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
This commit is contained in:
张威33321
141
auto-check/pkg/stress/runner.go
Normal file
141
auto-check/pkg/stress/runner.go
Normal file
@@ -0,0 +1,141 @@
|
||||
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
|
||||
}
|
||||
168
auto-check/pkg/stress/runners.go
Normal file
168
auto-check/pkg/stress/runners.go
Normal file
@@ -0,0 +1,168 @@
|
||||
package stress
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"golang.org/x/crypto/ssh"
|
||||
)
|
||||
|
||||
// truncate 截断过长输出
|
||||
func truncate(s string, maxLen int) string {
|
||||
s = strings.TrimSpace(s)
|
||||
if len(s) > maxLen {
|
||||
return s[:maxLen] + "\n ... (输出已截断)"
|
||||
}
|
||||
return s
|
||||
}
|
||||
|
||||
// ============================
|
||||
// stress-ng 压测
|
||||
// ============================
|
||||
|
||||
// CPUStress stress-ng CPU 压力测试
|
||||
func CPUStress(client *ssh.Client, ex Executor, cfg Config) Result {
|
||||
start := time.Now()
|
||||
fmt.Printf(" [CPU] stress-ng CPU 压力 (%d线程, %v)...\n", cfg.Threads, cfg.Duration)
|
||||
|
||||
info, err := EnsureTool(client, ex, "stress-ng", "apt install stress-ng / opkg install stress-ng")
|
||||
if err != nil {
|
||||
return Result{Type: TestCPU, Status: "skip", Error: err.Error(), Duration: time.Since(start)}
|
||||
}
|
||||
fmt.Printf(" [CPU] 使用 %s (%s)\n", info.Path, info.Version)
|
||||
|
||||
cmd := fmt.Sprintf("%s --cpu %d --cpu-method all --timeout %v --metrics-brief --temp-path /tmp 2>&1",
|
||||
info.Path, cfg.Threads, cfg.Duration)
|
||||
out, err := exec(client, ex, cmd)
|
||||
duration := time.Since(start)
|
||||
|
||||
if err != nil {
|
||||
return Result{Type: TestCPU, Status: "error", Output: truncate(out, 600), Error: err.Error(), Duration: duration}
|
||||
}
|
||||
return Result{Type: TestCPU, Status: "pass", Output: truncate(out, 600), Duration: duration}
|
||||
}
|
||||
|
||||
// MemoryStress stress-ng 内存压力测试
|
||||
func MemoryStress(client *ssh.Client, ex Executor, cfg Config) Result {
|
||||
start := time.Now()
|
||||
fmt.Printf(" [Memory] stress-ng 内存压力 (%d线程, %v)...\n", cfg.Threads, cfg.Duration)
|
||||
|
||||
info, err := EnsureTool(client, ex, "stress-ng", "apt install stress-ng / opkg install stress-ng")
|
||||
if err != nil {
|
||||
return Result{Type: TestMemory, Status: "skip", Error: err.Error(), Duration: time.Since(start)}
|
||||
}
|
||||
|
||||
memWorkers := cfg.Threads
|
||||
if memWorkers > 4 {
|
||||
memWorkers = 4
|
||||
}
|
||||
cmd := fmt.Sprintf("%s --vm %d --vm-bytes 256M --vm-method all --timeout %v --metrics-brief 2>&1",
|
||||
info.Path, memWorkers, cfg.Duration)
|
||||
out, err := exec(client, ex, cmd)
|
||||
duration := time.Since(start)
|
||||
|
||||
if err != nil {
|
||||
return Result{Type: TestMemory, Status: "error", Output: truncate(out, 600), Error: err.Error(), Duration: duration}
|
||||
}
|
||||
return Result{Type: TestMemory, Status: "pass", Output: truncate(out, 600), Duration: duration}
|
||||
}
|
||||
|
||||
// DiskIOStress stress-ng 磁盘IO压力测试
|
||||
func DiskIOStress(client *ssh.Client, ex Executor, cfg Config) Result {
|
||||
start := time.Now()
|
||||
