Files
membank/auto-check/pkg/stress/runner.go
2026-08-13 02:09:39 +08:00

389 lines
9.7 KiB
Go
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
package stress
import (
"fmt"
"net"
"os"
"strconv"
"strings"
"time"
"auto-check/pkg/config"
"auto-check/pkg/discovery"
"auto-check/pkg/sshclient"
"golang.org/x/crypto/ssh"
)
// Results 测试结果表IP → 报告),包级公开
var Results = make(map[string]*Report)
// ============================
// 执行器
// ============================
// Runner 压力测试执行器
type Runner struct {
cfg Config
types []TestType
client *ssh.Client
tools ToolSet
metrics []Metric
}
// 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(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)
// 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()
// 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()
// 按需读取各测试脚本,拼装成完整脚本
script := r.assembleScript()
fmt.Printf(" [脚本] 拼装完成,%d 字节\n", len(script))
// 通过 heredoc 传到远程执行
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)
}
// 解析结果
if output != "" {
r.parseResults(report, output)
}
// 生成采样折线图
for _, res := range report.Results {
if len(res.Samples) > 0 {
chartPath, err := WriteChartHTML(ip, res, "reports")
if err != nil {
fmt.Printf(" [图表] 生成失败: %v\n", err)
} else {
fmt.Printf(" [图表] %s\n", chartPath)
}
}
}
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
}
// 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
var currentSamples []Sample
inSamples := false
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,
Samples: currentSamples,
})
currentTest = ""
currentStatus = ""
currentDuration = ""
currentOutput = nil
currentSamples = nil
}
for _, line := range lines {
line = strings.TrimSpace(line)
// 采样数据块
if line == "===SAMPLES===" {
inSamples = true
continue
}
if line == "===END_SAMPLES===" {
inSamples = false
continue
}
if inSamples {
parts := strings.Split(line, ",")
if len(parts) >= 3 {
cpu, _ := strconv.ParseFloat(strings.TrimSpace(parts[1]), 64)
temp, _ := strconv.ParseFloat(strings.TrimSpace(parts[2]), 64)
currentSamples = append(currentSamples, Sample{
Time: strings.TrimSpace(parts[0]),
CPU: cpu,
Temp: temp,
})
}
continue
}
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
}
}
}
// assembleScript 根据 metrics 读取对应的独立脚本文件,拼装成完整脚本
func (r *Runner) assembleScript() string {
var sb strings.Builder
// shebang + 环境变量导出
sb.WriteString("#!/bin/bash\n")
sb.WriteString("set -e\n\n")
// 导出参数变量
sb.WriteString(fmt.Sprintf("export AUTOCHECK_DURATION=%d\n", int(r.cfg.Duration.Seconds())))
sb.WriteString(fmt.Sprintf("export AUTOCHECK_THREADS=%d\n", r.cfg.Threads))
sb.WriteString(fmt.Sprintf("export AUTOCHECK_DISK_SIZE_MB=%d\n", r.cfg.DiskSizeMB))
sb.WriteString(fmt.Sprintf("export AUTOCHECK_MEM_SIZE_MB=%d\n", r.cfg.MemSizeMB))
sb.WriteString(fmt.Sprintf("export AUTOCHECK_TEMP_INTERVAL=%d\n", int(r.cfg.TempLogInt.Seconds())))
sb.WriteString("export AUTOCHECK_SAMPLE_INTERVAL=2\n")
if r.cfg.DiskDir != "" {
sb.WriteString(fmt.Sprintf("export AUTOCHECK_DISK_DIR=%s\n", r.cfg.DiskDir))
}
sb.WriteString("\n")
// 系统信息
sb.Write(r.readScript("sysinfo"))
sb.WriteString("\n")
// 温度监控启动
hasTemp := false
for _, m := range r.metrics {
if m.Name == "temp" && m.Enabled {
hasTemp = true
break
}
}
if hasTemp {
sb.Write(r.readScript("temp_start"))
sb.WriteString("\n")
}
// 逐个测试脚本
for _, m := range r.metrics {
if m.IsMonitor || !m.Enabled {
continue
}
data := r.readScript(m.Name)
if data == nil {
sb.WriteString(fmt.Sprintf("echo '===TEST:%s==='\n", m.Name))
sb.WriteString("echo 'status:skip'\n")
sb.WriteString(fmt.Sprintf("echo 'output:脚本文件 scripts/%s.sh 不存在'\n", m.Name))
sb.WriteString("echo ''\n")
continue
}
sb.Write(data)
sb.WriteString("\n")
}
// 温度监控收尾
if hasTemp {
sb.Write(r.readScript("temp_end"))
sb.WriteString("\n")
}
// dmesg + 结束标记
sb.Write(r.readScript("dmesg"))
return sb.String()
}
// readScript 读取 scripts/ 目录下的脚本文件
func (r *Runner) readScript(name string) []byte {
data, err := os.ReadFile(fmt.Sprintf("scripts/%s.sh", name))
if err != nil {
return nil
}
return data
}
// ============================
// 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 {
return NewSSHFailReport(ip, err)
}
defer conn.Close()
// SSH 登录成功后回填远端 MAC发现阶段只做了 ping暂无 MAC
if mac := queryRemoteMAC(conn); mac != "" {
if dev, ok := discovery.Devices[ip]; ok {
dev.MAC = mac
discovery.Devices[ip] = dev
}
fmt.Printf(" [MAC] %s -> %s\n", ip, mac)
}
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,
}
}
// queryRemoteMAC 查询远端主机的 MAC 地址(取第一个有效的单播地址)。
// 失败或无有效地址时返回空字符串,不阻断主流程。
func queryRemoteMAC(client *ssh.Client) string {
out, err := sshclient.RunCommand(client, "cat /sys/class/net/*/address 2>/dev/null")
if err != nil {
return ""
}
for _, line := range strings.Split(out, "\n") {
mac := strings.TrimSpace(line)
if mac == "" {
continue
}
hw, e := net.ParseMAC(mac)
if e != nil || len(hw) != 6 {
continue
}
// 排除零地址、广播、组播
if hw[0] == 0 && hw[1] == 0 && hw[2] == 0 && hw[3] == 0 && hw[4] == 0 && hw[5] == 0 {
continue
}
if hw[0]&0x01 != 0 {
continue
}
return hw.String()
}
return ""
}
// ============================
// 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
func Status(rpt *Report) string {
if rpt == nil {
return StatusUnknown
}
if IsPassed(rpt) {
return StatusPass
}
return StatusFail
}
// ParseTypes 逗号分隔字符串 → TestType 列表
func ParseTypes(s string) []TestType {
var types []TestType
for _, t := range strings.Split(s, ",") {
if t = strings.TrimSpace(t); t != "" {
types = append(types, TestType(t))
}
}
return types
}