Golang slog.Errorf函数代码示例

func DatabaseConnect(c conf.Database) (*sql.DB, error) {
	var db *sql.DB
	var err error

	if c.Type == "mysql" {
		if c.Password == "" {
			db, err = sql.Open("mysql", c.Username+"@"+c.Protocol+"("+c.Address+":"+strconv.Itoa(c.Port)+")/")
		} else {
			db, err = sql.Open("mysql", c.Username+":"+c.Password+"@"+c.Protocol+"("+c.Address+":"+strconv.Itoa(c.Port)+")/")
		}
	}

	if err != nil {
		slog.Errorf("%v: %v: %v: %v", "Database", c.Type, "connect open error", err)
		return nil, fmt.Errorf("%v: %v", "connect open error", err)
	}

	// check if can connect to database, if fail, check again every minute until connected
	for {
		err = db.Ping()
		if err == nil {
			break
		}
		slog.Errorf("%v: %v: %v: %v", "Database", c.Type, "connect ping error", err)
		next := time.After(time.Minute)
		select {
		case <-next:
		case <-ContinuousCollectorVars.quit:
			db.Close()
			return nil, fmt.Errorf("connect while quitting")
		}
	}

	return db, nil
}

// Command executes the named program with the given arguments. If it does not
// exit within timeout, it is sent SIGINT (if supported by Go). After
// another timeout, it is killed.
func Command(timeout time.Duration, stdin io.Reader, name string, arg ...string) (io.Reader, error) {
	if _, err := exec.LookPath(name); err != nil {
		return nil, ErrPath
	}
	if Debug {
		slog.Infof("executing command: %v %v", name, arg)
	}
	c := exec.Command(name, arg...)
	b := &bytes.Buffer{}
	c.Stdout = b
	c.Stdin = stdin
	if err := c.Start(); err != nil {
		return nil, err
	}
	timedOut := false
	intTimer := time.AfterFunc(timeout, func() {
		slog.Errorf("Process taking too long. Interrupting: %s %s", name, strings.Join(arg, " "))
		c.Process.Signal(os.Interrupt)
		timedOut = true
	})
	killTimer := time.AfterFunc(timeout*2, func() {
		slog.Errorf("Process taking too long. Killing: %s %s", name, strings.Join(arg, " "))
		c.Process.Signal(os.Kill)
		timedOut = true
	})
	err := c.Wait()
	intTimer.Stop()
	killTimer.Stop()
	if timedOut {
		return nil, ErrTimeout
	}
	return b, err
}

func (s *Schedule) save() {
	if s.db == nil {
		return
	}
	s.Lock("Save")
	store := map[string]interface{}{
		dbMetric:        s.Search.Read.Metric,
		dbTagk:          s.Search.Read.Tagk,
		dbTagv:          s.Search.Read.Tagv,
		dbMetricTags:    s.Search.Read.MetricTags,
		dbNotifications: s.Notifications,
		dbSilence:       s.Silence,
		dbStatus:        s.status,
		dbMetadata:      s.Metadata,
		dbIncidents:     s.Incidents,
	}
	tostore := make(map[string][]byte)
	for name, data := range store {
		f := new(bytes.Buffer)
		gz := gzip.NewWriter(f)
		cw := &counterWriter{w: gz}
		enc := gob.NewEncoder(cw)
		if err := enc.Encode(data); err != nil {
			slog.Errorf("error saving %s: %v", name, err)
			s.Unlock()
			return
		}
		if err := gz.Flush(); err != nil {
			slog.Errorf("gzip flush error saving %s: %v", name, err)
		}
		if err := gz.Close(); err != nil {
			slog.Errorf("gzip close error saving %s: %v", name, err)
		}
		tostore[name] = f.Bytes()
		slog.Infof("wrote %s: %v", name, conf.ByteSize(cw.written))
		collect.Put("statefile.size", opentsdb.TagSet{"object": name}, cw.written)
	}
	s.Unlock()
	err := s.db.Update(func(tx *bolt.Tx) error {
		b, err := tx.CreateBucketIfNotExists([]byte(dbBucket))
		if err != nil {
			return err
		}
		for name, data := range tostore {
			if err := b.Put([]byte(name), data); err != nil {
				return err
			}
		}
		return nil
	})
	if err != nil {
		slog.Errorf("save db update error: %v", err)
		return
	}
	fi, err := os.Stat(s.Conf.StateFile)
	if err == nil {
		collect.Put("statefile.size", opentsdb.TagSet{"object": "total"}, fi.Size())
	}
	slog.Infoln("save to db complete")
}

