mirror of
https://github.com/prometheus/node_exporter.git
synced 2024-11-23 20:36:21 +01:00
Refactor node_exporter to support collectors.
A collector is a type matching 'Collector' interface. The following collectors where added: - NativeCollector wrapping the original functionality (attributes, load) - GmondCollector scraping ganglia's gmond (based on gmond_exporter) - MuninCollector scraping munin (based on munin_exporter)
This commit is contained in:
Johannes 'fish' Ziemke
parent
a6e8bcb1c4
commit
588ef8b62a
177
exporter/exporter.go
Normal file
177
exporter/exporter.go
Normal file
@ -0,0 +1,177 @@
|
||||
// Exporter is a prometheus exporter using multiple collectors to collect and export system metrics.
|
||||
package exporter
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"flag"
|
||||
"fmt"
|
||||
"github.com/prometheus/client_golang/prometheus"
|
||||
"github.com/prometheus/client_golang/prometheus/exp"
|
||||
"io/ioutil"
|
||||
"log"
|
||||
"net/http"
|
||||
"os"
|
||||
"os/signal"
|
||||
"runtime/pprof"
|
||||
"sync"
|
||||
"syscall"
|
||||
"time"
|
||||
)
|
||||
|
||||
var verbose = flag.Bool("verbose", false, "Verbose output.")
|
||||
|
||||
// Interface a collector has to implement.
|
||||
type Collector interface {
|
||||
// Get new metrics and expose them via prometheus registry.
|
||||
Update() (n int, err error)
|
||||
|
||||
// Returns the name of the collector
|
||||
Name() string
|
||||
}
|
||||
|
||||
type config struct {
|
||||
Attributes map[string]string `json:"attributes"`
|
||||
ListeningAddress string `json:"listeningAddress"`
|
||||
ScrapeInterval int `json:"scrapeInterval"`
|
||||
Collectors []string `json:"collectors"`
|
||||
}
|
||||
|
||||
func (e *exporter) loadConfig() (err error) {
|
||||
log.Printf("Reading config %s", e.configFile)
|
||||
bytes, err := ioutil.ReadFile(e.configFile)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
|
||||
return json.Unmarshal(bytes, &e.config) // Make sure this is safe
|
||||
}
|
||||
|
||||
type exporter struct {
|
||||
configFile string
|
||||
listeningAddress string
|
||||
scrapeInterval time.Duration
|
||||
scrapeDurations prometheus.Histogram
|
||||
metricsUpdated prometheus.Gauge
|
||||
config config
|
||||
registry prometheus.Registry
|
||||
collectors []Collector
|
||||
MemProfile string
|
||||
}
|
||||
|
||||
// New takes the path to a config file and returns an exporter instance
|
||||
func New(configFile string) (e exporter, err error) {
|
||||
registry := prometheus.NewRegistry()
|
||||
e = exporter{
|
||||
configFile: configFile,
|
||||
scrapeDurations: prometheus.NewDefaultHistogram(),
|
||||
metricsUpdated: prometheus.NewGauge(),
|
||||
listeningAddress: ":8080",
|
||||
scrapeInterval: 60 * time.Second,
|
||||
registry: registry,
|
||||
}
|
||||
|
||||
err = e.loadConfig()
|
||||
if err != nil {
|
||||
return e, fmt.Errorf("Couldn't read config: %s", err)
|
||||
}
|
||||
|
||||
cn, err := NewNativeCollector(e.config, e.registry)
|
||||
if err != nil {
|
||||
log.Fatalf("Couldn't attach collector: %s", err)
|
||||
}
|
||||
|
||||
cg, err := NewGmondCollector(e.config, e.registry)
|
||||
if err != nil {
|
||||
log.Fatalf("Couldn't attach collector: %s", err)
|
||||
}
|
||||
|
||||
cm, err := NewMuninCollector(e.config, e.registry)
|
||||
if err != nil {
|
||||
log.Fatalf("Couldn't attach collector: %s", err)
|
||||
}
|
||||
|
||||
e.collectors = []Collector{&cn, &cg, &cm}
|
||||
|
||||
if e.config.ListeningAddress != "" {
|
||||
e.listeningAddress = e.config.ListeningAddress
|
||||
}
|
||||
if e.config.ScrapeInterval != 0 {
|
||||
e.scrapeInterval = time.Duration(e.config.ScrapeInterval) * time.Second
|
||||
}
|
||||
|
||||
registry.Register("node_exporter_scrape_duration_seconds", "node_exporter: Duration of a scrape job.", prometheus.NilLabels, e.scrapeDurations)
|
||||