fmt.Printf(" [DiskIO] stress-ng 磁盘IO压力 (%v)...\n", cfg.Duration)
|
||||
|
||||
info, err := EnsureTool(client, ex, "stress-ng", "apt install stress-ng / opkg install stress-ng")
|
||||
if err != nil {
|
||||
return Result{Type: TestDiskIO, Status: "skip", Error: err.Error(), Duration: time.Since(start)}
|
||||
}
|
||||
|
||||
testDir := cfg.DiskDir
|
||||
if testDir == "" {
|
||||
testDir = "/tmp/stress-disk-test"
|
||||
}
|
||||
_, _ = exec(client, ex, fmt.Sprintf("mkdir -p %s", testDir))
|
||||
|
||||
cmd := fmt.Sprintf("%s --iomix 2 --iomix-bytes %dM --timeout %v --metrics-brief 2>&1",
|
||||
info.Path, cfg.DiskSizeMB, cfg.Duration)
|
||||
out, err := exec(client, ex, cmd)
|
||||
duration := time.Since(start)
|
||||
|
||||
_, _ = exec(client, ex, fmt.Sprintf("rm -rf %s", testDir))
|
||||
|
||||
if err != nil {
|
||||
return Result{Type: TestDiskIO, Status: "error", Output: truncate(out, 600), Error: err.Error(), Duration: duration}
|
||||
}
|
||||
return Result{Type: TestDiskIO, Status: "pass", Output: truncate(out, 600), Duration: duration}
|
||||
}
|
||||
|
||||
// ============================
|
||||
// stressapptest 内存精压
|
||||
// ============================
|
||||
|
||||
// MemNativeStress stressapptest 内存稳定性精压
|
||||
func MemNativeStress(client *ssh.Client, ex Executor, cfg Config) Result {
|
||||
start := time.Now()
|
||||
fmt.Printf(" [MemNative] stressapptest 内存精压 (%v)...\n", cfg.Duration)
|
||||
|
||||
info, err := EnsureTool(client, ex, "stressapptest", "apt install stressapptest / opkg install stressapptest")
|
||||
if err != nil {
|
||||
return Result{Type: TestMemNative, Status: "skip", Error: err.Error(), Duration: time.Since(start)}
|
||||
}
|
||||
fmt.Printf(" [MemNative] 使用 %s\n", info.Path)
|
||||
|
||||
memSize := cfg.MemSizeMB
|
||||
if memSize <= 0 {
|
||||
out, _ := exec(client, ex, "free -m 2>/dev/null | awk '/Mem:/{print int($7*0.6)}'")
|
||||
out = strings.TrimSpace(out)
|
||||
if out != "" {
|
||||
fmt.Sscanf(out, "%d", &memSize)
|
||||
}
|
||||
if memSize <= 0 {
|
||||
memSize = 512
|
||||
}
|
||||
}
|
||||
|
||||
cmd := fmt.Sprintf("%s -s %d -M %d -f 0 -v 2>&1",
|
||||
info.Path, int(cfg.Duration.Seconds()), memSize)
|
||||
out, err := exec(client, ex, cmd)
|
||||
duration := time.Since(start)
|
||||
|
||||
output := truncate(out, 800)
|
||||
if strings.Contains(strings.ToUpper(output), "PASS") {
|
||||
return Result{Type: TestMemNative, Status: "pass", Output: output, Duration: duration}
|
||||
}
|
||||
if strings.Contains(strings.ToUpper(output), "FAIL") {
|
||||
return Result{Type: TestMemNative, Status: "fail", Output: output, Duration: duration}
|
||||
}
|
||||
if err != nil {
|
||||
return Result{Type: TestMemNative, Status: "error", Output: output, Error: err.Error(), Duration: duration}
|
||||
}
|
||||
return Result{Type: TestMemNative, Status: "pass", Output: output, Duration: duration}
|
||||
}
|
||||
|
||||
// ============================
|
||||
// 综合压测
|
||||
// ============================
|
||||
|
||||
// FullStress 三合一综合压测 (CPU + 内存 + 磁盘IO 同时进行)
|
||||
func FullStress(client *ssh.Client, ex Executor, cfg Config) Result {
|
||||
start := time.Now()
|
||||
