[#502] Add Dropped logs (by sampling) metric

Signed-off-by: Pavel Pogodaev <p.pogodaev@yadro.com>

cmd/s3-gw/app.go

@@ -83,11 +83,17 @@ type (
 		bucketResolver *resolver.BucketResolver
 		services       []*Service
 		settings       *appSettings
+		loggerSettings *loggerSettings
 
 		webDone chan struct{}
 		wrkDone chan struct{}
 	}
 
+	loggerSettings struct {
+		mu         sync.RWMutex
+		appMetrics *metrics.AppMetrics
+	}
+
 	appSettings struct {
 		logLevel            zap.AtomicLevel
 		httpLogging         s3middleware.LogHTTPConfig
@@ -132,7 +138,25 @@ type (
 	}
 )
 
-func newApp(ctx context.Context, log *Logger, v *viper.Viper) *App {
+func (s *loggerSettings) DroppedLogsInc() {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+
+	if s.appMetrics != nil {
+		s.appMetrics.Statistic().DroppedLogsInc()
+	}
+}
+
+func (s *loggerSettings) setMetrics(appMetrics *metrics.AppMetrics) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	s.appMetrics = appMetrics
+}
+
+func newApp(ctx context.Context, v *viper.Viper) *App {
+	logSettings := &loggerSettings{}
+	log := pickLogger(v, logSettings)
 	settings := newAppSettings(log, v)
 
 	objPool, treePool, key := getPools(ctx, log.logger, v, settings.dialerSource)
@@ -148,6 +172,7 @@ func newApp(ctx context.Context, log *Logger, v *viper.Viper) *App {
 		wrkDone: make(chan struct{}, 1),
 
 		settings:       settings,
+		loggerSettings: logSettings,
 	}
 
 	app.init(ctx)
@@ -521,6 +546,7 @@ func (a *App) initMetrics() {
 
 	a.metrics = metrics.NewAppMetrics(cfg)
 	a.metrics.State().SetHealth(metrics.HealthStatusStarting)
+	a.loggerSettings.setMetrics(a.metrics)
 }
 
 func (a *App) initFrostfsID(ctx context.Context) {

cmd/s3-gw/app_settings.go

@@ -1094,7 +1094,11 @@ func mergeConfig(v *viper.Viper, fileName string) error {
 	return v.MergeConfig(cfgFile)
 }
 
-func pickLogger(v *viper.Viper) *Logger {
+type LoggerAppSettings interface {
+	DroppedLogsInc()
+}
+
+func pickLogger(v *viper.Viper, settings LoggerAppSettings) *Logger {
 	lvl, err := getLogLevel(v)
 	if err != nil {
 		panic(err)
@@ -1104,9 +1108,9 @@ func pickLogger(v *viper.Viper) *Logger {
 
 	switch dest {
 	case destinationStdout:
-		return newStdoutLogger(v, lvl)
+		return newStdoutLogger(v, lvl, settings)
 	case destinationJournald:
-		return newJournaldLogger(v, lvl)
+		return newJournaldLogger(v, lvl, settings)
 	default:
 		panic(fmt.Sprintf("wrong destination for logger: %s", dest))
 	}
@@ -1126,13 +1130,13 @@ func pickLogger(v *viper.Viper) *Logger {
 // Logger records a stack trace for all messages at or above fatal level.
 //
 // See also zapcore.Level, zap.NewProductionConfig, zap.AddStacktrace.
-func newStdoutLogger(v *viper.Viper, lvl zapcore.Level) *Logger {
+func newStdoutLogger(v *viper.Viper, lvl zapcore.Level, settings LoggerAppSettings) *Logger {
 	stdout := zapcore.AddSync(os.Stderr)
 	level := zap.NewAtomicLevelAt(lvl)
 
 	consoleOutCore := zapcore.NewCore(newLogEncoder(), stdout, level)
 
-	consoleOutCore = samplingEnabling(v, consoleOutCore)
+	consoleOutCore = applyZapCoreMiddlewares(consoleOutCore, v, settings)
 
 	return &Logger{
 		logger: zap.New(consoleOutCore, zap.AddStacktrace(zap.NewAtomicLevelAt(zap.FatalLevel))),
@@ -1140,7 +1144,7 @@ func newStdoutLogger(v *viper.Viper, lvl zapcore.Level) *Logger {
 	}
 }
 
-func newJournaldLogger(v *viper.Viper, lvl zapcore.Level) *Logger {
+func newJournaldLogger(v *viper.Viper, lvl zapcore.Level, settings LoggerAppSettings) *Logger {
 	level := zap.NewAtomicLevelAt(lvl)
 
 	encoder := zapjournald.NewPartialEncoder(newLogEncoder(), zapjournald.SyslogFields)
@@ -1152,7 +1156,7 @@ func newJournaldLogger(v *viper.Viper, lvl zapcore.Level) *Logger {
 		zapjournald.SyslogPid(),
 	})
 