registry.Register("node_exporter_metrics_updated", "node_exporter: Number of metrics updated.", prometheus.NilLabels, e.metricsUpdated)
|
||||
|
||||
return e, nil
|
||||
}
|
||||
|
||||
func (e *exporter) serveStatus() {
|
||||
exp.Handle(prometheus.ExpositionResource, e.registry.Handler())
|
||||
http.ListenAndServe(e.listeningAddress, exp.DefaultCoarseMux)
|
||||
}
|
||||
|
||||
func (e *exporter) Execute(c Collector) {
|
||||
begin := time.Now()
|
||||
updates, err := c.Update()
|
||||
duration := time.Since(begin)
|
||||
|
||||
label := map[string]string{
|
||||
"collector": c.Name(),
|
||||
}
|
||||
if err != nil {
|
||||
log.Printf("ERROR: %s failed after %fs: %s", c.Name(), duration.Seconds(), err)
|
||||
label["result"] = "error"
|
||||
} else {
|
||||
log.Printf("OK: %s success after %fs.", c.Name(), duration.Seconds())
|
||||
label["result"] = "success"
|
||||
}
|
||||
e.scrapeDurations.Add(label, duration.Seconds())
|
||||
e.metricsUpdated.Set(label, float64(updates))
|
||||
}
|
||||
|
||||
func (e *exporter) Loop() {
|
||||
sigHup := make(chan os.Signal)
|
||||
sigUsr1 := make(chan os.Signal)
|
||||
signal.Notify(sigHup, syscall.SIGHUP)
|
||||
signal.Notify(sigUsr1, syscall.SIGUSR1)
|
||||
|
||||
go e.serveStatus()
|
||||
|
||||
tick := time.Tick(e.scrapeInterval)
|
||||
for {
|
||||
select {
|
||||
case <-sigHup:
|
||||
err := e.loadConfig()
|
||||
if err != nil {
|
||||
log.Printf("Couldn't reload config: %s", err)
|
||||
continue
|
||||
}
|
||||
log.Printf("Got new config")
|
||||
tick = time.Tick(e.scrapeInterval)
|
||||
|
||||
case <-tick:
|
||||
log.Printf("Starting new scrape interval")
|
||||
wg := sync.WaitGroup{}
|
||||
wg.Add(len(e.collectors))
|
||||
for _, c := range e.collectors {
|
||||
go func(c Collector) {
|
||||
e.Execute(c)
|
||||
wg.Done()
|
||||
}(c)
|
||||
}
|
||||
wg.Wait()
|
||||
|
||||
case <-sigUsr1:
|
||||
log.Printf("got signal")
|
||||
if e.MemProfile != "" {
|
||||
log.Printf("Writing memory profile to %s", e.MemProfile)
|
||||
f, err := os.Create(e.MemProfile)
|
||||
if err != nil {
|
||||
log.Fatal(err)
|
||||
}
|
||||
pprof.WriteHeapProfile(f)
|
||||
f.Close()
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
61
exporter/ganglia/format.go
Normal file
61
exporter/ganglia/format.go
Normal file
@ -0,0 +1,61 @@
|
||||
// Types for unmarshalling gmond's XML output.
|
||||
//
|
||||
// Not used elements in gmond's XML output are commented.
|
||||
// In case you want to use them, please change the names so that one
|
||||
// can understand without needing to know what the acronym stands for.
|
||||
package ganglia
|
||||
|
||||
import "encoding/xml"
|
||||
|
||||
type ExtraElement struct {
|
||||
Name string `xml:"NAME,attr"`
|
||||
Val string `xml:"VAL,attr"`
|
||||
}
|
||||
|
||||
type ExtraData struct {
|
||||
ExtraElements []ExtraElement `xml:"EXTRA_ELEMENT"`
|
||||
}
|
||||
|
||||
type Metric struct {
|
||||
Name string `xml:"NAME,attr"`
|
||||
Value float64 `xml:"VAL,attr"`
|
||||
/*
|
||||
Unit string `xml:"UNITS,attr"`
|
||||
Slope string `xml:"SLOPE,attr"`
|
||||
Tn int `xml:"TN,attr"`
|
||||
Tmax int `xml:"TMAX,attr"`
|
||||
Dmax int `xml:"DMAX,attr"`
|
||||
*/
|
||||
ExtraData ExtraData `xml:"EXTRA_DATA"`
|
||||
}
|
||||
|
||||
type Host struct {
|
||||
Name string `xml:"NAME,attr"`
|
||||
/*
|
||||
Ip string `xml:"IP,attr"`
|
||||
Tags string `xml:"TAGS,attr"`
|
||||
Reported int `xml:"REPORTED,attr"`
|
||||
Tn int `xml:"TN,attr"`
|
||||
Tmax int `xml:"TMAX,attr"`
|
||||
Dmax int `xml:"DMAX,attr"`
|
||||
Location string `xml:"LOCATION,attr"`
|
||||
GmondStarted int `xml:"GMOND_STARTED",attr"`
|
||||
*/
|
||||
Metrics []Metric `xml:"METRIC"`
|
||||
}
|
||||
|
||||
type Cluster struct {
|
||||
Name string `xml:"NAME,attr"`
|
||||
/*
|
||||
Owner string `xml:"OWNER,attr"`
|
||||
LatLong string `xml:"LATLONG,attr"`
|
||||
