dn42promsrv/dns.go

//////////////////////////////////////////////////////////////////////////
// DNS Metrics
//////////////////////////////////////////////////////////////////////////

package main

//////////////////////////////////////////////////////////////////////////

import (
	"encoding/json"
	"errors"
	//	"fmt"
	dns "github.com/miekg/dns"
	"github.com/prometheus/client_golang/prometheus"
	log "github.com/sirupsen/logrus"
	"io/ioutil"
	"math"
	"net/http"
	"strconv"
	"strings"
	"time"
)

//////////////////////////////////////////////////////////////////////////
// data structures

// for holding the DNS metrics

type DNSMetrics struct {
	soa       *prometheus.GaugeVec
	rtt       *prometheus.GaugeVec
	valid     *prometheus.GaugeVec
	stime     *prometheus.GaugeVec
	label_map []prometheus.Labels
}

// for specifying DNS servers

type DNSServer struct {
	role  string
	owner string
	name  string
	ip    uint8
	addr  string
	soa   uint64
}

// hardcoded :( list of DNS servers to query

var dns_servers = []*DNSServer{
	&DNSServer{"master", "jrb0001", "j.master.delegation-servers.dn42", 6, "[fd42:180:3de0:10:5054:ff:fe87:ea39]:53", 0},
	&DNSServer{"delegation", "burble", "b.delegation-servers.dn42", 4, "172.20.129.1:53", 0},
	&DNSServer{"delegation", "burble", "b.delegation-servers.dn42", 6, "[fd42:4242:2601:ac53::1]:53", 0},
	&DNSServer{"delegation", "jrb0001", "j.delegation-servers.dn42", 4, "172.20.1.18:53", 0},
	&DNSServer{"delegation", "jrb0001", "j.delegation-servers.dn42", 6, "[fd42:5d71:219:0:1::42]:53", 0},
	&DNSServer{"delegation", "yamakaja", "y.delegation-servers.dn42", 4, "172.20.20.66:53", 0},
	&DNSServer{"delegation", "yamakaja", "y.delegation-servers.dn42", 6, "[fd42:c01d:beef::3]:53", 0},
	&DNSServer{"recursive", "yamakaja", "a.recursive-servers.dn42", 4, "172.20.0.53:53", 0},
	&DNSServer{"recursive", "yamakaja", "a.recursive-servers.dn42", 6, "[fd42:d42:d42:54::1]:53", 0},
	&DNSServer{"recursive", "burble", "b.recursive-servers.dn42", 4, "172.20.129.2:53", 0},
	&DNSServer{"recursive", "burble", "b.recursive-servers.dn42", 6, "[fd42:4242:2601:ac53::53]:53", 0},
	&DNSServer{"recursive", "jrb0001", "j.recursive-servers.dn42", 4, "172.20.1.19:53", 0},
	&DNSServer{"recursive", "jrb0001", "j.recursive-servers.dn42", 6, "[fd42:5d71:219:0:1::43]:53", 0},
	&DNSServer{"recursive", "yamakaja", "y.recursive-servers.dn42", 4, "172.20.20.65:53", 0},
	&DNSServer{"recursive", "yamakaja", "y.recursive-servers.dn42", 6, "[fd42:c01d:beef::2]:53", 0},
	&DNSServer{"burble.dn42", "burble", "fr-rbx1", 6, "[fd42:4242:2601:36::ac:53]:53", 0},
	&DNSServer{"burble.dn42", "burble", "us-dal3", 6, "[fd42:4242:2601:2a::ac:53]:53", 0},
	&DNSServer{"burble.dn42", "burble", "sg-sin2", 6, "[fd42:4242:2601:37::ac:53]:53", 0},
	&DNSServer{"burble.dn42", "burble", "ca-bhs2", 6, "[fd42:4242:2601:2d::ac:53]:53", 0},
	&DNSServer{"burble.dn42", "burble", "lt-vil1", 6, "[fd42:4242:2601:3d::ac:53]:53", 0},
}