fmt.Printf(" [Full] 三合一综合压测 (CPU %d线程 + 内存 + 磁盘IO, %v)...\n", cfg.Threads, cfg.Duration)
|
||||
|
||||
info, err := EnsureTool(client, ex, "stress-ng", "apt install stress-ng / opkg install stress-ng")
|
||||
if err != nil {
|
||||
return Result{Type: TestFull, Status: "skip", Error: err.Error(), Duration: time.Since(start)}
|
||||
}
|
||||
|
||||
cmd := fmt.Sprintf("%s --cpu %d --vm 2 --vm-bytes 128M --iomix 1 --iomix-bytes 256M --timeout %v --metrics-brief 2>&1",
|
||||
info.Path, cfg.Threads, cfg.Duration)
|
||||
out, err := exec(client, ex, cmd)
|
||||
duration := time.Since(start)
|
||||
|
||||
if err != nil {
|
||||
return Result{Type: TestFull, Status: "error", Output: truncate(out, 600), Error: err.Error(), Duration: duration}
|
||||
}
|
||||
return Result{Type: TestFull, Status: "pass", Output: truncate(out, 600), Duration: duration}
|
||||
}
|
||||
269
auto-check/pkg/stress/tools.go
Normal file
269
auto-check/pkg/stress/tools.go
Normal file
@@ -0,0 +1,269 @@
|
||||
package stress
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"golang.org/x/crypto/ssh"
|
||||
)
|
||||
|
||||
// ============================
|
||||
// 远程命令执行接口
|
||||
// ============================
|
||||
|
||||
// Executor 远程命令执行器
|
||||
type Executor interface {
|
||||
RunCommand(client *ssh.Client, command string) (string, error)
|
||||
}
|
||||
|
||||
// exec 执行远程命令（带超时）
|
||||
func exec(client *ssh.Client, executor Executor, cmd string) (string, error) {
|
||||
return executor.RunCommand(client, cmd)
|
||||
}
|
||||
|
||||
// execf 格式化执行
|
||||
func execf(client *ssh.Client, executor Executor, format string, args ...interface{}) (string, error) {
|
||||
return exec(client, executor, fmt.Sprintf(format, args...))
|
||||
}
|
||||
|
||||
// ============================
|
||||
// 工具检测
|
||||
// ============================
|
||||
|
||||
// ToolInfo 远程工具信息
|
||||
type ToolInfo struct {
|
||||
Available bool
|
||||
Path string
|
||||
Version string
|
||||
}
|
||||
|
||||
// DetectTool 检测远程工具是否可用
|
||||
func DetectTool(client *ssh.Client, executor Executor, name string) ToolInfo {
|
||||
// 检查路径
|
||||
out, err := execf(client, executor, "which %s 2>/dev/null", name)
|
||||
path := strings.TrimSpace(out)
|
||||
if err != nil || path == "" {
|
||||
return ToolInfo{Available: false}
|
||||
}
|
||||
|
||||
// 获取版本
|
||||
var version string
|
||||
switch name {
|
||||
case "stress-ng":
|
||||
out, _ = execf(client, executor, "%s --version 2>&1 | head -1", path)
|
||||
version = strings.TrimSpace(out)
|
||||
case "stressapptest":
|
||||
out, _ = execf(client, executor, "%s --help 2>&1 | head -1", path)
|
||||
version = strings.TrimSpace(out)
|
||||
case "lm-sensors":
|
||||
out, _ = execf(client, executor, "%s -v 2>&1 | head -1", path)
|
||||
version = strings.TrimSpace(out)
|
||||
}
|
||||
|
||||
return ToolInfo{Available: true, Path: path, Version: version}
|
||||
}
|
||||
|
||||
// EnsureTool 检测工具，不可用则返回错误提示
|
||||
func EnsureTool(client *ssh.Client, executor Executor, name string, installHint string) (ToolInfo, error) {
|
||||
info := DetectTool(client, executor, name)
|
||||
if !info.Available {
|
||||
return info, fmt.Errorf("%s 未安装。安装方式: %s", name, installHint)
|
||||
}
|
||||