// structProcessor.add() takes in a metric name prefix, an arbitrary struct, and a tagset.
// The processor recurses through the struct and builds metrics. The field tags direct how
// the field should be processed, as well as the metadata for the resulting metric.
//
// The field tags used are described as follows:
//
// version: typically set to '1' or '2'.
//	This is compared against the elastic cluster version. If the version from the tag
//      does not match the version in production, the metric will not be sent for this field.
//
// exclude:
//      If this tag is set to 'true', a metric will not be sent for this field.
//
// rate: one of 'gauge', 'counter', 'rate'
//	This tag dictates the metadata.RateType we send.
//
// unit: 'bytes', 'pages', etc
//	This tag dictates the metadata.Unit we send.
//
// metric:
//      This is the metric name which will be sent. If not present, the 'json'
//      tag is sent as the metric name.
//
// Special handling:
//
// Metrics having the json tag suffix of 'in_milliseconds' are automagically
// divided by 1000 and sent as seconds. The suffix is stripped from the name.
//
// Metrics having the json tag suffix of 'in_bytes' are automatically sent as
// gauge bytes. The suffix is stripped from the metric name.
func (s *structProcessor) add(prefix string, st interface{}, ts opentsdb.TagSet) {
	t := reflect.TypeOf(st)
	valueOf := reflect.ValueOf(st)
	for i := 0; i < t.NumField(); i++ {
		field := t.Field(i)
		value := valueOf.Field(i).Interface()
		if field.Tag.Get("exclude") == "true" {
			continue
		}
		var (
			jsonTag    = field.Tag.Get("json")
			metricTag  = field.Tag.Get("metric")
			versionTag = field.Tag.Get("version")
			rateTag    = field.Tag.Get("rate")
			unitTag    = field.Tag.Get("unit")
		)
		metricName := jsonTag
		if metricTag != "" {
			metricName = metricTag
		}
		if metricName == "" {
			slog.Errorf("Unable to determine metric name for field %s. Skipping.", field.Name)
			continue
		}
		if versionTag == "" || strings.HasPrefix(s.elasticVersion, versionTag) {
			switch value := value.(type) {
			case int, float64: // Number types in our structs are only ints and float64s.
				// Turn all millisecond metrics into seconds
				if strings.HasSuffix(metricName, "_in_millis") {
					switch value.(type) {
					case int:
						value = float64(value.(int)) / 1000
					case float64:
						value = value.(float64) / 1000
					}
					unitTag = "seconds"
					metricName = strings.TrimSuffix(metricName, "_in_millis")
				}
				// Set rate and unit for all "_in_bytes" metrics, and strip the "_in_bytes"
				if strings.HasSuffix(metricName, "_in_bytes") {
					if rateTag == "" {
						rateTag = "gauge"
					}
					unitTag = "bytes"
					metricName = strings.TrimSuffix(metricName, "_in_bytes")
				}
				Add(s.md, prefix+"."+metricName, value, ts, metadata.RateType(rateTag), metadata.Unit(unitTag), field.Tag.Get("desc"))
			case string:
				// The json data has a lot of strings, and we don't care about em.
			default:
				// If we hit another struct, recurse
				if reflect.ValueOf(value).Kind() == reflect.Struct {
					s.add(prefix+"."+metricName, value, ts)
				} else {
					slog.Errorf("Field %s for metric %s is non-numeric type. Cannot record as a metric.\n", field.Name, prefix+"."+metricName)
				}
			}
		}
	}
}

