//SanitizeInfluxInput adds backslashes to special chars.
func SanitizeInfluxInput(input string) string {
if len(input) == 0 {
return input
}
if string(input[0]) == `"` && string(input[len(input)-1]) == `"` {
return input
}
if config.GetConfig().Influx.NastyString != "" {
input = strings.Replace(input, config.GetConfig().Influx.NastyString, config.GetConfig().Influx.NastyStringToReplace, -1)
}
input = strings.Trim(input, `'`)
input = strings.Replace(input, " ", `\ `, -1)
input = strings.Replace(input, ",", `\,`, -1)
return input
}
//PrintForInfluxDB prints the data in influxdb lineformat
func (p PerformanceData) PrintForInfluxDB(version string) string {
if helper.VersionOrdinal(version) >= helper.VersionOrdinal("0.9") {
tableName := fmt.Sprintf(`metrics,host=%s`, helper.SanitizeInfluxInput(p.hostname))
if p.service == "" {
tableName += fmt.Sprintf(`,service=%s`, helper.SanitizeInfluxInput(config.GetConfig().Influx.HostcheckAlias))
} else {
tableName += fmt.Sprintf(`,service=%s`, helper.SanitizeInfluxInput(p.service))
}
tableName += fmt.Sprintf(`,command=%s,performanceLabel=%s`,
helper.SanitizeInfluxInput(p.command),
helper.SanitizeInfluxInput(p.performanceLabel),
)
if len(p.tags) > 0 {
tableName += fmt.Sprintf(`,%s`, helper.PrintMapAsString(helper.SanitizeMap(p.tags), ",", "="))
}
if p.unit != "" {
tableName += fmt.Sprintf(`,unit=%s`, p.unit)
}
tableName += fmt.Sprintf(` %s`, helper.PrintMapAsString(helper.SanitizeMap(p.fields), ",", "="))
tableName += fmt.Sprintf(" %s\n", p.time)
return tableName
}
return ""
}
//createTemplate creates the nagflux template.
func (connector *Connector) createTemplate() bool {
mapping := fmt.Sprintf(NagfluxTemplate, connector.index, config.GetConfig().Elasticsearch.NumberOfShards, config.GetConfig().Elasticsearch.NumberOfReplicas)
createIndex, _ := helper.SentReturnCodeIsOK(connector.httpClient, connector.connectionHost+"_template/"+connector.index, "PUT", mapping)
if !createIndex {
return false
}
return true
}
func (live Data) genElasticLineWithValue(index, typ, value, timestamp string) string {
value = strings.Replace(value, `"`, `\"`, -1)
if live.serviceDisplayName == "" {
live.serviceDisplayName = config.GetConfig().Elasticsearch.HostcheckAlias
}
head := fmt.Sprintf(`{"index":{"_index":"%s","_type":"messages"}}`, helper.GenIndex(index, timestamp)) + "\n"
data := fmt.Sprintf(`{"timestamp":%s,"message":"%s","author":"%s","host":"%s","service":"%s","type":"%s"}`+"\n",
helper.CastStringTimeFromSToMs(timestamp), value, live.author, live.hostName, live.serviceDisplayName, typ,
)
return head + data
}
//GenIndex generates an index depending on the config, ending with year and month
func GenIndex(index, timeString string) string {
rotation := config.GetConfig().Elasticsearch.IndexRotation
year, month := GetYearMonthFromStringTimeMs(timeString)
switch rotation {
case "monthly":
return fmt.Sprintf("%s-%d.%02d", index, year, month)
case "yearly":
return fmt.Sprintf("%s-%d", index, year)
default:
panic(fmt.Sprintf("The given IndexRotation[%s] is not supported", rotation))
}
}
func getLivestatusVersion(live *Collector) int {
printables := make(chan collector.Printable, 1)
finished := make(chan bool, 1)
var version string
live.requestPrintablesFromLivestatus(QueryLivestatusVersion, false, printables, finished)
i := 0
oneMinute := time.Duration(1) * time.Minute
roundsToWait := config.GetConfig().Livestatus.MinutesToWait
Loop:
