
	done := make(chan result, 1)
	go func() {
		out, err := runSSHCommand(client, req.Command)
		done <- result{out, err}
	}()

	select {
	case r := <-done:
		if r.err != nil {
			c.JSON(http.StatusOK, model.Response{Code: 200, Message: "执行完成", Data: gin.H{"output": r.out, "error": r.err.Error()}})
			return
		}
		c.JSON(http.StatusOK, model.Response{Code: 200, Message: "执行成功", Data: gin.H{"output": r.out}})
	case <-time.After(time.Duration(req.Timeout) * time.Second):
		c.JSON(http.StatusOK, model.Response{Code: 408, Message: fmt.Sprintf("命令执行超时(%ds)", req.Timeout)})
	}
}