// data structures for querying the current commit metric

type DNSCommitMetrics struct {
	match        *prometheus.GaugeVec
	lastUpdate   time.Time
	mismatch     bool
	since        time.Time
	labelInvalid prometheus.Labels
	labelUpdate  prometheus.Labels
}

type DNSExplorerCommit struct {
	Commit string
}

type DNSMasterCommit struct {
	Type   string `json:"type"`
	URI    string `json:"uri"`
	Branch string `json:"branch"`
	Commit string `json:"commit"`
}

//////////////////////////////////////////////////////////////////////////
// initialisation function to register metrics

func (m *DNSMetrics) Register() {

	m.soa = prometheus.NewGaugeVec(prometheus.GaugeOpts{
		Name: "dn42_dns_soa",
		Help: "SOA for .dn42 domain",
	}, []string{"role", "owner", "name", "ip", "addr"})
	prometheus.MustRegister(m.soa)

	m.rtt = prometheus.NewGaugeVec(prometheus.GaugeOpts{
		Name: "dn42_dns_rtt",
		Help: "RTT when collecting SOA for .dn42 domain",
	}, []string{"role", "owner", "name", "ip", "addr"})
	prometheus.MustRegister(m.rtt)

	m.valid = prometheus.NewGaugeVec(prometheus.GaugeOpts{
		Name: "dn42_dns_valid",
		Help: "0 = response and correct serial, 1 = response but incorrect serial, 2 = no response",
	}, []string{"role", "owner", "name", "ip", "addr"})
	prometheus.MustRegister(m.valid)

	m.stime = prometheus.NewGaugeVec(prometheus.GaugeOpts{
		Name: "dn42_dns_stime",
		Help: "Returns the time between now and the SOA serial number",
	}, []string{"role", "owner", "name", "ip", "addr"})
	prometheus.MustRegister(m.stime)

	// pre-populate the labels
	m.label_map = make([]prometheus.Labels, len(dns_servers))

	for ix, server := range dns_servers {
		m.label_map[ix] = prometheus.Labels{
			"role":  server.role,
			"owner": server.owner,
			"name":  server.name,
			"ip":    strconv.Itoa(int(server.ip)),
			"addr":  server.addr,
		}
	}

}

//////////////////////////////////////////////////////////////////////////
// collect metrics for all DNS servers

func (m *DNSMetrics) Collect() {

	now := uint64(time.Now().Unix())

	// go through each server in turn
	for ix, server := range dns_servers {

		// query it
		soa, rtt := server.Query()
		server.soa = uint64(soa)

		// SOA and RTT are direct metrics returned from the query
		m.soa.With(m.label_map[ix]).Set(soa)
		m.rtt.With(m.label_map[ix]).Set(rtt)

		// check if the returned SOA matches j.master.delegation-servers.dn42
		var valid uint = 0

		if server.soa == 0 {
			// didn't get a result, server issue
			valid = 2
		} else {
			if server.soa != dns_servers[0].soa {
				// SOA didn't match
				valid = 1
			} else {
				valid = 0
			}
		}

		// before setting whether the server is valid, calculate the stime
		// (difference in time between now and the SOA, to allow checking that
		// it is not stale)
		// it's possible that the SOA could be in the future if there is a
		// clock mismatch between monitor and DNS server, in which case this
		// is flagged as a server error

		if server.soa > now {
			// server error
			valid = 2
		} else {
			m.stime.With(m.label_map[ix]).Set(float64(now - server.soa))
		}

		// finally set the valid status
		m.valid.With(m.label_map[ix]).Set(float64(valid))
	}

}

//////////////////////////////////////////////////////////////////////////

func (s *DNSServer) Query() (float64, float64) {

	msg := new(dns.Msg)
	msg.Id = dns.Id()
	msg.RecursionDesired = (s.role == "recursive")
	msg.Question = []dns.Question{{"dn42.", dns.TypeSOA, dns.ClassINET}}

	client := new(dns.Client)
	resp, rtt, err := client.Exchange(msg, s.addr)
	if err != nil || len(resp.Answer) != 1 {
		log.WithFields(log.Fields{
			"error":  err,
			"resp":   resp,
			"server": s,
		}).Warn("Unable to query DNS server")

		return 0, 0
	}

	if soa, ok := resp.Answer[0].(*dns.SOA); ok {
		return float64(soa.Serial), math.Round(rtt.Seconds() * 1000)
	}

