重构 auto-check：状态数据下沉各包，workflow 精简为纯调度

- discovery: ping 探测存活 + ARP 解析 MAC，包级 Devices map，串行扫描 - stress: 脚本生成模式（BuildScript/ParseResults），ToolSet 工具检测，包级 Results map - report: 包级 Printed map，SaveAndPrintDeviceReport 保存+打印 - sshclient: 包级 RunCommand - workflow: 仅持有 config，Start() 固定循环，直接调用各包方法 - config: 移除 ScanConfig.Concurrency（串行扫描无需并发数）
2026-08-12 17:11:59 +08:00
parent 700ba91883
commit 906bb54aaa
15 changed files with 708 additions and 871 deletions
--- a/auto-check/pkg/stress/runner.go
+++ b/auto-check/pkg/stress/runner.go
@@ -2,6 +2,7 @@ package stress

 import (
 	"fmt"
+	"strconv"
 	"strings"
 	"time"

@@ -14,145 +15,212 @@ import (
 // Results 测试结果表（IP → 报告），包级公开
 var Results = make(map[string]*Report)

-// Runner 压力测试调度器
+// ============================
+// 执行器
+// ============================
+
+// Runner 压力测试执行器
 type Runner struct {
-	Config   Config
-	Types    []TestType
-	Executor Executor
+	cfg     Config
+	types   []TestType
+	client  *ssh.Client
+	tools   ToolSet
+	metrics []Metric
 }

-// NewRunner 创建调度器
-func NewRunner(cfg Config, types []TestType, executor Executor) *Runner {
-	return &Runner{
-		Config:   cfg,
-		Types:    types,
-		Executor: executor,
-	}
+// NewRunner 创建执行器（检测工具 → 构建指标 → 生成脚本 → 执行）
+func NewRunner(cfg Config, types []TestType, client *ssh.Client) *Runner {
+	tools := DetectTools(client)
+	metrics := BuildMetrics(types, tools)
+	return &Runner{cfg: cfg, types: types, client: client, tools: tools, metrics: metrics}
 }

-// Run 执行压力测试（单台主机）
-func (r *Runner) Run(client *ssh.Client, ip string) *Report {
+// Run 执行压力测试（生成脚本 → 远程执行 → 解析结果）
+func (r *Runner) Run(ip string) *Report {
 	report := &Report{IP: ip, StartTime: time.Now()}
 	fmt.Printf("\n════════ [%s] 压力测试开始 ════════\n", ip)

-	// 系统信息探测
-	info := ProbeSystem(client, r.Executor)
-	fmt.Printf("  主机: %s | CPU: %s (%s核) | 内存: %s | 内核: %s\n",
-		info.Hostname, info.CPUModel, info.CPUCores, info.MemTotal, info.KernelVer)
+	// 显示可用工具
+	fmt.Printf("  工具: ")
+	for name, info := range r.tools.Tools {
+		fmt.Printf("%s(%s) ", name, info.Version)
+	}
+	fmt.Println()

-	// 启动温度监控
-	tempMon := NewTempMonitor(client, r.Executor, r.Config.TempLogInt)
-	tempMon.Start()
-
-	// 启动 dmesg 监控
-	dmesgMon := NewDmesgMonitor(client, r.Executor)
-	dmesgMon.Start()
-
-	// 逐项执行测试
-	for _, t := range r.Types {
-		fmt.Printf("\n  ── 开始: %s ──\n", t)
-		var res Result
-
-		switch t {
-		case TestCPU:
-			res = CPUStress(client, r.Executor, r.Config)
-		case TestMemory:
-			res = MemoryStress(client, r.Executor, r.Config)
-		case TestDiskIO:
-			res = DiskIOStress(client, r.Executor, r.Config)
-		case TestMemNative:
-			res = MemNativeStress(client, r.Executor, r.Config)
-		case TestFull:
-			res = FullStress(client, r.Executor, r.Config)
-		default:
-			res = Result{Type: t, Status: "skip", Error: fmt.Sprintf("未知测试类型: %s", t)}
+	// 显示将执行的指标
+	fmt.Printf("  指标: ")
+	for _, m := range r.metrics {
+		if m.Enabled {
+			fmt.Printf("%s ", m.Name)
 		}
+	}
+	fmt.Println()