-	coreWithContext = samplingEnabling(v, coreWithContext)
+	coreWithContext = applyZapCoreMiddlewares(coreWithContext, v, settings)
 
 	l := zap.New(coreWithContext, zap.AddStacktrace(zap.NewAtomicLevelAt(zap.FatalLevel)))
 
@@ -1169,19 +1173,17 @@ func newLogEncoder() zapcore.Encoder {
 	return zapcore.NewConsoleEncoder(c)
 }
 
-func samplingEnabling(v *viper.Viper, core zapcore.Core) zapcore.Core {
+func applyZapCoreMiddlewares(core zapcore.Core, v *viper.Viper, settings LoggerAppSettings) zapcore.Core {
-	// Zap samples by logging the first cgfLoggerSamplingInitial entries with a given level
-	// and message within the specified time interval.
-	// In the above config, only the first cgfLoggerSamplingInitial log entries with the same level and message
-	// are recorded in cfgLoggerSamplingInterval interval. Every other log entry will be dropped within the interval since
-	// cfgLoggerSamplingThereafter is specified here.
 	if v.GetBool(cfgLoggerSamplingEnabled) {
-		core = zapcore.NewSamplerWithOptions(
+		core = zapcore.NewSamplerWithOptions(core,
-			core,
 			v.GetDuration(cfgLoggerSamplingInterval),
 			v.GetInt(cfgLoggerSamplingInitial),
 			v.GetInt(cfgLoggerSamplingThereafter),
-		)
+			zapcore.SamplerHook(func(_ zapcore.Entry, dec zapcore.SamplingDecision) {
+				if dec&zapcore.LogDropped > 0 {
+					settings.DroppedLogsInc()
+				}
+			}))
 	}
 
 	return core

cmd/s3-gw/main.go

@@ -8,11 +8,9 @@ import (
 
 func main() {
 	g, _ := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM)
-
 	v := newSettings()
-	l := pickLogger(v)
 
-	a := newApp(g, l, v)
+	a := newApp(g, v)
 
 	go a.Serve(g)
 

metrics/desc.go

@@ -93,6 +93,13 @@ var appMetricsDesc = map[string]map[string]Description{
 		},
 	},
 	statisticSubsystem: {
+		droppedLogs: Description{
+			Type:      dto.MetricType_COUNTER,
+			Namespace: namespace,
+			Subsystem: statisticSubsystem,
+			Name:      droppedLogs,
+			Help:      "Dropped logs (by sampling) count",
+		},
 		requestsSecondsMetric: Description{
 			Type:           dto.MetricType_HISTOGRAM,
 			Namespace:      namespace,
@@ -252,3 +259,12 @@ func mustNewHistogramVec(description Description, buckets []float64) *prometheus
 		description.VariableLabels,
 	)
 }
+
+func mustNewCounter(description Description) prometheus.Counter {
+	if description.Type != dto.MetricType_COUNTER {
+		panic("invalid metric type")
+	}
+	return prometheus.NewCounter(
+		prometheus.CounterOpts(newOpts(description)),
+	)
+}

metrics/stats.go

@@ -34,6 +34,7 @@ type (
 	APIStatMetrics struct {
 		stats                *httpStats
 		httpRequestsDuration *prometheus.HistogramVec
+		droppedLogs          prometheus.Counter
 	}
 )
 
@@ -47,6 +48,7 @@ const (
 	requestsTotalMetric   = "requests_total"
 	errorsTotalMetric     = "errors_total"
 	bytesTotalMetric      = "bytes_total"
+	droppedLogs           = "dropped_logs"
 )
 
 const (
@@ -61,6 +63,7 @@ func newAPIStatMetrics() *APIStatMetrics {
 		stats: newHTTPStats(),
 		httpRequestsDuration: mustNewHistogramVec(histogramDesc,
 			[]float64{.05, .1, .25, .5, 1, 2.5, 5, 10}),
+		droppedLogs: mustNewCounter(appMetricsDesc[statisticSubsystem][droppedLogs]),
 	}
 }
 
@@ -119,6 +122,7 @@ func (a *APIStatMetrics) Describe(ch chan<- *prometheus.Desc) {
 		return
 	}
 	a.stats.Describe(ch)
+	a.droppedLogs.Describe(ch)
 	a.httpRequestsDuration.Describe(ch)
 }
 
@@ -127,9 +131,17 @@ func (a *APIStatMetrics) Collect(ch chan<- prometheus.Metric) {
 		return
 	}
 	a.stats.Collect(ch)
+	a.droppedLogs.Collect(ch)
 	a.httpRequestsDuration.Collect(ch)
 }
 
+func (a *APIStatMetrics) DroppedLogsInc() {
+	if a == nil {
+		return
+	}
+	a.droppedLogs.Inc()
+}
+
 func newHTTPStats() *httpStats {
 	return &httpStats{
 		currentS3RequestsDesc: newDesc(appMetricsDesc[statisticSubsystem][requestsCurrentMetric]),