Url string `xml:"URL,attr"`
|
||||
Localtime int `xml:"LOCALTIME,attr"`
|
||||
*/
|
||||
Hosts []Host `xml:"HOST"`
|
||||
}
|
||||
|
||||
type Ganglia struct {
|
||||
XMLNAME xml.Name `xml:"GANGLIA_XML"`
|
||||
Clusters []Cluster `xml:"CLUSTER"`
|
||||
}
|
||||
103
exporter/gmond_collector.go
Normal file
103
exporter/gmond_collector.go
Normal file
@ -0,0 +1,103 @@
|
||||
package exporter
|
||||
|
||||
import (
|
||||
"bufio"
|
||||
"encoding/xml"
|
||||
"fmt"
|
||||
"github.com/prometheus/client_golang/prometheus"
|
||||
"github.com/prometheus/node_exporter/exporter/ganglia"
|
||||
"io"
|
||||
"net"
|
||||
"time"
|
||||
"strings"
|
||||
)
|
||||
|
||||
const (
|
||||
gangliaAddress = "127.0.0.1:8649"
|
||||
gangliaProto = "tcp"
|
||||
gangliaTimeout = 30 * time.Second
|
||||
)
|
||||
|
||||
type gmondCollector struct {
|
||||
name string
|
||||
Metrics map[string]prometheus.Gauge
|
||||
config config
|
||||
registry prometheus.Registry
|
||||
}
|
||||
|
||||
// Takes a config struct and prometheus registry and returns a new Collector scraping ganglia.
|
||||
func NewGmondCollector(config config, registry prometheus.Registry) (collector gmondCollector, err error) {
|
||||
collector = gmondCollector{
|
||||
name: "gmond_collector",
|
||||
config: config,
|
||||
Metrics: make(map[string]prometheus.Gauge),
|
||||
registry: registry,
|
||||
}
|
||||
|
||||
return collector, nil
|
||||
}
|
||||
|
||||
func (c *gmondCollector) Name() string { return c.name }
|
||||
|
||||
func (c *gmondCollector) setMetric(name string, labels map[string]string, metric ganglia.Metric) {
|
||||
if _, ok := c.Metrics[name]; !ok {
|
||||
var desc string
|
||||
var title string
|
||||
for _, element := range metric.ExtraData.ExtraElements {
|
||||
switch element.Name {
|
||||
case "DESC":
|
||||
desc = element.Val
|
||||
case "TITLE":
|
||||
title = element.Val
|
||||
}
|
||||
if title != "" && desc != "" {
|
||||
break
|
||||
}
|
||||
}
|
||||
debug(c.Name(), "Register %s: %s", name, desc)
|
||||
gauge := prometheus.NewGauge()
|
||||
c.Metrics[name] = gauge
|
||||
c.registry.Register(name, desc, prometheus.NilLabels, gauge) // one gauge per metric!
|
||||
}
|
||||
debug(c.Name(), "Set %s{%s}: %f", name, labels, metric.Value)
|
||||
c.Metrics[name].Set(labels, metric.Value)
|
||||
}
|
||||
|
||||
func (c *gmondCollector) Update() (updates int, err error) {
|
||||
conn, err := net.Dial(gangliaProto, gangliaAddress)
|
||||
debug(c.Name(), "gmondCollector Update")
|
||||
if err != nil {
|
||||
return updates, fmt.Errorf("Can't connect to gmond: %s", err)
|
||||
}
|
||||
conn.SetDeadline(time.Now().Add(gangliaTimeout))
|
||||
|
||||
ganglia := ganglia.Ganglia{}
|
||||
decoder := xml.NewDecoder(bufio.NewReader(conn))
|
||||
decoder.CharsetReader = toUtf8
|
||||
|
||||
err = decoder.Decode(&ganglia)
|
||||
if err != nil {
|
||||
return updates, fmt.Errorf("Couldn't parse xml: %s", err)
|
||||
}
|
||||
|
||||
for _, cluster := range ganglia.Clusters {
|
||||
for _, host := range cluster.Hosts {
|
||||
|
||||
for _, metric := range host.Metrics {
|
||||
name := strings.ToLower(metric.Name)
|
||||
|
||||
var labels = map[string]string{
|
||||
"hostname": host.Name,
|
||||
"cluster": cluster.Name,
|
||||
}
|
||||
c.setMetric(name, labels, metric)
|
||||
updates++
|
||||
}
|
||||
}
|
||||
}
|
||||
return updates, err
|
||||
}
|
||||
|
||||
func toUtf8(charset string, input io.Reader) (io.Reader, error) {
|
||||
return input, nil //FIXME
|
||||
}
|
||||
26
exporter/helper.go
Normal file
26
exporter/helper.go
Normal file
@ -0,0 +1,26 @@
|
||||
package exporter
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"log"
|
||||
"strconv"
|
||||
"strings"
|
||||
)
|
||||
|
||||
func debug(name string, format string, a ...interface{}) {
|
||||
if *verbose {
|
||||
f := fmt.Sprintf("%s: %s", name, format)
|
||||
log.Printf(f, a...)
