Files
membank/auto-check/pkg/stress/runner.go
张威33321 19f6050147 新增 auto-check 工具:网段扫描+SSH登录+压力测试永久循环工作流
- pkg/discovery: TCP端口探测存活主机,ARP表解析MAC作为设备唯一标识
- pkg/sshclient: SSH连接(密码/密钥认证)与远程命令执行
- pkg/stress: stress-ng/stressapptest 压测(cpu/memory/disk/memnative/full)+温度与dmesg监控
- pkg/report: 检测报告生成(文本/JSON)
- pkg/workflow: 永久循环工作流(间隔可配),MAC唯一标识设备,增量测试+状态变化打印
- pkg/config: YAML分类配置(scan/ssh/stress/report/workflow)
- cmd: cobra入口,仅 --config 指定配置文件
2026-08-10 20:30:23 +08:00

142 lines
3.5 KiB
Go

package stress
import (
"fmt"
"strings"
"time"
"golang.org/x/crypto/ssh"
)
// Runner 压力测试调度器
type Runner struct {
Config Config
Types []TestType
Executor Executor
}
// NewRunner 创建调度器
func NewRunner(cfg Config, types []TestType, executor Executor) *Runner {
return &Runner{
Config: cfg,
Types: types,
Executor: executor,
}
}
// Run 执行压力测试(单台主机)
func (r *Runner) Run(client *ssh.Client, 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)
// 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)}
}
report.AddResult(res)
fmt.Printf(" ── 完成: %s [%s] %s ──\n", t, res.Status, res.Duration.Round(time.Millisecond))
}
// 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),
})
}
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))
}
return reports
}
func min(a, b int) int {
if a < b {
return a
}
return b
}