Skip samples that have a NaN value. (#186)

retrieval/transform.go

@@ -40,6 +40,10 @@ func (b *sampleBuilder) next(ctx context.Context, samples []tsdb.RefSample) (*mo
 	sample := samples[0]
 	tailSamples := samples[1:]
 
+	if (math.IsNaN(sample.V)) {
+		return nil, 0, tailSamples, nil
+	}
+
 	entry, ok, err := b.series.get(ctx, sample.Ref)
 	if err != nil {
 		return nil, 0, samples, errors.Wrap(err, "get series information")

retrieval/transform_test.go

@@ -15,6 +15,7 @@ package retrieval
 
 import (
 	"context"
+	"math"
 	"reflect"
 	"testing"
 
@@ -760,6 +761,80 @@ func TestSampleBuilder(t *testing.T) {
 				},
 			},
 		},
+		// Samples with a NaN value should be dropped.
+		{
+			targets: targetMap{
+				"job1/instance1": &targets.Target{
+					Labels:           promlabels.FromStrings("job", "job1", "instance", "instance1"),
+					DiscoveredLabels: promlabels.FromStrings("__resource_a", "resource2_a"),
+				},
+			},
+			series: seriesMap{
+				1: labels.FromStrings("job", "job1", "instance", "instance1", "__name__", "metric1_count"),
+			},
+			metadata: metadataMap{
+				"job1/instance1/metric1": &scrape.MetricMetadata{Type: textparse.MetricTypeSummary, Metric: "metric1_count"},
+			},
+			metricPrefix: "test.googleapis.com",
+			input: []tsdb.RefSample{
+				// A first non-NaN sample is necessary to avoid false-positives, since the
+				// first result will always be nil due to reset timestamp handling.
+				{Ref: 1, T: 2000, V: 5},
+				{Ref: 1, T: 4000, V: math.NaN()},
+			},
+			result: []*monitoring_pb.TimeSeries{
+				nil, // due to reset timestamp handling
+				nil, // due to NaN
+			},
+		},
+		// Samples with a NaN value should be dropped.
+		{
+			targets: targetMap{
+				"job1/instance1": &targets.Target{
+					Labels:           promlabels.FromStrings("job", "job1", "instance", "instance1"),
+					DiscoveredLabels: promlabels.FromStrings("__resource_a", "resource2_a"),
+				},
+			},
+			series: seriesMap{
+				1: labels.FromStrings("job", "job1", "instance", "instance1", "__name__", "metric1_count"),
+			},
+			metadata: metadataMap{
+				"job1/instance1/metric1": &scrape.MetricMetadata{Type: textparse.MetricTypeSummary, Metric: "metric1_count"},
+			},
+			metricPrefix: "test.googleapis.com",
+			input: []tsdb.RefSample{
+				// A first non-NaN sample is necessary to avoid false-positives, since the
+				// first result will always be nil due to reset timestamp handling.
+				{Ref: 1, T: 2000, V: 5},
+				{Ref: 1, T: 4000, V: math.NaN()},
+				{Ref: 1, T: 5000, V: 9},
+			},
+			result: []*monitoring_pb.TimeSeries{
+				nil, // due to reset timestamp handling
+				nil, // due to NaN
+				{
+					Resource: &monitoredres_pb.MonitoredResource{
+						Type:   "resource2",
+						Labels: map[string]string{"resource_a": "resource2_a"},
+					},
+					Metric: &metric_pb.Metric{
+						Type:   "test.googleapis.com/metric1_count",
+						Labels: map[string]string{},
+					},
+					MetricKind: metric_pb.MetricDescriptor_CUMULATIVE,
+					ValueType:  metric_pb.MetricDescriptor_INT64,
+					Points: []*monitoring_pb.Point{{
+						Interval: &monitoring_pb.TimeInterval{
+							StartTime: &timestamp_pb.Timestamp{Seconds: 2},
+							EndTime:   &timestamp_pb.Timestamp{Seconds: 5},
+						},
+						Value: &monitoring_pb.TypedValue{
+							Value: &monitoring_pb.TypedValue_Int64Value{4},
+						},
+					}},
+				},
+			},
+		},
 	}
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()