|
||||
}
|
||||
}
|
||||
|
||||
func splitToInts(str string, sep string) (ints []int, err error) {
|
||||
for _, part := range strings.Split(str, sep) {
|
||||
i, err := strconv.Atoi(part)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("Could not split '%s' because %s is no int: %s", str, part, err)
|
||||
}
|
||||
ints = append(ints, i)
|
||||
}
|
||||
return ints, nil
|
||||
}
|
||||
233
exporter/munin_collector.go
Normal file
233
exporter/munin_collector.go
Normal file
@ -0,0 +1,233 @@
|
||||
package exporter
|
||||
|
||||
import (
|
||||
"bufio"
|
||||
"fmt"
|
||||
"github.com/prometheus/client_golang/prometheus"
|
||||
"io"
|
||||
"net"
|
||||
"regexp"
|
||||
"strconv"
|
||||
"strings"
|
||||
)
|
||||
|
||||
const (
|
||||
muninAddress = "127.0.0.1:4949"
|
||||
muninProto = "tcp"
|
||||
)
|
||||
|
||||
var muninBanner = regexp.MustCompile(`# munin node at (.*)`)
|
||||
|
||||
type muninCollector struct {
|
||||
name string
|
||||
hostname string
|
||||
graphs []string
|
||||
gaugePerMetric map[string]prometheus.Gauge
|
||||
config config
|
||||
registry prometheus.Registry
|
||||
connection net.Conn
|
||||
}
|
||||
|
||||
// Takes a config struct and prometheus registry and returns a new Collector scraping munin.
|
||||
func NewMuninCollector(config config, registry prometheus.Registry) (c muninCollector, err error) {
|
||||
c = muninCollector{
|
||||
name: "munin_collector",
|
||||
config: config,
|
||||
registry: registry,
|
||||
gaugePerMetric: make(map[string]prometheus.Gauge),
|
||||
}
|
||||
|
||||
return c, err
|
||||
}
|
||||
|
||||
func (c *muninCollector) Name() string { return c.name }
|
||||
|
||||
func (c *muninCollector) connect() (err error) {
|
||||
c.connection, err = net.Dial(muninProto, muninAddress)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
debug(c.Name(), "Connected.")
|
||||
|
||||
reader := bufio.NewReader(c.connection)
|
||||
head, err := reader.ReadString('\n')
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
matches := muninBanner.FindStringSubmatch(head)
|
||||
if len(matches) != 2 { // expect: # munin node at <hostname>
|
||||
return fmt.Errorf("Unexpected line: %s", head)
|
||||
}
|
||||
c.hostname = matches[1]
|
||||
debug(c.Name(), "Found hostname: %s", c.hostname)
|
||||
return err
|
||||
}
|
||||
|
||||
func (c *muninCollector) muninCommand(cmd string) (reader *bufio.Reader, err error) {
|
||||
if c.connection == nil {
|
||||
err := c.connect()
|
||||
if err != nil {
|
||||
return reader, fmt.Errorf("Couldn't connect to munin: %s", err)
|
||||
}
|
||||
}
|
||||
reader = bufio.NewReader(c.connection)
|
||||
|
||||
fmt.Fprintf(c.connection, cmd+"\n")
|
||||
|
||||
_, err = reader.Peek(1)
|
||||
switch err {
|
||||
case io.EOF:
|
||||
debug(c.Name(), "not connected anymore, closing connection and reconnect.")
