yyfbaloot/servers/micros/micromonitor/main.go

package main

import (
	"context"
	"encoding/json"
	"fmt"
	"os"
	"os/exec"
	"time"

	"bet24.com/log"
	"bet24.com/servers/micros/common"
	"bet24.com/utils"
	"github.com/smallnest/rpcx/client"
)

var consulAddr = common.Default_Consul_Addr

func getClient(servcieName string) client.XClient {
	return common.GetClientPool().GetClient(servcieName, consulAddr)
}

type serviceInfo struct {
	Name      string
	Exe       string
	FailCount int
}

func (si *serviceInfo) getExecuteFile() string {
	if si.Exe == "" {
		return si.Name
	}
	return si.Exe
}

var services []serviceInfo

const max_failed_count = 2

func waitInput() {
	for {
		var endInput string
		fmt.Scanln(&endInput)
		switch endInput {
		case "list":
			dumpServices()
		default:
			log.Release("unknown command")
		}
	}
}

func dumpServices() {
	log.Release("listing services")
	for _, v := range services {
		log.Release("  %s:[%s][%d]", v.Name, v.Exe, v.FailCount)
	}
	log.Release("")
}

func tryRestart(exe string) {
	cmd := exec.Command("cmd.exe", "/c", fmt.Sprintf("start cmd.exe /k %v", exe))
	cmd.Start()
}

func loadServices() {
	data, err := os.ReadFile("serviceconf/monitor.json")
	if err != nil {
		log.Debug("read config failed %v", err)
		time.AfterFunc(time.Second*10, loadServices)
		return
	}

	err = json.Unmarshal(data, &services)
	if err != nil {
		log.Debug("Unmarshal config failed err:%v", err)
		time.AfterFunc(time.Second*10, loadServices)
		return
	}

	time.AfterFunc(time.Second*600, loadServices)
}

type Request struct {
	Name string
}

type Response struct {
	Data string
}

func sayHelloToService(si *serviceInfo) {
	xclient := getClient(si.Name)

	args := &Request{
		Name: si.Name,
	}

	reply := &Response{}

	err := xclient.Call(context.Background(), "SayHello", args, reply)
	if err != nil {
		log.Debug("failed to call service[%s]: %v", si.Name, err)
		common.GetClientPool().RemoveClient(si.Name)
		// 重启服务
		//exe := addFailCount(serviceName)
		si.FailCount++
		if si.FailCount >= max_failed_count {
			log.Release("service[%s] dead,try to restart", si.Name)
			si.FailCount = 0
			tryRestart(si.getExecuteFile())
		}
	} else {
		si.FailCount = 0
	}
}

func checkAllServices() {
	time.AfterFunc(time.Second*10, checkAllServices)

	for k := range services {
		sayHelloToService(&services[k])
	}
}

func main() {
	defer waitInput()
	// 读取配置
	loadServices()
	utils.SetConsoleTitle(fmt.Sprintf("MicroMonitor0 %d services", len(services)))
	checkAllServices()
}