Add grace period for metrics late for aggregation (#6049)

internal/models/running_aggregator.go

@@ -59,6 +59,7 @@ type AggregatorConfig struct {
 	DropOriginal bool
 	Period       time.Duration
 	Delay        time.Duration
+	Grace        time.Duration
 
 	NameOverride      string
 	MeasurementPrefix string
@@ -135,9 +136,9 @@ func (r *RunningAggregator) Add(m telegraf.Metric) bool {
 	r.Lock()
 	defer r.Unlock()
 
-	if m.Time().Before(r.periodStart) || m.Time().After(r.periodEnd.Add(r.Config.Delay)) {
-		log.Printf("D! [%s] metric is outside aggregation window; discarding. %s: m: %s e: %s",
-			r.Name(), m.Time(), r.periodStart, r.periodEnd)
+	if m.Time().Before(r.periodStart.Add(-r.Config.Grace)) || m.Time().After(r.periodEnd.Add(r.Config.Delay)) {
+		log.Printf("D! [%s] metric is outside aggregation window; discarding. %s: m: %s e: %s g: %s",
+			r.Name(), m.Time(), r.periodStart, r.periodEnd, r.Config.Grace)
 		r.MetricsDropped.Incr(1)
 		return r.Config.DropOriginal
 	}

internal/models/running_aggregator_test.go

@@ -89,6 +89,68 @@ func TestAddMetricsOutsideCurrentPeriod(t *testing.T) {
 	require.Equal(t, int64(101), acc.Metrics[0].Fields["sum"])
 }
 
+func TestAddMetricsOutsideCurrentPeriodWithGrace(t *testing.T) {
+	a := &TestAggregator{}
+	ra := NewRunningAggregator(a, &AggregatorConfig{
+		Name: "TestRunningAggregator",
+		Filter: Filter{
+			NamePass: []string{"*"},
+		},
+		Period: time.Millisecond * 1500,
+		Grace:  time.Millisecond * 500,
+	})
+	require.NoError(t, ra.Config.Filter.Compile())
+	acc := testutil.Accumulator{}
+	now := time.Now()
+	ra.UpdateWindow(now, now.Add(ra.Config.Period))
+
+	m := testutil.MustMetric("RITest",
+		map[string]string{},
+		map[string]interface{}{
+			"value": int64(101),
+		},
+		now.Add(-time.Hour),
+		telegraf.Untyped,
+	)
+	require.False(t, ra.Add(m))
+
+	// metric before current period (late)
+	m = testutil.MustMetric("RITest",
+		map[string]string{},
+		map[string]interface{}{
+			"value": int64(100),
+		},
+		now.Add(-time.Millisecond*1000),
+		telegraf.Untyped,
+	)
+	require.False(t, ra.Add(m))
+
+	// metric before current period, but within grace period (late)
+	m = testutil.MustMetric("RITest",
+		map[string]string{},
+		map[string]interface{}{
+			"value": int64(102),
+		},
+		now.Add(-time.Millisecond*200),
+		telegraf.Untyped,
+	)
+	require.False(t, ra.Add(m))
+
+	// "now" metric
+	m = testutil.MustMetric("RITest",
+		map[string]string{},
+		map[string]interface{}{
+			"value": int64(101),
+		},
+		time.Now().Add(time.Millisecond*50),
+		telegraf.Untyped)
+	require.False(t, ra.Add(m))
+
+	ra.Push(&acc)
+	require.Equal(t, 1, len(acc.Metrics))
+	require.Equal(t, int64(203), acc.Metrics[0].Fields["sum"])
+}
+
 func TestAddAndPushOnePeriod(t *testing.T) {
 	a := &TestAggregator{}
 	ra := NewRunningAggregator(a, &AggregatorConfig{