// Package main 是 AIOps Agent 的入口。 // Agent 定时采集 CPU、内存、磁盘及监听端口指标,并通过 HTTP 上报至 AIOps 后端。 // // 以下变量可在编译时通过 -ldflags "-X main.version=... -X main.buildTime=..." 注入。 // // 署名 Logan.Li package main import ( "context" "flag" "log" "os" "os/signal" "syscall" "time" "ezops/agent/collector" "ezops/agent/model" "ezops/agent/reporter" ) // 编译时可注入的版本信息 var ( version = "dev" buildTime = "unknown" ) func main() { server := flag.String("server", "http://localhost:8080", "AIOps 后端上报地址(含 scheme 与端口)") interval := flag.Int("interval", 10, "采集与上报间隔,单位:秒") token := flag.String("token", "", "Bearer 鉴权令牌") hostnameFlag := flag.String("hostname", "", "上报主机名,默认使用系统主机名") flag.Parse() // 统一日志格式:带微秒时间戳,便于排查问题 logger := log.New(os.Stdout, "[aiops-agent] ", log.LstdFlags|log.Lmicroseconds) // 解析主机名:优先使用 -hostname,否则回退到 os.Hostname() hostname := *hostnameFlag if hostname == "" { if h, err := os.Hostname(); err == nil { hostname = h } else { hostname = "unknown" } } // 参数校验 if *server == "" { logger.Fatal("参数 -server 不能为空") } if *interval < 1 { logger.Fatalf("参数 -interval 必须 >= 1,当前为 %d", *interval) } logger.Printf("AIOps Agent 启动: version=%s buildTime=%s", version, buildTime) logger.Printf("配置: server=%s interval=%ds hostname=%q token=%s", *server, *interval, hostname, maskToken(*token)) // 初始化各采集器与上报器 cpuCollector := collector.NewCPUCollector() memCollector := collector.NewMemCollector() diskCollector := collector.NewDiskCollector() portCollector := collector.NewPortCollector() rep := reporter.New(*server, *token, logger) // 监听 SIGINT / SIGTERM,支持优雅退出 ctx, stop := signal.NotifyContext(context.Background(), os.Interrupt, syscall.SIGTERM) defer stop() ticker := time.NewTicker(time.Duration(*interval) * time.Second) defer ticker.Stop() // collectAndReport 执行单次采集并上报;各项采集独立进行,单项失败不影响整体上报 collectAndReport := func() { report := &model.Report{ Hostname: hostname, Timestamp: time.Now().UTC().Format(time.RFC3339), } if stat, err := cpuCollector.Collect(); err != nil { logger.Printf("CPU 采集异常: %v", err) } else { report.CPU = stat } if stat, err := memCollector.Collect(); err != nil { logger.Printf("内存采集异常: %v", err) } else { report.Memory = stat } if disks, err := diskCollector.Collect(); err != nil { logger.Printf("磁盘采集异常: %v", err) } else { report.Disks = disks } if ports, err := portCollector.Collect(); err != nil { logger.Printf("端口采集异常: %v", err) } else { report.Ports = ports } if err := rep.Send(ctx, report); err != nil { logger.Printf("上报失败: %v", err) return } logger.Printf("上报成功: cpu=%.2f%% mem=%.2f%% disks=%d ports=%d", report.CPU.UsagePercent, report.Memory.UsagePercent, len(report.Disks), len(report.Ports)) } // 启动后立即执行一次,再按 ticker 周期执行 collectAndReport() for { select { case <-ticker.C: collectAndReport() case <-ctx.Done(): logger.Println("收到退出信号,Agent 正在关闭...") return } } } // maskToken 对令牌做脱敏处理,仅用于日志输出 func maskToken(token string) string { if token == "" { return "(empty)" } if len(token) <= 4 { return "****" } return token[:4] + "****" }