for roundsToWait != 0 {
select {
case versionPrintable := <-printables:
version = versionPrintable.PrintForInfluxDB("0")
break Loop
case <-time.After(oneMinute):
if i < roundsToWait {
go live.requestPrintablesFromLivestatus(QueryLivestatusVersion, false, printables, finished)
} else {
break Loop
}
i++
case fin := <-finished:
if !fin {
live.log.Infof(
"Could not detect livestatus version, waiting for %s %d times( %d/%d )...",
oneMinute, roundsToWait, i, roundsToWait,
)
}
}
}
live.log.Info("Livestatus version: ", version)
if icinga2, _ := regexp.MatchString(`^r[\d\.-]+$`, version); icinga2 {
return Icinga2
} else if nagios, _ := regexp.MatchString(`^[\d\.]+p[\d\.]+$`, version); nagios {
return Nagios
} else if neamon, _ := regexp.MatchString(`^[\d\.]+-naemon$`, version); neamon {
return Naemon
}
live.log.Warn("Could not detect livestatus type, with version: ", version, ". Asuming Nagios")
return -1
}
func (g GearmanWorker) handleLoad() {
bufferLimit := int(float32(config.GetConfig().Main.BufferSize) * 0.90)
for {
for _, r := range g.results {
if len(r) > bufferLimit && g.worker != nil {
g.worker.Lock()
for len(r) > bufferLimit {
time.Sleep(time.Duration(100) * time.Millisecond)
}
g.worker.Unlock()
}
}
select {
case <-g.quit:
g.quit <- true
return
case <-time.After(time.Duration(1) * time.Second):
}
}
}
//PrintForElasticsearch prints in the elasticsearch json format
func (p PerformanceData) PrintForElasticsearch(version, index string) string {
if helper.VersionOrdinal(version) >= helper.VersionOrdinal("2.0") {
if p.service == "" {
p.service = config.GetConfig().Influx.HostcheckAlias
}
head := fmt.Sprintf(`{"index":{"_index":"%s","_type":"metrics"}}`, helper.GenIndex(index, p.time)) + "\n"
data := fmt.Sprintf(
`{"timestamp":%s,"host":"%s","service":"%s","command":"%s","performanceLabel":"%s"`,
p.time,
helper.SanitizeElasicInput(p.hostname),
helper.SanitizeElasicInput(p.service),
helper.SanitizeElasicInput(p.command),
helper.SanitizeElasicInput(p.performanceLabel),
)
if p.unit != "" {
data += fmt.Sprintf(`,"unit":"%s"`, helper.SanitizeElasicInput(p.unit))
}
data += helper.CreateJSONFromStringMap(p.tags)
data += helper.CreateJSONFromStringMap(p.fields)
data += "}\n"
return head + data
}
return ""
}
//Generates the Influxdb tablename.
func (live Data) getTablename() string {
if live.serviceDisplayName == "" {
live.serviceDisplayName = config.GetConfig().Influx.HostcheckAlias
}
return fmt.Sprintf("messages,host=%s,service=%s", live.hostName, live.serviceDisplayName)
}
func main() {
//Parse Args
var configPath string
var printver bool
flag.Usage = func() {
fmt.Println(`Nagflux by Philip Griesbacher`, nagfluxVersion, `
Commandline Parameter:
-configPath Path to the config file. If no file path is given the default is ./config.gcfg.
-V Print version and exit`)
}
flag.StringVar(&configPath, "configPath", "config.gcfg", "path to the config file")
flag.BoolVar(&printver, "V", false, "print version and exit")
flag.Parse()
//Print version and exit
if printver {
fmt.Println(nagfluxVersion)
os.Exit(0)
}
//Load config
config.InitConfig(configPath)
cfg := config.GetConfig()
//Create Logger
logging.InitLogger(cfg.Log.LogFile, cfg.Log.MinSeverity)
log = logging.GetLogger()
log.Info(`Started Nagflux `, nagfluxVersion)
resultQueues := map[data.Datatype]chan collector.Printable{}
stoppables := []Stoppable{}
if len(cfg.Main.FieldSeparator) < 1 {
panic("FieldSeparator is too short!")
