2013-05-07 16:40:10 +02:00
package main
import (
2014-02-18 12:35:11 +01:00
"encoding/json"
2013-05-07 16:40:10 +02:00
"flag"
2014-06-04 15:09:33 +02:00
"fmt"
2014-02-18 12:35:11 +01:00
"io/ioutil"
2013-05-07 16:40:10 +02:00
"log"
2014-02-18 12:35:11 +01:00
"net/http"
"os"
"os/signal"
"runtime/pprof"
2014-06-04 13:12:34 +02:00
"strings"
2014-02-18 12:35:11 +01:00
"sync"
"syscall"
"time"
2014-02-07 17:09:39 +01:00
2014-06-04 13:12:34 +02:00
"github.com/golang/glog"
2014-02-18 12:35:11 +01:00
"github.com/prometheus/client_golang/prometheus"
"github.com/prometheus/node_exporter/collector"
2013-05-07 16:40:10 +02:00
)
2014-06-26 19:20:36 +02:00
const subsystem = "exporter"
2013-05-07 16:40:10 +02:00
var (
2014-06-04 13:12:34 +02:00
configFile = flag . String ( "config" , "node_exporter.conf" , "config file." )
memProfile = flag . String ( "memprofile" , "" , "write memory profile to this file" )
listeningAddress = flag . String ( "listen" , ":8080" , "address to listen on" )
2014-11-11 16:24:35 +01:00
enabledCollectors = flag . String ( "enabledCollectors" , "attributes,diskstats,filesystem,loadavg,meminfo,stat,time,netdev,netstat" , "comma-seperated list of collectors to use" )
2014-06-04 15:09:33 +02:00
printCollectors = flag . Bool ( "printCollectors" , false , "If true, print available collectors and exit" )
2014-06-26 19:20:36 +02:00
collectorLabelNames = [ ] string { "collector" , "result" }
scrapeDurations = prometheus . NewSummaryVec (
prometheus . SummaryOpts {
Namespace : collector . Namespace ,
Subsystem : subsystem ,
Name : "scrape_duration_seconds" ,
Help : "node_exporter: Duration of a scrape job." ,
} ,
collectorLabelNames ,
)
2013-05-07 16:40:10 +02:00
)
2014-10-29 15:16:43 +01:00
// Implements Collector.
type NodeCollector struct {
collectors map [ string ] collector . Collector
}
// Implements Collector.
func ( n NodeCollector ) Describe ( ch chan <- * prometheus . Desc ) {
scrapeDurations . Describe ( ch )
}
// Implements Collector.
func ( n NodeCollector ) Collect ( ch chan <- prometheus . Metric ) {
wg := sync . WaitGroup { }
wg . Add ( len ( n . collectors ) )
for name , c := range n . collectors {
go func ( name string , c collector . Collector ) {
Execute ( name , c , ch )
wg . Done ( )
} ( name , c )
}
wg . Wait ( )
scrapeDurations . Collect ( ch )
}
func Execute ( name string , c collector . Collector , ch chan <- prometheus . Metric ) {
begin := time . Now ( )
err := c . Update ( ch )
duration := time . Since ( begin )
var result string
if err != nil {
glog . Infof ( "ERROR: %s failed after %fs: %s" , name , duration . Seconds ( ) , err )
result = "error"
} else {
glog . Infof ( "OK: %s success after %fs." , name , duration . Seconds ( ) )
result = "success"
}
scrapeDurations . WithLabelValues ( name , result ) . Observe ( duration . Seconds ( ) )
}
func getConfig ( file string ) ( * collector . Config , error ) {
config := & collector . Config { }
glog . Infof ( "Reading config %s" , * configFile )
bytes , err := ioutil . ReadFile ( * configFile )
if err != nil {
return nil , err
}
return config , json . Unmarshal ( bytes , & config )
}
func loadCollectors ( file string ) ( map [ string ] collector . Collector , error ) {
collectors := map [ string ] collector . Collector { }
config , err := getConfig ( file )
if err != nil {
2014-11-24 23:15:13 +01:00
return nil , fmt . Errorf ( "couldn't read config %s: %s" , file , err )
2014-10-29 15:16:43 +01:00
}
for _ , name := range strings . Split ( * enabledCollectors , "," ) {
fn , ok := collector . Factories [ name ]
if ! ok {
2014-11-24 23:15:13 +01:00
return nil , fmt . Errorf ( "collector '%s' not available" , name )
2014-10-29 15:16:43 +01:00
}
c , err := fn ( * config )
if err != nil {
return nil , err
}
collectors [ name ] = c
}
return collectors , nil
}
2013-05-07 16:40:10 +02:00
func main ( ) {
flag . Parse ( )
2014-06-04 15:09:33 +02:00
if * printCollectors {
fmt . Printf ( "Available collectors:\n" )
for n , _ := range collector . Factories {
fmt . Printf ( " - %s\n" , n )
}
return
}
2014-06-26 19:20:36 +02:00
collectors , err := loadCollectors ( * configFile )
2013-05-07 16:40:10 +02:00
if err != nil {
2014-02-18 12:35:11 +01:00
log . Fatalf ( "Couldn't load config and collectors: %s" , err )
2013-05-07 16:40:10 +02:00
}
2014-02-18 12:35:11 +01:00
2014-06-04 13:12:34 +02:00
glog . Infof ( "Enabled collectors:" )
for n , _ := range collectors {
glog . Infof ( " - %s" , n )
2014-02-07 17:09:39 +01:00
}
2014-02-18 12:35:11 +01:00
2014-10-29 15:16:43 +01:00
nodeCollector := NodeCollector { collectors : collectors }
prometheus . MustRegister ( nodeCollector )
2014-02-18 12:35:11 +01:00
sigUsr1 := make ( chan os . Signal )
signal . Notify ( sigUsr1 , syscall . SIGUSR1 )
2014-10-29 15:16:43 +01:00
go func ( ) {
http . Handle ( "/metrics" , prometheus . Handler ( ) )
http . ListenAndServe ( * listeningAddress , nil )
} ( )
2014-02-18 12:35:11 +01:00
for {
select {
case <- sigUsr1 :
2014-06-04 13:12:34 +02:00
glog . Infof ( "got signal" )
2014-02-18 12:35:11 +01:00
if * memProfile != "" {
2014-06-04 13:12:34 +02:00
glog . Infof ( "Writing memory profile to %s" , * memProfile )
2014-02-18 12:35:11 +01:00
f , err := os . Create ( * memProfile )
if err != nil {
log . Fatal ( err )
}
pprof . WriteHeapProfile ( f )
f . Close ( )
}
}
}
}