|
||||
c.connection.Close()
|
||||
err = c.connect()
|
||||
if err != nil {
|
||||
return reader, fmt.Errorf("Couldn't connect to %s: %s", muninAddress)
|
||||
}
|
||||
return c.muninCommand(cmd)
|
||||
case nil: //no error
|
||||
break
|
||||
default:
|
||||
return reader, fmt.Errorf("Unexpected error: %s", err)
|
||||
}
|
||||
|
||||
return reader, err
|
||||
}
|
||||
|
||||
func (c *muninCollector) muninList() (items []string, err error) {
|
||||
munin, err := c.muninCommand("list")
|
||||
if err != nil {
|
||||
return items, fmt.Errorf("Couldn't get list: %s", err)
|
||||
}
|
||||
|
||||
response, err := munin.ReadString('\n') // we are only interested in the first line
|
||||
if err != nil {
|
||||
return items, fmt.Errorf("Couldn't read response: %s", err)
|
||||
}
|
||||
|
||||
if response[0] == '#' { // # not expected here
|
||||
return items, fmt.Errorf("Error getting items: %s", response)
|
||||
}
|
||||
items = strings.Fields(strings.TrimRight(response, "\n"))
|
||||
return items, err
|
||||
}
|
||||
|
||||
func (c *muninCollector) muninConfig(name string) (config map[string]map[string]string, graphConfig map[string]string, err error) {
|
||||
graphConfig = make(map[string]string)
|
||||
config = make(map[string]map[string]string)
|
||||
|
||||
resp, err := c.muninCommand("config " + name)
|
||||
if err != nil {
|
||||
return config, graphConfig, fmt.Errorf("Couldn't get config for %s: %s", name, err)
|
||||
}
|
||||
|
||||
for {
|
||||
line, err := resp.ReadString('\n')
|
||||
if err == io.EOF {
|
||||
debug(c.Name(), "EOF, retrying")
|
||||
return c.muninConfig(name)
|
||||
}
|
||||
if err != nil {
|
||||
return nil, nil, err
|
||||
}
|
||||
if line == ".\n" { // munin end marker
|
||||
break
|
||||
}
|
||||
if line[0] == '#' { // here it's just a comment, so ignore it
|
||||
continue
|
||||
}
|
||||
parts := strings.Fields(line)
|
||||
if len(parts) < 2 {
|
||||
return nil, nil, fmt.Errorf("Line unexpected: %s", line)
|
||||
}
|
||||
key, value := parts[0], strings.TrimRight(strings.Join(parts[1:], " "), "\n")
|
||||
|
||||
key_parts := strings.Split(key, ".")
|
||||
if len(key_parts) > 1 { // it's a metric config (metric.label etc)
|
||||
if _, ok := config[key_parts[0]]; !ok {
|
||||
config[key_parts[0]] = make(map[string]string)
|
||||
}
|
||||
config[key_parts[0]][key_parts[1]] = value
|
||||
} else {
|
||||
graphConfig[key_parts[0]] = value
|
||||
}
|
||||
}
|
||||
return config, graphConfig, err
|
||||
}
|
||||
|
||||
func (c *muninCollector) registerMetrics() (err error) {
|
||||
items, err := c.muninList()
|
||||
if err != nil {
|
||||
return fmt.Errorf("Couldn't get graph list: %s", err)
|
||||
}
|
||||
|
||||
for _, name := range items {
|
||||
c.graphs = append(c.graphs, name)
|
||||
configs, graphConfig, err := c.muninConfig(name)
|
||||
if err != nil {
|
||||
return fmt.Errorf("Couldn't get config for graph %s: %s", name, err)
|
||||
}
|
||||
|
||||
for metric, config := range configs {
|
||||
metricName := name + "-" + metric
|
||||
desc := graphConfig["graph_title"] + ": " + config["label"]
|
||||
if config["info"] != "" {
|
||||
desc = desc + ", " + config["info"]
|
||||
}
|
||||
gauge := prometheus.NewGauge()
|
||||
debug(c.Name(), "Register %s: %s", metricName, desc)
|
||||
c.gaugePerMetric[metricName] = gauge
|
||||
c.registry.Register(metricName, desc, prometheus.NilLabels, gauge)
|
||||
}
|
||||
}
|
||||
return err
|
||||
}
|
||||
|
||||
func (c *muninCollector) Update() (updates int, err error) {
|
||||
err = c.registerMetrics()
|
||||
if err != nil {
|
||||
return updates, fmt.Errorf("Couldn't register metrics: %s", err)
|
||||
}
|
||||
|
||||
for _, graph := range c.graphs {
|
||||
munin, err := c.muninCommand("fetch " + graph)
|
||||
if err != nil {
|
||||
return updates, err
|
||||
}
|
||||
|
||||
for {
|
||||
line, err := munin.ReadString('\n')
|
||||
line = strings.TrimRight(line, "\n")
|
||||
if err == io.EOF {
|
||||
debug(c.Name(), "unexpected EOF, retrying")
|
||||
return c.Update()
|
||||
}
|
||||
if err != nil {
|
||||
return updates, err
|
||||
}
|
||||
if len(line) == 1 && line[0] == '.' {
|
||||
break // end of list
|
||||
}
|
||||
|
||||
parts := strings.Fields(line)
|
||||
if len(parts) != 2 {
|
||||
debug(c.Name(), "unexpected line: %s", line)
|
||||
continue
|
||||
}
|
||||
key, value_s := strings.Split(parts[0], ".")[0], parts[1]
|
||||
value, err := strconv.ParseFloat(value_s, 64)
|
||||
if err != nil {
|
||||
debug(c.Name(), "Couldn't parse value in line %s, malformed?", line)
|
||||
continue
|
||||
}
|
||||
labels := map[string]string{
|
||||
"hostname": c.hostname,
|
||||
}
|
||||
name := graph + "-" + key
|
||||
debug(c.Name(), "Set %s{%s}: %f\n", name, labels, value)
|
||||
c.gaugePerMetric[name].Set(labels, value)
|
||||
updates++
|
||||
}
|
||||
}
|
||||
return updates, err