-		report.AddResult(res)
-		fmt.Printf("  ── 完成: %s [%s] %s ──\n", t, res.Status, res.Duration.Round(time.Millisecond))
+	// 生成脚本
+	script := BuildScript(r.metrics, r.cfg)
+
+	// 远程执行脚本（通过 stdin 传入）
+	fmt.Printf("  [执行] 生成脚本 %d 字节，开始远程执行...\n", len(script))
+	scriptCmd := fmt.Sprintf("cat <<'AUTOCHECKSCRIPT' > /tmp/auto-check.sh\n%s\nAUTOCHECKSCRIPT\nchmod +x /tmp/auto-check.sh && bash /tmp/auto-check.sh 2>&1", script)
+	output, err := sshclient.RunCommand(r.client, scriptCmd)
+
+	if err != nil {
+		fmt.Printf("  [执行] 脚本执行出错: %v\n", err)
 	}

-	// 停止监控，收集结果
-	tempLogs := tempMon.Stop()
-	dmesgErrors := dmesgMon.Stop()
-
-	// 添加温度报告
-	if len(tempLogs) > 0 {
-		maxTemp := ""
-		for _, line := range tempLogs {
-			// 提取温度值
-			parts := strings.Fields(line)
-			for _, p := range parts {
-				if strings.HasSuffix(p, "°C") || strings.HasSuffix(p, "C") {
-					maxTemp = p
-				}
-			}
-		}
-		tempSummary := fmt.Sprintf("采样 %d 次", len(tempLogs))
-		if maxTemp != "" {
-			tempSummary += ", 最高温度: " + maxTemp
-		}
-		status := "pass"
-		if strings.Contains(strings.ToLower(strings.Join(tempLogs, " ")), "throttl") {
-			status = "fail"
-			tempSummary += " [检测到降频!]"
-		}
-		report.AddResult(Result{
-			Type:     MonitorTemp,
-			Status:   status,
-			Output:   tempSummary,
-			Duration: time.Since(report.StartTime),
-		})
-	}
-
-	// 添加 dmesg 报告
-	if len(dmesgErrors) > 0 {
-		report.AddResult(Result{
-			Type:     MonitorDmesg,
-			Status:   "fail",
-			Output:   fmt.Sprintf("检测到 %d 条硬件相关报错:\n%s", len(dmesgErrors), strings.Join(dmesgErrors[:min(10, len(dmesgErrors))], "\n")),
-			Duration: time.Since(report.StartTime),
-		})
-	} else {
-		report.AddResult(Result{
-			Type:     MonitorDmesg,
-			Status:   "pass",
-			Output:   "无硬件相关内核报错",
-			Duration: time.Since(report.StartTime),
-		})
+	// 解析结果
+	if output != "" {
+		r.parseResults(report, output)
 	}

 	report.EndTime = time.Now()
 	report.Duration = report.EndTime.Sub(report.StartTime)
-
 	fmt.Printf("\n════════ [%s] 压力测试完成 ════════\n", ip)
 	fmt.Println(report.ToText())
 	return report
 }

-// RunAll 对多台主机执行压力测试
-func (r *Runner) RunAll(clients map[string]*ssh.Client) []*Report {
-	var reports []*Report
-	for ip, client := range clients {
-		reports = append(reports, r.Run(client, ip))
+// parseResults 解析脚本的结构化输出
+func (r *Runner) parseResults(report *Report, output string) {
+	lines := strings.Split(output, "\n")
+	var currentTest string
+	var currentStatus string
+	var currentDuration string
+	var currentOutput []string
+
+	flush := func() {
+		if currentTest == "" {
+			return
+		}
+		status := StatusPass
+		switch currentStatus {
+		case "fail":
+			status = StatusFail
+		case "skip":
+			status = StatusSkip
+		case "error":
+			status = StatusError
+		}
+		duration, _ := time.ParseDuration(currentDuration)
+		report.AddResult(Result{
+			Type:     TestType(currentTest),
+			Status:   status,
+			Output:   strings.Join(currentOutput, "\n"),
+			Duration: duration,
+		})
+		currentTest = ""
+		currentStatus = ""
+		currentDuration = ""
+		currentOutput = nil
+	}
+
+	for _, line := range lines {
+		line = strings.TrimSpace(line)
+
+		if strings.HasPrefix(line, "===TEST:") {
+			flush()
+			currentTest = strings.TrimSuffix(strings.TrimPrefix(line, "===TEST:"), "===")
+			continue
+		}
+
+		if strings.HasPrefix(line, "===MONITOR:") {
+			flush()
+			currentTest = strings.TrimSuffix(strings.TrimPrefix(line, "===MONITOR:"), "===")
+			continue
+		}
+
+		if line == "===END===" {
+			flush()
+			continue
+		}
+
+		// 解析 key:value
+		if strings.HasPrefix(line, "status:") {
+			currentStatus = strings.TrimPrefix(line, "status:")
+			continue
+		}
+		if strings.HasPrefix(line, "duration:") {
+			durStr := strings.TrimPrefix(line, "duration:")
+			ms, _ := strconv.ParseInt(strings.TrimSuffix(durStr, "ms"), 10, 64)
+			currentDuration = fmt.Sprintf("%dms", ms)
+			continue
+		}
+		if strings.HasPrefix(line, "output:") {
+			currentOutput = append(currentOutput, strings.TrimPrefix(line, "output:"))
+			continue
+		}
+		if strings.HasPrefix(line, "samples:") || strings.HasPrefix(line, "max:") {
+			currentOutput = append(currentOutput, line)
+			continue
+		}
+
+		// 普通输出行
+		if currentTest != "" && line != "" {
+			currentOutput = append(currentOutput, line)
+		}
+	}
+	flush()
+
+	// 从系统信息输出解析主机信息
+	for _, line := range lines {
+		if strings.HasPrefix(line, "hostname:") {
+			// 可扩展：存入 report 的系统信息字段
+		}
 	}
-	return reports
 }