return info, nil
|
||||
}
|
||||
|
||||
// ============================
|
||||
// 环境探测
|
||||
// ============================
|
||||
|
||||
// SystemInfo 系统基础信息
|
||||
type SystemInfo struct {
|
||||
Hostname string
|
||||
CPUModel string
|
||||
CPUCores string
|
||||
MemTotal string
|
||||
KernelVer string
|
||||
DiskInfo string
|
||||
}
|
||||
|
||||
// ProbeSystem 探测远程系统基础信息
|
||||
func ProbeSystem(client *ssh.Client, executor Executor) SystemInfo {
|
||||
info := SystemInfo{}
|
||||
|
||||
out, _ := exec(client, executor, "hostname 2>/dev/null")
|
||||
info.Hostname = strings.TrimSpace(out)
|
||||
|
||||
out, _ = exec(client, executor, "lscpu 2>/dev/null | grep 'Model name' | sed 's/Model name:\\s*//'")
|
||||
info.CPUModel = strings.TrimSpace(out)
|
||||
|
||||
out, _ = exec(client, executor, "nproc 2>/dev/null")
|
||||
info.CPUCores = strings.TrimSpace(out)
|
||||
|
||||
out, _ = exec(client, executor, "free -h 2>/dev/null | awk '/Mem:/{print $2}'")
|
||||
info.MemTotal = strings.TrimSpace(out)
|
||||
|
||||
out, _ = exec(client, executor, "uname -r 2>/dev/null")
|
||||
info.KernelVer = strings.TrimSpace(out)
|
||||
|
||||
out, _ = exec(client, executor, "lsblk -d -o NAME,SIZE,TYPE 2>/dev/null | head -5")
|
||||
info.DiskInfo = strings.TrimSpace(out)
|
||||
|
||||
return info
|
||||
}
|
||||
|
||||
// ============================
|
||||
// 温度监控
|
||||
// ============================
|
||||
|
||||
// TempMonitor 温度监控器
|
||||
type TempMonitor struct {
|
||||
client *ssh.Client
|
||||
executor Executor
|
||||
interval time.Duration
|
||||
stopCh chan struct{}
|
||||
logs []string
|
||||
}
|
||||
|
||||
// NewTempMonitor 创建温度监控器
|
||||
func NewTempMonitor(client *ssh.Client, executor Executor, interval time.Duration) *TempMonitor {
|
||||
if interval <= 0 {
|
||||
interval = 10 * time.Second
|
||||
}
|
||||
return &TempMonitor{
|
||||
client: client,
|
||||
executor: executor,
|
||||
interval: interval,
|
||||
stopCh: make(chan struct{}),
|
||||
}
|
||||
}
|
||||
|
||||
// Start 后台启动温度监控
|
||||
func (m *TempMonitor) Start() {
|
||||
go m.loop()
|
||||
}
|
||||
|
||||
// Stop 停止监控并返回所有采样
|
||||
func (m *TempMonitor) Stop() []string {
|
||||
close(m.stopCh)
|
||||
return m.logs
|
||||
}
|
||||
|
||||
func (m *TempMonitor) loop() {
|
||||
ticker := time.NewTicker(m.interval)
|
||||
defer ticker.Stop()
|
||||
|
||||
// 初始温度
|
||||
m.sample()
|
||||
|
||||
for {
|
||||
select {
|
||||
case <-m.stopCh:
|
||||
return
|
||||
case <-ticker.C:
|
||||
m.sample()
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func (m *TempMonitor) sample() {
|
||||
// 尝试 sensors
|
||||
out, err := execf(m.client, m.executor, "sensors 2>/dev/null | grep -i 'temp\\|core\\|cpu' | head -10")
|
||||
if err == nil && strings.TrimSpace(out) != "" {
|
||||
ts := time.Now().Format("15:04:05")
|
||||
for _, line := range strings.Split(out, "\n") {
|
||||
line = strings.TrimSpace(line)
|
||||
if line != "" {
|
||||
m.logs = append(m.logs, fmt.Sprintf("[%s] %s", ts, line))
|
||||
}
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
// 回退: 读 sysfs
|
||||
out, _ = execf(m.client, m.executor, `cat /sys/class/thermal/thermal_zone*/temp 2>/dev/null | while read t; do echo "$((t/1000))°C"; done`)