|
||||
}
|
||||
154
exporter/native_collector.go
Normal file
154
exporter/native_collector.go
Normal file
@ -0,0 +1,154 @@
|
||||
package exporter
|
||||
|
||||
import (
|
||||
"bufio"
|
||||
"fmt"
|
||||
"github.com/prometheus/client_golang/prometheus"
|
||||
"io"
|
||||
"io/ioutil"
|
||||
"os"
|
||||
"os/exec"
|
||||
"strconv"
|
||||
"strings"
|
||||
"time"
|
||||
)
|
||||
|
||||
const (
|
||||
procLoad = "/proc/loadavg"
|
||||
)
|
||||
|
||||
type nativeCollector struct {
|
||||
loadAvg prometheus.Gauge
|
||||
attributes prometheus.Gauge
|
||||
lastSeen prometheus.Gauge
|
||||
name string
|
||||
config config
|
||||
}
|
||||
|
||||
// Takes a config struct and prometheus registry and returns a new Collector exposing
|
||||
// load, seconds since last login and a list of tags as specified by config.
|
||||
func NewNativeCollector(config config, registry prometheus.Registry) (collector nativeCollector, err error) {
|
||||
hostname, err := os.Hostname()
|
||||
if err != nil {
|
||||
return nativeCollector{}, fmt.Errorf("Couldn't get hostname: %s", err)
|
||||
}
|
||||
|
||||
collector = nativeCollector{
|
||||
name: "native_collector",
|
||||
config: config,
|
||||
loadAvg: prometheus.NewGauge(),
|
||||
attributes: prometheus.NewGauge(),
|
||||
lastSeen: prometheus.NewGauge(),
|
||||
}
|
||||
|
||||
registry.Register(
|
||||
"node_load",
|
||||
"node_exporter: system load.",
|
||||
map[string]string{"hostname": hostname},
|
||||
collector.loadAvg,
|
||||
)
|
||||
|
||||
registry.Register(
|
||||
"node_last_login_seconds",
|
||||
"node_exporter: seconds since last login.",
|
||||
map[string]string{"hostname": hostname},
|
||||
collector.lastSeen,
|
||||
)
|
||||
|
||||
registry.Register(
|
||||
"node_attributes",
|
||||
"node_exporter: system attributes.",
|
||||
map[string]string{"hostname": hostname},
|
||||
collector.attributes,
|
||||
)
|
||||
|
||||
return collector, nil
|
||||
}
|
||||
|
||||
func (c *nativeCollector) Name() string { return c.name }
|
||||
|
||||
func (c *nativeCollector) Update() (updates int, err error) {
|
||||
last, err := getSecondsSinceLastLogin()
|
||||
if err != nil {
|
||||
return updates, fmt.Errorf("Couldn't get last seen: %s", err)
|
||||
} else {
|
||||
updates++
|
||||
debug(c.Name(), "Set node_last_login_seconds: %f", last)
|
||||
c.lastSeen.Set(nil, last)
|
||||
}
|
||||
|
||||
load, err := getLoad()
|
||||
if err != nil {
|
||||
return updates, fmt.Errorf("Couldn't get load: %s", err)
|
||||
} else {
|
||||
updates++
|
||||
debug(c.Name(), "Set node_load: %f", load)
|
||||
c.loadAvg.Set(nil, load)
|
||||
}
|
||||
debug(c.Name(), "Set node_attributes{%v}: 1", c.config.Attributes)
|
||||
c.attributes.Set(c.config.Attributes, 1)
|
||||
return updates, err
|
||||
}
|
||||
|
||||
func getLoad() (float64, error) {
|
||||
data, err := ioutil.ReadFile(procLoad)
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
parts := strings.Fields(string(data))
|
||||
load, err := strconv.ParseFloat(parts[0], 64)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("Could not parse load '%s': %s", parts[0], err)
|
||||
}
|
||||
return load, nil
|
||||
}
|
||||
|
||||
func getSecondsSinceLastLogin() (float64, error) {
|
||||
who := exec.Command("who", "/var/log/wtmp", "-l", "-u", "-s")
|
||||
|
||||
output, err := who.StdoutPipe()
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
|
||||
err = who.Start()
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
|
||||
reader := bufio.NewReader(output)
|
||||
|
||||
var last time.Time
|
||||
for {
|
||||
line, isPrefix, err := reader.ReadLine()
|
||||
if err == io.EOF {
|
||||
break
|
||||
}
|
||||
if isPrefix {
|
||||
return 0, fmt.Errorf("line to long: %s(...)", line)
|
||||
}
|
||||
|
||||
fields := strings.Fields(string(line))
|
||||
lastDate := fields[2]
|
||||
lastTime := fields[3]
|
||||
|
||||
dateParts, err := splitToInts(lastDate, "-") // 2013-04-16
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("Couldn't parse date in line '%s': %s", fields, err)
|
||||
}
|
||||
|
||||
timeParts, err := splitToInts(lastTime, ":") // 11:33
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("Couldn't parse time in line '%s': %s", fields, err)
|
||||
}
|
||||
|
||||
last_t := time.Date(dateParts[0], time.Month(dateParts[1]), dateParts[2], timeParts[0], timeParts[1], 0, 0, time.UTC)
|
||||
last = last_t
|
||||
}
|
||||
err = who.Wait()
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
|
||||
return float64(time.Now().Sub(last).Seconds()), nil
|
||||
}
|
||||
22
main.go
Normal file
22
main.go
Normal file
@ -0,0 +1,22 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"flag"
|
||||
"github.com/prometheus/node_exporter/exporter"
|
||||
"log"
|
||||
)
|
||||
|
||||
var (
|
||||
configFile = flag.String("config", "node_exporter.conf", "config file.")