+// ============================
+// 业务入口（workflow 调用）
+// ============================
+
+// TestDevice 对单台设备执行 SSH 登录 + 压力测试（业务入口）
+func TestDevice(ip string, cfg config.Config) *Report {
+	sshCfg := sshclient.NewConfig(cfg.SSH.User, cfg.SSH.Password, cfg.SSH.KeyFile, cfg.SSH.Port, cfg.Scan.Timeout)
+
+	conn, err := sshclient.Connect(ip, sshCfg)
+	if err != nil {
+		fmt.Printf("  [测试] %s SSH 连接失败: %v\n", ip, err)
+		return NewSSHFailReport(ip, err)
+	}
+	defer conn.Close()
+
+	stressCfg := Config{
+		Duration:   cfg.Stress.Duration,
+		Threads:    cfg.Stress.Threads,
+		DiskSizeMB: 1024,
+		TempLogInt: 10 * time.Second,
+	}
+	return NewRunner(stressCfg, ParseTypes(cfg.Stress.Types), conn).Run(ip)
+}
+
+// NewSSHFailReport 构造 SSH 连接失败报告
+func NewSSHFailReport(ip string, err error) *Report {
+	return &Report{
+		IP:        ip,
+		StartTime: time.Now(),
+		Results:   []Result{{Type: "ssh", Status: StatusFail, Error: err.Error()}},
+		Failed:    1,
+	}
+}
+
+// ============================
+// 工具函数
+// ============================
+
 // IsPassed 报告是否通过
 func IsPassed(rpt *Report) bool {
 	return rpt != nil && rpt.Failed == 0 && rpt.Errors == 0
 }

-// Status 报告状态字符串：pass / fail / untested
+// Status 报告状态字符串
 func Status(rpt *Report) string {
 	if rpt == nil {
-		return "untested"
+		return StatusUnknown
 	}
 	if IsPassed(rpt) {
-		return "pass"
+		return StatusPass
 	}
-	return "fail"
+	return StatusFail
 }

 // ParseTypes 逗号分隔字符串 → TestType 列表
@@ -165,39 +233,3 @@ func ParseTypes(s string) []TestType {
 	}
 	return types
 }
-
-// TestDevice 对单台设备执行 SSH 登录 + 压力测试（业务入口）
-func TestDevice(ip string, cfg config.Config) *Report {
-	client := sshclient.NewClient(cfg.SSH.User, cfg.SSH.Password, cfg.SSH.KeyFile, cfg.SSH.Port, cfg.Scan.Timeout)
-
-	conn, err := client.Connect(ip)
-	if err != nil {
-		fmt.Printf("  [测试] %s SSH 连接失败: %v\n", ip, err)
-		return NewSSHFailReport(ip, err)
-	}
-	defer conn.Close()
-
-	return NewRunner(
-		Config{Duration: cfg.Stress.Duration, Threads: cfg.Stress.Threads, DiskSizeMB: 1024, TempLogInt: 10 * time.Second},
-		ParseTypes(cfg.Stress.Types),
-		client,
-	).Run(conn, ip)
-}
-
-// NewSSHFailReport 构造 SSH 连接失败报告
-func NewSSHFailReport(ip string, err error) *Report {
-	return &Report{
-		IP:        ip,
-		StartTime: time.Now(),
-		Results:   []Result{{Type: "ssh", Status: "fail", Error: err.Error()}},
-		Failed:    1,
-	}
-}
-
-// min 取较小值
-func min(a, b int) int {
-	if a < b {
-		return a
-	}
-	return b
-}