|
||||
if strings.TrimSpace(out) != "" {
|
||||
ts := time.Now().Format("15:04:05")
|
||||
m.logs = append(m.logs, fmt.Sprintf("[%s] %s", ts, strings.TrimSpace(out)))
|
||||
}
|
||||
}
|
||||
|
||||
// ============================
|
||||
// dmesg 监控
|
||||
// ============================
|
||||
|
||||
// DmesgMonitor 内核日志监控
|
||||
type DmesgMonitor struct {
|
||||
client *ssh.Client
|
||||
executor Executor
|
||||
stopCh chan struct{}
|
||||
baseline string // 启动时的 dmesg 行数
|
||||
logs []string
|
||||
}
|
||||
|
||||
// NewDmesgMonitor 创建 dmesg 监控器
|
||||
func NewDmesgMonitor(client *ssh.Client, executor Executor) *DmesgMonitor {
|
||||
return &DmesgMonitor{
|
||||
client: client,
|
||||
executor: executor,
|
||||
stopCh: make(chan struct{}),
|
||||
}
|
||||
}
|
||||
|
||||
// Start 记录基线，后台监控
|
||||
func (m *DmesgMonitor) Start() {
|
||||
out, _ := exec(m.client, m.executor, "dmesg 2>/dev/null | wc -l")
|
||||
m.baseline = strings.TrimSpace(out)
|
||||
go m.loop()
|
||||
}
|
||||
|
||||
// Stop 停止监控，返回新增的硬件报错
|
||||
func (m *DmesgMonitor) Stop() []string {
|
||||
close(m.stopCh)
|
||||
|
||||
// 获取新增的 dmesg 日志中的硬件错误
|
||||
out, _ := execf(m.client, m.executor,
|
||||
`dmesg --level=err,crit,alert,emerg 2>/dev/null | tail -30`)
|
||||
if strings.TrimSpace(out) != "" {
|
||||
m.logs = append(m.logs, strings.Split(out, "\n")...)
|
||||
}
|
||||
|
||||
// 过滤硬件相关关键词
|
||||
var hwErrors []string
|
||||
keywords := []string{"error", "fail", "fault", "warn", "critical", "oom", "panic", "hardware", "thermal", "throttl"}
|
||||
for _, line := range m.logs {
|
||||
lower := strings.ToLower(line)
|
||||
for _, kw := range keywords {
|
||||
if strings.Contains(lower, kw) {
|
||||
hwErrors = append(hwErrors, strings.TrimSpace(line))
|
||||
break
|
||||
}
|
||||
}
|
||||
}
|
||||
return hwErrors
|
||||
}
|
||||
|
||||
func (m *DmesgMonitor) loop() {
|
||||
ticker := time.NewTicker(5 * time.Second)
|
||||
defer ticker.Stop()
|
||||
|
||||
for {
|
||||
select {
|
||||
case <-m.stopCh:
|
||||
return
|
||||
case <-ticker.C:
|
||||
// 周期性检查是否有新的硬件相关日志（轻量级）
|
||||
out, _ := execf(m.client, m.executor,
|
||||
`dmesg --level=err,crit 2>/dev/null | tail -3`)
|
||||
if strings.TrimSpace(out) != "" {
|
||||
ts := time.Now().Format("15:04:05")
|
||||
for _, line := range strings.Split(out, "\n") {
|
||||
line = strings.TrimSpace(line)
|
||||
if line != "" {
|
||||
m.logs = append(m.logs, fmt.Sprintf("[%s] %s", ts, line))
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
139
auto-check/pkg/stress/types.go
Normal file
139
auto-check/pkg/stress/types.go
Normal file
@@ -0,0 +1,139 @@
|
||||
package stress
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"strings"
|
||||
"time"
|
||||
)
|
||||
|
||||
// ============================
|
||||
// 测试类型定义
|
||||
// ============================
|
||||
|
||||
// TestType 压力测试类型
|
||||
type TestType string
|
||||
|
||||
const (
|
||||
// 单项压测
|
||||
TestCPU TestType = "cpu" // stress-ng CPU 压力
|
||||
TestMemory TestType = "memory" // stress-ng 内存压力
|
||||
TestDiskIO TestType = "disk" // stress-ng 磁盘IO压力
|
||||
TestMemNative TestType = "memnative" // stressapptest 内存精压