func runService(name string, isDebug bool) {
	errFix := fixEventMessageFile(name) //Temp fix. Remove after a few weeks.
	if errFix != nil {
		slog.Errorf("%s fixEventMessageFile failed: %v", name, errFix)
		return
	}
	if isDebug {
		slog.SetEventLog(debug.New(name), 1)
	} else {
		elog, err := eventlog.Open(name)
		if err != nil {
			return
		}
		slog.SetEventLog(elog, 1)
		defer elog.Close()
	}
	slog.Infof("starting %s service version %v (%v)", name, version.Version, version.VersionSHA)
	run := svc.Run
	if isDebug {
		run = debug.Run
	}
	err := run(name, &s{})
	if err != nil {
		slog.Errorf("%s service failed: %v", name, err)
		return
	}
	slog.Infof("%s service stopped", name)
	os.Exit(0)
}

func c_redis_counters(server string, db int) (opentsdb.MultiDataPoint, error) {
	var md opentsdb.MultiDataPoint
	conn, err := redis.Dial("tcp", server, redis.DialDatabase(db))
	if err != nil {
		return md, err
	}
	defer conn.Close()
	if _, err := conn.Do("CLIENT", "SETNAME", "scollector"); err != nil {
		return md, err
	}
	cursor := 0
	for {
		vals, err := redis.Values(conn.Do("HSCAN", collect.RedisCountersKey, cursor))
		if err != nil {
			return md, err
		}
		if len(vals) != 2 {
			return md, fmt.Errorf("Unexpected number of values")
		}
		cursor, err = redis.Int(vals[0], nil)
		if err != nil {
			return md, err
		}
		pairs, err := redis.StringMap(vals[1], nil)
		if err != nil {
			return md, err
		}
		for mts, val := range pairs {
			parts := strings.Split(mts, ":")
			if len(parts) != 2 {
				slog.Errorf("Invalid metric tag set counter: %s", mts)
				continue
			}
			metric := parts[0]
			tags, err := opentsdb.ParseTags(parts[1])
			if err != nil {
				slog.Errorf("Invalid tags: %s", parts[1])
				continue
			}
			v, err := strconv.Atoi(val)
			if err != nil {
				slog.Errorf("Invalid counter value: %s", val)
				continue
			}
			Add(&md, metric, v, tags, metadata.Counter, metadata.Count, "")
		}
		if cursor == 0 {
			break
		}
	}
	return md, nil
}

func (s *Schedule) PutMetadata(k metadata.Metakey, v interface{}) {

	isCoreMeta := (k.Name == "desc" || k.Name == "unit" || k.Name == "rate")
	if !isCoreMeta {
		s.metaLock.Lock()
		s.Metadata[k] = &Metavalue{time.Now().UTC(), v}
		s.metaLock.Unlock()
		return
	}
	if k.Metric == "" {
		slog.Error("desc, rate, and unit require metric name")
		return
	}
	strVal, ok := v.(string)
	if !ok {
		slog.Errorf("desc, rate, and unit require value to be string. Found: %s", reflect.TypeOf(v))
		return
	}
	s.metricMetaLock.Lock()
	metricData, ok := s.metricMetadata[k.Metric]
	if !ok {
		metricData = &MetadataMetric{}
		s.metricMetadata[k.Metric] = metricData
	}
	switch k.Name {
	case "desc":
		metricData.Description = strVal
	case "unit":
		metricData.Unit = strVal
	case "rate":
		metricData.Type = strVal
	}
	s.metricMetaLock.Unlock()
}

func timeTSDBRequest(e *State, T miniprofiler.Timer, req *opentsdb.Request) (s opentsdb.ResponseSet, err error) {
	e.tsdbQueries = append(e.tsdbQueries, *req)
	if e.autods > 0 {
		for _, q := range req.Queries {
			if q.Downsample == "" {
				if err := req.AutoDownsample(e.autods); err != nil {
					return nil, err
				}
			}
		}
	}
	b, _ := json.MarshalIndent(req, "", "  ")
	tries := 1
	for {
		T.StepCustomTiming("tsdb", "query", string(b), func() {
			getFn := func() (interface{}, error) {
				return e.tsdbContext.Query(req)
			}
			var val interface{}
			val, err = e.cache.Get(string(b), getFn)
			s = val.(opentsdb.ResponseSet).Copy()