}
pro := statistics.NewPrometheusServer(cfg.Monitoring.PrometheusAddress)
pro.WatchResultQueueLength(resultQueues)
fieldSeparator := []rune(cfg.Main.FieldSeparator)[0]
config.PauseNagflux.Store(false)
if cfg.Influx.Enabled {
resultQueues[data.InfluxDB] = make(chan collector.Printable, cfg.Main.BufferSize)
influx := influx.ConnectorFactory(resultQueues[data.InfluxDB], cfg.Influx.Address, cfg.Influx.Arguments, cfg.Main.DumpFile, cfg.Influx.Version, cfg.Main.InfluxWorker, cfg.Main.MaxInfluxWorker, cfg.Influx.CreateDatabaseIfNotExists)
stoppables = append(stoppables, influx)
influxDumpFileCollector := nagflux.NewDumpfileCollector(resultQueues[data.InfluxDB], cfg.Main.DumpFile, data.InfluxDB)
stoppables = append(stoppables, influxDumpFileCollector)
}
if cfg.Elasticsearch.Enabled {
resultQueues[data.Elasticsearch] = make(chan collector.Printable, cfg.Main.BufferSize)
elasticsearch := elasticsearch.ConnectorFactory(resultQueues[data.Elasticsearch], cfg.Elasticsearch.Address, cfg.Elasticsearch.Index, cfg.Main.DumpFile, cfg.Elasticsearch.Version, cfg.Main.InfluxWorker, cfg.Main.MaxInfluxWorker, true)
stoppables = append(stoppables, elasticsearch)
elasticDumpFileCollector := nagflux.NewDumpfileCollector(resultQueues[data.Elasticsearch], cfg.Main.DumpFile, data.Elasticsearch)
stoppables = append(stoppables, elasticDumpFileCollector)
}
//Some time for the dumpfile to fill the queue
time.Sleep(time.Duration(100) * time.Millisecond)
liveconnector := &livestatus.Connector{log, cfg.Livestatus.Address, cfg.Livestatus.Type}
livestatusCollector := livestatus.NewLivestatusCollector(resultQueues, liveconnector, true)
livestatusCache := livestatus.NewLivestatusCacheBuilder(liveconnector)
for name, data := range cfg.ModGearman {
if data == nil || !(*data).Enabled {
continue
}
log.Infof("Mod_Gearman: %s - %s [%s]", name, (*data).Address, (*data).Queue)
secret := modGearman.GetSecret((*data).Secret, (*data).SecretFile)
for i := 0; i < (*data).Worker; i++ {
gearmanWorker := modGearman.NewGearmanWorker((*data).Address,
(*data).Queue,
secret,
resultQueues,
livestatusCache,
)
stoppables = append(stoppables, gearmanWorker)
}
}
log.Info("Nagios Spoolfile Folder: ", cfg.Main.NagiosSpoolfileFolder)
nagiosCollector := spoolfile.NagiosSpoolfileCollectorFactory(cfg.Main.NagiosSpoolfileFolder, cfg.Main.NagiosSpoolfileWorker, resultQueues, livestatusCache)
log.Info("Nagflux Spoolfile Folder: ", cfg.Main.NagfluxSpoolfileFolder)
nagfluxCollector := nagflux.NewNagfluxFileCollector(resultQueues, cfg.Main.NagfluxSpoolfileFolder, fieldSeparator)
//Listen for Interrupts
interruptChannel := make(chan os.Signal, 1)
signal.Notify(interruptChannel, syscall.SIGINT)
signal.Notify(interruptChannel, syscall.SIGTERM)
go func() {
<-interruptChannel
log.Warn("Got Interrupted")
stoppables = append(stoppables, []Stoppable{livestatusCollector, livestatusCache, nagiosCollector, nagfluxCollector}...)
cleanUp(stoppables, resultQueues)
quit <- true
}()
loop:
//Main loop
for {
select {
case <-time.After(time.Duration(updateRate) * time.Second):
/*queriesSend, measureTime, err := statisticUser.GetData("send")
if err != nil {
continue
}
idleTime := (measureTime.Seconds() - queriesSend.Time.Seconds() / float64(influx.AmountWorkers())) / updateRate
log.Debugf("Buffer len: %d - Idletime in percent: %0.2f ", len(resultQueues[0]), idleTime * 100)
//TODO: fix worker spawn by type
if idleTime > 0.25 {
influx.RemoveWorker()
} else if idleTime < 0.1 && float64(len(resultQueues[0])) > resultQueueLength * 0.8 {
influx.AddWorker()
}*/
case <-quit:
break loop
}
}
}