|
||||
memprofile = flag.String("memprofile", "", "write memory profile to this file")
|
||||
)
|
||||
|
||||
func main() {
|
||||
flag.Parse()
|
||||
|
||||
exporter, err := exporter.New(*configFile)
|
||||
if err != nil {
|
||||
log.Fatalf("Couldn't instantiate exporter: %s", err)
|
||||
}
|
||||
exporter.Loop()
|
||||
}
|
||||
@ -1,4 +1,10 @@
|
||||
{
|
||||
"scrapeInterval": 10,
|
||||
"collectors": [
|
||||
"NativeCollector",
|
||||
"GmondCollector",
|
||||
"MuninCollector"
|
||||
],
|
||||
"attributes" : {
|
||||
"web-server" : "1",
|
||||
"zone" : "a",
|
||||
|
||||
224
node_exporter.go
224
node_exporter.go
@ -1,224 +0,0 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"bufio"
|
||||
"encoding/json"
|
||||
"flag"
|
||||
"fmt"
|
||||
"github.com/prometheus/client_golang"
|
||||
"github.com/prometheus/client_golang/metrics"
|
||||
"io"
|
||||
"io/ioutil"
|
||||
"log"
|
||||
"net/http"
|
||||
_ "net/http/pprof"
|
||||
"os"
|
||||
"os/exec"
|
||||
"os/signal"
|
||||
"strconv"
|
||||
"strings"
|
||||
"syscall"
|
||||
"time"
|
||||
)
|
||||
|
||||
const (
|
||||
proto = "tcp"
|
||||
procLoad = "/proc/loadavg"
|
||||
)
|
||||
|
||||
var (
|
||||
verbose = flag.Bool("verbose", false, "Verbose output.")
|
||||
listeningAddress = flag.String("listeningAddress", ":8080", "Address on which to expose JSON metrics.")
|
||||
metricsEndpoint = flag.String("metricsEndpoint", "/metrics.json", "Path under which to expose JSON metrics.")
|
||||
configFile = flag.String("config", "node_exporter.conf", "Config file.")
|
||||
scrapeInterval = flag.Int("interval", 60, "Scrape interval.")
|
||||
|
||||
loadAvg = metrics.NewGauge()
|
||||
attributes = metrics.NewGauge()
|
||||
lastSeen = metrics.NewGauge()
|
||||
)
|
||||
|
||||
type config struct {
|
||||
Attributes map[string]string `json:"attributes"`
|
||||
}
|
||||
|
||||
func init() {
|
||||
hostname, err := os.Hostname()
|
||||
if err != nil {
|
||||
log.Fatalf("Couldn't get hostname: %s", err)
|
||||
}
|
||||
|
||||
registry.DefaultRegistry.Register(
|
||||
"node_load",
|
||||
"node_exporter: system load.",
|
||||
map[string]string{"hostname": hostname},
|
||||
loadAvg,
|
||||
)
|
||||
|
||||
registry.DefaultRegistry.Register(
|
||||
"node_last_login_seconds",
|
||||
"node_exporter: seconds since last login.",
|
||||
map[string]string{"hostname": hostname},
|
||||
lastSeen,
|
||||
)
|
||||
|
||||
registry.DefaultRegistry.Register(
|
||||
"node_attributes",
|
||||
"node_exporter: system attributes.",
|
||||
map[string]string{"hostname": hostname},
|
||||
attributes,
|
||||
)
|
||||
}
|
||||
|
||||
func debug(format string, a ...interface{}) {
|
||||
if *verbose {
|
||||
log.Printf(format, a...)