	return 0, 0
}

//////////////////////////////////////////////////////////////////////////

func (m *DNSCommitMetrics) Register() {
	m.match = prometheus.NewGaugeVec(prometheus.GaugeOpts{
		Name: "dn42_dns_commit",
		Help: "Time since commits last matched (or -1 for error)",
	}, []string{"metric"})
	prometheus.MustRegister(m.match)

	m.labelInvalid = prometheus.Labels{
		"metric": "invalid",
	}

	m.labelUpdate = prometheus.Labels{
		"metric": "update",
	}

}

func (m *DNSCommitMetrics) Collect() {

	now := time.Now()
	interval := now.Sub(m.lastUpdate)

	// only check if it's more than 60 mins since the last successful match
	if (interval.Hours() >= 1.0) || m.mismatch {

		// fetch the current commit from the explorer
		ec, err := m.FetchExplorerCommit()
		if err != nil {
			// couldn't fetch the explorer commit value, set the metric to bad
			m.match.With(m.labelUpdate).Set(-1.0)
			m.match.With(m.labelInvalid).Set(-1.0)
			m.mismatch = true
			return
		}

		mc, err := m.FetchMasterCommit()
		if err != nil {
			// couldn't fetch master commit value, set the metric to bad
			m.match.With(m.labelUpdate).Set(-2.0)
			m.match.With(m.labelInvalid).Set(-1.0)
			m.mismatch = true
			return
		}

		m.lastUpdate = now
		m.match.With(m.labelUpdate).Set(0.0)

		if ec == mc {

			// if the commits match, null the metric
			m.match.With(m.labelInvalid).Set(0.0)
			m.mismatch = false

		} else {

			// was this the first mismatch ?
			if !m.mismatch {
				m.since = now
				m.mismatch = true
			}

			// set the metric to be the interval since last good match
			m.match.With(m.labelInvalid).Set(float64(now.Sub(m.since).Seconds()))
		}

	} else {

		// update time since last check
		m.match.With(m.labelUpdate).Set(float64(interval.Seconds()))
	}

}

//////////////////////////////////////////////////////////////////////////

func (m *DNSCommitMetrics) FetchExplorerCommit() (string, error) {

	response, err := http.Get("http://collector.dn42:8043/api/registry/.meta")
	if err != nil {
		log.WithFields(log.Fields{
			"error": err,
		}).Warn("Unable to query registry explorer")
		return "", err
	}

	data, err := ioutil.ReadAll(response.Body)
	if err != nil {
		log.WithFields(log.Fields{
			"error": err,
		}).Warn("Unable read explorer response")
		return "", err
	}

	var ec DNSExplorerCommit
	if err := json.Unmarshal(data, &ec); err != nil {
		log.WithFields(log.Fields{
			"error": err,
		}).Warn("Unable to parse explorer JSON")
		return "", err
	}

	return strings.ToLower(ec.Commit), nil

}

func (m *DNSCommitMetrics) FetchMasterCommit() (string, error) {

	response, err := http.Get("http://[2a0c:3800:1:1011:5054:ff:fe87:ea39]:8080/api/git-db-state")
	if err != nil {
		log.WithFields(log.Fields{
			"error": err,
		}).Warn("Unable to query registry master")
		return "", err
	}

	data, err := ioutil.ReadAll(response.Body)
	if err != nil {
		log.WithFields(log.Fields{
			"error": err,
		}).Warn("Unable read master response")
		return "", err
	}

	var mc []DNSMasterCommit
	if err := json.Unmarshal(data, &mc); err != nil {
		log.WithFields(log.Fields{
			"error": err,
		}).Warn("Unable to parse master JSON")
		return "", err
	}

	for _, c := range mc {
		if c.URI == "https://git.dn42.us/dn42/registry.git" {
			return strings.ToLower(c.Commit), nil
		}
	}

	log.WithFields(log.Fields{
		"MasterCommit": mc,
	}).Warn("Unable to find registry commit from master")
	return "", errors.New("Unable to find registry commit from master")
}

//////////////////////////////////////////////////////////////////////////
// end of code