virtigo/poll.go

package main

import (
	"fmt"
	"strings"
	"time"

	"go.wit.com/lib/gui/shell"
	"go.wit.com/log"
)

func (h *HyperT) pollHypervisor() {
	url := "http://" + h.pb.Hostname + ":2520/vms"
	log.Log(POLL, "wget url =", url)
	s := shell.Wget(url)
	if s == nil {
		return
	}
	var bytesSplice []byte
	bytesSplice = s.Bytes()
	// fmt.Fprintln(w, string(bytesSplice))
	for _, line := range strings.Split(string(bytesSplice), "\n") {
		if line == "" {
			continue
		}
		fields := strings.Fields(line)
		if len(fields) < 2 {
			continue
		}
		state := fields[0]
		name := fields[1]
		if state == "ON" {
			log.Log(POLL, h.pb.Hostname, "STATE:", state, "HOST:", name, "rest:", fields[2:])
			d := findDroplet(name)
			if d == nil {
				// this is a new unknown droplet (not in the config file)
				d = new(DropletT)
				d.pb.Hostname = name
				d.h = h
				d.lastpoll = time.Now()
				d.CurrentState = "ON"
				me.droplets = append(me.droplets, d)
				log.Log(EVENT, name, "IS NEW. ADDED ON", h.pb.Hostname)
			}
			log.Log(INFO, "ALREADY RECORDED", d.pb.Hostname)

			// update the status to ON and the last polled value
			d.CurrentState = "ON"
			d.lastpoll = time.Now()

			if d.h == nil {
				// this means the droplet was in the config file
				// but this is the first time it's shown up as running

				// this should mean a droplet is running where the config file says it probably should be running
				if d.pb.PreferredHypervisor == h.pb.Hostname {
					log.Log(EVENT, "new droplet", d.pb.Hostname, "(matches config hypervisor", h.pb.Hostname+")")
					d.h = h
					continue
				}

				log.Log(EVENT, "new droplet", d.pb.Hostname, "on", h.pb.Hostname, "(in config file without preferred hypervisor)")
				d.h = h
				continue
			}

			// this means the droplet is still where it was before
			if d.h.pb.Hostname != h.pb.Hostname {
				log.Log(EVENT, "droplet", d.h.pb.Hostname, "moved to", h.pb.Hostname)
				continue
			}
			d.h = h
		}
		continue
	}
	h.lastpoll = time.Now()
	h.killcount = 0 // poll worked. reset killcount
}

func findDroplet(name string) *DropletT {
	for _, d := range me.droplets {
		if d.pb.Hostname == name {
			return d
		}
	}
	return nil
}

func findHypervisor(name string) *HyperT {
	for _, h := range me.hypers {
		if h.pb.Hostname == name {
			return h
		}
	}
	return nil
}

// check the state of the cluster and return a string
// that is intended to be sent to an uptime monitor like Kuma
func clusterHealthy() (bool, string) {
	var good bool = true
	var total int
	var working int
	var failed int
	var missing int
	var unknown int
	var unknownList []string

	for _, d := range me.droplets {
		total += 1
		if d.pb.StartState != "ON" {
			continue
		}
		dur := time.Since(d.lastpoll) // Calculate the elapsed time
		if d.CurrentState == "" {
			// log.Info("SKIP. hostname has not been polled yet", d.pb.Hostname, d.hname)
			unknown += 1
			unknownList = append(unknownList, d.pb.Hostname)
			continue
		}
		var hname string
		if d.h != nil {
			hname = d.h.pb.Hostname
		}
		if d.CurrentState != "ON" {
			log.Info("BAD  STATE", d.pb.StartState, d.pb.Hostname, hname, "CurrentState =", d.CurrentState, shell.FormatDuration(dur))
			good = false
			failed += 1
		} else {
			dur := time.Since(d.lastpoll) // Calculate the elapsed time
			if dur > time.Minute {
				log.Info("GOOD STATE MISSING", d.pb.Hostname, hname, shell.FormatDuration(dur))
				good = false
				d.CurrentState = "MISSING"
				failed += 1
				continue
			}
			l := shell.FormatDuration(dur)
			if l == "" {
				log.Info("DUR IS EMPTY", dur)
				missing += 1
				continue
			}
			working += 1
			// log.Info("GOOD STATE ON", d.pb.Hostname, d.hname, "dur =", l)
		}
	}
	var summary string = "("
	summary += fmt.Sprintf("total = %d ", total)
	summary += fmt.Sprintf("working = %d ", working)
	if missing > 0 {
		summary += fmt.Sprintf("missing = %d ", missing)
	}
	if unknown > 0 {
		summary += fmt.Sprintf("unknown = %d ", unknown, unknownList)
	}
	if failed > 0 {
		summary += fmt.Sprintf("failed = %d ", failed)
	}
	summary = strings.TrimSpace(summary)
	summary += ")"
	if me.killcount > 0 {
		summary += "(killcount=" + fmt.Sprintf("%d", me.killcount) + ")"
	}
	last := time.Since(me.unstable)
	if last > 133*time.Second {
		// the cluster has not been stable for 10 seconds
		s := strings.TrimSpace(shell.FormatDuration(last))
		summary += "(stable=" + s + ")"
	}
	if good {
		return good, "GOOD=true " + summary
	}
	me.unstable = time.Now()
	return good, "GOOD=false " + summary
}