esxi-monitor/cmd/watchdog/main.go

package main

import (
	"context"
	"fmt"
	"os"
	"time"

	"github.com/go-ping/ping"
	_ "github.com/go-ping/ping"

	"github.com/tursom/esxi-monitor/vmomi"
)

func main() {
	cfg, err := parseConfig("config.yaml")
	if err != nil {
		panic(err)
	}

	watchTargets := make(map[string]*watchVm)
	for _, watch := range cfg.Watch {
		watchTargets[watch.Name] = watch
	}

	client, err := vmomi.Connect("https://esxi/sdk", cfg.User, cfg.Password)
	if err != nil {
		panic(err)
	}

	vms, err := vmomi.ListVms(client.Client)
	if err != nil {
		panic(err)
	}

	// ip -> vm instance
	watches := make(map[string]*vmomi.VirtualMachine)
	for _, vm := range vms {
		target, contains := watchTargets[vm.Name()]
		if !contains {
			continue
		}

		fmt.Printf("watching virtual machine: addr=%s, name=%s\n", target.Addr, vm.Name())
		watches[target.Addr] = vm
	}

	doWatch(watches)
}

// watches map ip -> vm instance
func doWatch(watches map[string]*vmomi.VirtualMachine) {
	t := time.NewTicker(time.Minute)
	for {
		<-t.C

		for ip, vm := range watches {
			pinger, err := ping.NewPinger(ip)
			if err != nil {
				_, _ = fmt.Fprintf(os.Stderr, "ping to target failed: %s\n", err)

				continue
			}

			pinger.Count = 4

			err = pinger.Run()
			if err != nil {
				_, _ = fmt.Fprintf(os.Stderr, "ping to target failed: %s\n", err)
				continue
			}

			statistics := pinger.Statistics()
			if statistics.PacketsRecv == 0 {
				fmt.Printf("restarting vm %s(%s)", ip, vm.Name())

				task, err := vm.Reset()
				if err != nil {
					_, _ = fmt.Fprintf(os.Stderr, "restart vm failed: %s\n", err)
					continue
				}

				ctx, _ := context.WithTimeout(context.Background(), time.Minute)
				err = task.Wait(ctx)
				if err != nil {
					_, _ = fmt.Fprintf(os.Stderr, "restart vm failed: %s\n", err)
					continue
				}
			}
		}
	}
}