		})
		if err == nil || tries == tsdbMaxTries {
			break
		}
		slog.Errorf("Error on tsdb query %d: %s", tries, err.Error())
		tries++
	}
	return
}

func (s *Schedule) CheckAlert(T miniprofiler.Timer, r *RunHistory, a *conf.Alert) {
	slog.Infof("check alert %v start", a.Name)
	start := utcNow()
	for _, ak := range s.findUnknownAlerts(r.Start, a.Name) {
		r.Events[ak] = &models.Event{Status: models.StUnknown}
	}
	var warns, crits models.AlertKeys
	d, err := s.executeExpr(T, r, a, a.Depends)
	var deps expr.ResultSlice
	if err == nil {
		deps = filterDependencyResults(d)
		crits, err = s.CheckExpr(T, r, a, a.Crit, models.StCritical, nil)
		if err == nil {
			warns, err = s.CheckExpr(T, r, a, a.Warn, models.StWarning, crits)
		}
	}
	unevalCount, unknownCount := markDependenciesUnevaluated(r.Events, deps, a.Name)
	if err != nil {
		slog.Errorf("Error checking alert %s: %s", a.Name, err.Error())
		removeUnknownEvents(r.Events, a.Name)
		s.markAlertError(a.Name, err)
	} else {
		s.markAlertSuccessful(a.Name)
	}
	collect.Put("check.duration", opentsdb.TagSet{"name": a.Name}, time.Since(start).Seconds())
	slog.Infof("check alert %v done (%s): %v crits, %v warns, %v unevaluated, %v unknown", a.Name, time.Since(start), len(crits), len(warns), unevalCount, unknownCount)
}

func init() {
	registerInit(func(c *conf.Conf) {
		for _, filter := range c.ElasticIndexFilters {
			err := AddElasticIndexFilter(filter)
			if err != nil {
				slog.Errorf("Error processing ElasticIndexFilter: %s", err)
			}
		}
		collectors = append(collectors, &IntervalCollector{
			F: func() (opentsdb.MultiDataPoint, error) {
				return c_elasticsearch(false)
			},
			name:   "elasticsearch",
			Enable: enableURL("http://localhost:9200/"),
		})
		collectors = append(collectors, &IntervalCollector{
			F: func() (opentsdb.MultiDataPoint, error) {
				return c_elasticsearch(true)
			},
			name:     "elasticsearch-indices",
			Interval: time.Minute * 15,
			Enable:   enableURL("http://localhost:9200/"),
		})
	})
}

func (s *Schedule) findUnknownAlerts(now time.Time, alert string) []models.AlertKey {
	keys := []models.AlertKey{}
	if utcNow().Sub(bosunStartupTime) < s.SystemConf.GetCheckFrequency() {
		return keys
	}
	if !s.AlertSuccessful(alert) {
		return keys
	}
	a := s.RuleConf.GetAlert(alert)
	t := a.Unknown
	if t == 0 {
		runEvery := s.SystemConf.GetDefaultRunEvery()
		if a.RunEvery != 0 {
			runEvery = a.RunEvery
		}
		t = s.SystemConf.GetCheckFrequency() * 2 * time.Duration(runEvery)
	}
	maxTouched := now.UTC().Unix() - int64(t.Seconds())
	untouched, err := s.DataAccess.State().GetUntouchedSince(alert, maxTouched)
	if err != nil {
		slog.Errorf("Error finding unknown alerts for alert %s: %s.", alert, err)
		return keys
	}
	for _, ak := range untouched {
		if a.Squelch.Squelched(ak.Group()) {
			continue
		}
		keys = append(keys, ak)
	}
	return keys
}

func c_fastly(c fastlyClient) (opentsdb.MultiDataPoint, error) {
	var md opentsdb.MultiDataPoint
	to := time.Now().UTC().Truncate(time.Minute)
	from := to.Add(-15 * time.Minute) // "Minutely data will be delayed by roughly 10 to 15 minutes from the current time -- Fastly Docs"