|
||||
|
||||
// 综合压测
|
||||
TestFull TestType = "full" // CPU+内存+磁盘IO 三合一
|
||||
|
||||
// 监控（自动附加）
|
||||
MonitorTemp TestType = "temp" // lm-sensors 温度监控
|
||||
MonitorDmesg TestType = "dmesg" // dmesg 内核报错监控
|
||||
)
|
||||
|
||||
// ============================
|
||||
// 配置
|
||||
// ============================
|
||||
|
||||
// Config 压力测试配置
|
||||
type Config struct {
|
||||
Duration time.Duration // 总持续时间
|
||||
Threads int // CPU/内存线程数
|
||||
MemSizeMB int // 内存测试大小 (MB)，0=自动(可用60%)
|
||||
DiskSizeMB int // 磁盘测试大小 (MB)，0=默认1024
|
||||
DiskDir string // 磁盘测试目录，空=自动临时目录
|
||||
TempLogInt time.Duration // 温度采样间隔，0=10s
|
||||
}
|
||||
|
||||
// DefaultConfig 默认配置
|
||||
func DefaultConfig() Config {
|
||||
return Config{
|
||||
Duration: 30 * time.Second,
|
||||
Threads: 4,
|
||||
MemSizeMB: 0, // 自动
|
||||
DiskSizeMB: 1024,
|
||||
DiskDir: "",
|
||||
TempLogInt: 10 * time.Second,
|
||||
}
|
||||
}
|
||||
|
||||
// ============================
|
||||
// 单项测试结果
|
||||
// ============================
|
||||
|
||||
// Result 单项测试结果
|
||||
type Result struct {
|
||||
Type TestType
|
||||
Status string // "pass" / "fail" / "skip" / "error"
|
||||
Output string // 工具输出摘要
|
||||
Error string // 错误信息
|
||||
Duration time.Duration
|
||||
}
|
||||
|
||||
// ============================
|
||||
// 完整测试报告
|
||||
// ============================
|
||||
|
||||
// Report 一次完整压测的报告
|
||||
type Report struct {
|
||||
IP string
|
||||
StartTime time.Time
|
||||
EndTime time.Time
|
||||
Duration time.Duration
|
||||
Results []Result
|
||||
Passed int
|
||||
Failed int
|
||||
Skipped int
|
||||
Errors int
|
||||
}
|
||||
|
||||
// AddResult 添加测试结果
|
||||
func (r *Report) AddResult(res Result) {
|
||||
r.Results = append(r.Results, res)
|
||||
switch res.Status {
|
||||
case "pass":
|
||||
r.Passed++
|
||||
case "fail":
|
||||
r.Failed++
|
||||
case "skip":
|
||||
r.Skipped++
|
||||
case "error":
|
||||
r.Errors++
|
||||
}
|
||||
}
|
||||
|
||||
// Summary 汇总信息
|
||||
func (r *Report) Summary() string {
|
||||
return fmt.Sprintf("共 %d 项: %d 通过, %d 失败, %d 跳过, %d 错误",
|
||||
len(r.Results), r.Passed, r.Failed, r.Skipped, r.Errors)
|
||||
}
|
||||
|
||||
// ToText 文本格式报告
|
||||
func (r *Report) ToText() string {
|
||||
var sb strings.Builder
|
||||
sb.WriteString(fmt.Sprintf("══ [%s] 压力测试报告 ══\n", r.IP))
|
||||
sb.WriteString(fmt.Sprintf(" 耗时: %s\n\n", r.Duration.Round(time.Millisecond)))
|
||||
|
||||
for _, res := range r.Results {
|
||||
icon := "✓"
|
||||
switch res.Status {
|
||||
case "fail":
|
||||
icon = "✗"
|
||||
case "skip":
|
||||
icon = "⊘"
|
||||
case "error":
|
||||
icon = "⚠"
|
||||
}
|
||||
sb.WriteString(fmt.Sprintf(" %s %-12s %s\n", icon, res.Type, res.Status))
|
||||
if res.Output != "" {
|
||||
for _, line := range strings.Split(res.Output, "\n") {
|
||||
if strings.TrimSpace(line) != "" {
|
||||
sb.WriteString(fmt.Sprintf(" %s\n", line))
|
||||
}
|
||||
}
|
||||
}
|
||||
if res.Error != "" {
|
||||
sb.WriteString(fmt.Sprintf(" 错误: %s\n", res.Error))
|
||||
}
|
||||
}
|
||||
|
||||
sb.WriteString(fmt.Sprintf("\n %s\n", r.Summary()))
|
||||
return sb.String()
|
||||
}
|
||||
Reference in New Issue
Block a user