|
||||
}
|
||||
}
|
||||
|
||||
func newConfig(filename string) (conf config, err error) {
|
||||
log.Printf("Reading config %s", filename)
|
||||
bytes, err := ioutil.ReadFile(filename)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
|
||||
err = json.Unmarshal(bytes, &conf)
|
||||
return
|
||||
}
|
||||
|
||||
func serveStatus() {
|
||||
exporter := registry.DefaultRegistry.Handler()
|
||||
|
||||
http.Handle(*metricsEndpoint, exporter)
|
||||
http.ListenAndServe(*listeningAddress, nil)
|
||||
}
|
||||
|
||||
// Takes a string, splits it, converts each element to int and returns them as new list.
|
||||
// It will return an error in case any element isn't an int.
|
||||
func splitToInts(str string, sep string) (ints []int, err error) {
|
||||
for _, part := range strings.Split(str, sep) {
|
||||
i, err := strconv.Atoi(part)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("Could not split '%s' because %s is no int: %s", str, part, err)
|
||||
}
|
||||
ints = append(ints, i)
|
||||
}
|
||||
return ints, nil
|
||||
}
|
||||
|
||||
func main() {
|
||||
flag.Parse()
|
||||
sig := make(chan os.Signal, 1)
|
||||
signal.Notify(sig, syscall.SIGHUP)
|
||||
configChan := make(chan config)
|
||||
go func() {
|
||||
for _ = range sig {
|
||||
config, err := newConfig(*configFile)
|
||||
if err != nil {
|
||||
log.Printf("Couldn't reload config: %s", err)
|
||||
continue
|
||||
}
|
||||
configChan <- config
|
||||
}
|
||||
}()
|
||||
|
||||
conf, err := newConfig(*configFile)
|
||||
if err != nil {
|
||||
log.Fatalf("Couldn't read config: %s", err)
|
||||
}
|
||||
|
||||
go serveStatus()
|
||||
|
||||
tick := time.Tick(time.Duration(*scrapeInterval) * time.Second)
|
||||
for {
|
||||
select {
|
||||
case conf = <-configChan:
|
||||
log.Printf("Got new config")
|
||||
case <-tick:
|
||||
log.Printf("Starting new scrape interval")
|
||||
|
||||
last, err := getSecondsSinceLastLogin()
|
||||
if err != nil {
|
||||
log.Printf("Couldn't get last seen: %s", err)
|
||||
} else {
|
||||
debug("last: %f", last)
|
||||
lastSeen.Set(nil, last)
|
||||
}
|
||||
|
||||
load, err := getLoad()
|
||||
if err != nil {
|
||||
log.Printf("Couldn't get load: %s", err)
|
||||
} else {
|
||||
debug("load: %f", load)
|
||||
loadAvg.Set(nil, load)
|
||||
}
|
||||
|
||||
debug("attributes: %s", conf.Attributes)
|
||||
attributes.Set(conf.Attributes, 1)
|
||||
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func getLoad() (float64, error) {
|
||||
data, err := ioutil.ReadFile(procLoad)
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
parts := strings.Fields(string(data))
|
||||
load, err := strconv.ParseFloat(parts[0], 64)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("Could not parse load '%s': %s", parts[0], err)
|
||||
}
|
||||
return load, nil
|
||||
}
|
||||
|
||||
func getSecondsSinceLastLogin() (float64, error) {
|
||||
who := exec.Command("who", "/var/log/wtmp", "-l", "-u", "-s")
|
||||
|
||||
output, err := who.StdoutPipe()
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
|
||||
err = who.Start()
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
|
||||
reader := bufio.NewReader(output)
|
||||
var last time.Time
|
||||
for {
|
||||
line, isPrefix, err := reader.ReadLine()
|
||||
if err == io.EOF {
|
||||
break
|
||||
}
|
||||
if isPrefix {
|
||||
return 0, fmt.Errorf("line to long: %s(...)", line)
|
||||
}
|
||||
|
||||
fields := strings.Fields(string(line))
|
||||
lastDate := fields[2]
|
||||
lastTime := fields[3]
|
||||
|
||||
dateParts, err := splitToInts(lastDate, "-") // 2013-04-16
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
|
||||
timeParts, err := splitToInts(lastTime, ":") // 11:33
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
|
||||
last_t := time.Date(dateParts[0], time.Month(dateParts[1]), dateParts[2], timeParts[0], timeParts[1], 0, 0, time.UTC)
|
||||
last = last_t
|
||||
}
|
||||
err = who.Wait()
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
|
||||
return float64(time.Now().Sub(last).Seconds()), nil
|
||||
}
|
||||
Loading…
Reference in New Issue
Block a user