	// Aggregate
	statsCollection, err := c.GetAggregateStats(from, to)
	if err != nil {
		return md, err
	}
	for _, stats := range statsCollection {
		fastlyReflectAdd(&md, "fastly", "", stats, stats.StartTime, nil)
	}

	// By Service
	services, err := c.GetServices()
	if err != nil {
		return md, err
	}
	for _, service := range services {
		statsCollection, err := c.GetServiceStats(from, to, service.Id)
		if err != nil {
			slog.Errorf("couldn't get stats for service %v with id %v: %v", service.Name, service.Id, err)
			continue
		}
		for _, stats := range statsCollection {
			fastlyReflectAdd(&md, "fastly", "_by_service", stats, stats.StartTime, service.TagSet())
		}
	}

	// By Region
	regions, err := c.GetRegions()
	if err != nil {
		return md, err
	}
	for _, region := range regions {
		statsCollection, err := c.GetRegionStats(from, to, region)
		if err != nil {
			slog.Errorf("couldn't get stats for region %v: %v", region, err)
			continue
		}
		for _, stats := range statsCollection {
			fastlyReflectAdd(&md, "fastly", "_by_region", stats, stats.StartTime, region.TagSet())
		}
	}
	return md, nil
}

func incrementRedisCounter(data []byte, addr string, conn redis.Conn) {
	if len(data) < 5 {
		slog.Errorf("Insufficient data for increment from %s.", addr)
		return
	}
	r := bytes.NewReader(data)
	var i int32
	err := binary.Read(r, binary.BigEndian, &i)
	if err != nil {
		slog.Error(err)
		return
	}
	mts := string(data[4:])
	if _, err = conn.Do("HINCRBY", RedisCountersKey, mts, i); err != nil {
		slog.Errorf("Error incrementing counter %s by %d. From %s. %s", mts, i, addr, err)
	}
}

func (s *Schedule) GetUnknownAndUnevaluatedAlertKeys(alert string) (unknown, uneval []models.AlertKey) {
	unknown, uneval, err := s.DataAccess.State().GetUnknownAndUnevalAlertKeys(alert)
	if err != nil {
		slog.Errorf("Error getting unknown/unevaluated alert keys: %s", err)
		return nil, nil
	}
	return unknown, uneval
}

// snmp_subtree takes an oid and returns all data exactly one level below it. It
// produces an error if there is more than one level below.
func snmp_subtree(host, community, oid string) (map[string]interface{}, error) {
	rows, err := snmp.Walk(host, community, oid)
	if err != nil {
		return nil, err
	}
	m := make(map[string]interface{})
	for rows.Next() {
		key := ""
		var a interface{}
		switch oid {
		case ifHCInBroadcastPkts:
			a = new(big.Int)
			id, err := rows.Scan(&a)
			if err != nil {
				slog.Errorf("Error scanning oid %v on host %v: %v", oid, host, err)
				continue
			}
			switch t := id.(type) {
			case int:
				key = fmt.Sprint(t)
			default:
				return nil, fmt.Errorf("snmp subtree: only one level allowed")
			}
		default:
			id, err := rows.Scan(&a)
			if err != nil {
				slog.Errorf("Error scanning oid %v on host %v: %v", oid, host, err)
				continue
			}
			switch t := id.(type) {
			case int:
				key = fmt.Sprint(t)
			case []int:
				key = snmpOidArrayToString(t)
			default:
				return nil, fmt.Errorf("Unknown key type: %s", reflect.TypeOf(id).String())
			}
		}
		m[key] = a
	}
	if err := rows.Err(); err != nil && err != io.EOF {
		return nil, err
	}
	return m, nil
}