Golang model.SampleValue函数代码示例

// === changes(matrix ExprMatrix) Vector ===
func funcChanges(ev *evaluator, args Expressions) Value {
	in := ev.evalMatrix(args[0])
	out := make(Vector, 0, len(in))

	for _, samples := range in {
		changes := 0
		prev := clientmodel.SampleValue(samples.Values[0].Value)
		for _, sample := range samples.Values[1:] {
			current := sample.Value
			if current != prev {
				changes++
			}
			prev = current
		}

		rs := &Sample{
			Metric:    samples.Metric,
			Value:     clientmodel.SampleValue(changes),
			Timestamp: ev.Timestamp,
		}
		rs.Metric.Delete(clientmodel.MetricNameLabel)
		out = append(out, rs)
	}
	return out
}

// scalarBinop evaluates a binary operation between two scalars.
func scalarBinop(op itemType, lhs, rhs clientmodel.SampleValue) clientmodel.SampleValue {
	switch op {
	case itemADD:
		return lhs + rhs
	case itemSUB:
		return lhs - rhs
	case itemMUL:
		return lhs * rhs
	case itemDIV:
		return lhs / rhs
	case itemMOD:
		if rhs != 0 {
			return clientmodel.SampleValue(int(lhs) % int(rhs))
		}
		return clientmodel.SampleValue(math.NaN())
	case itemEQL:
		return btos(lhs == rhs)
	case itemNEQ:
		return btos(lhs != rhs)
	case itemGTR:
		return btos(lhs > rhs)
	case itemLSS:
		return btos(lhs < rhs)
	case itemGTE:
		return btos(lhs >= rhs)
	case itemLTE:
		return btos(lhs <= rhs)
	}
	panic(fmt.Errorf("operator %q not allowed for scalar operations", op))
}

// === predict_linear(node ExprMatrix, k ExprScalar) Vector ===
func funcPredictLinear(ev *evaluator, args Expressions) Value {
	vector := funcDeriv(ev, args[0:1]).(Vector)
	duration := clientmodel.SampleValue(clientmodel.SampleValue(ev.evalFloat(args[1])))

	excludedLabels := map[clientmodel.LabelName]struct{}{
		clientmodel.MetricNameLabel: {},
	}

	// Calculate predicted delta over the duration.
	signatureToDelta := map[uint64]clientmodel.SampleValue{}
	for _, el := range vector {
		signature := clientmodel.SignatureWithoutLabels(el.Metric.Metric, excludedLabels)
		signatureToDelta[signature] = el.Value * duration
	}

	// add predicted delta to last value.
	matrixBounds := ev.evalMatrixBounds(args[0])
	outVec := make(Vector, 0, len(signatureToDelta))
	for _, samples := range matrixBounds {
		if len(samples.Values) < 2 {
			continue
		}
		signature := clientmodel.SignatureWithoutLabels(samples.Metric.Metric, excludedLabels)
		delta, ok := signatureToDelta[signature]
		if ok {
			samples.Metric.Delete(clientmodel.MetricNameLabel)
			outVec = append(outVec, &Sample{
				Metric:    samples.Metric,
				Value:     delta + samples.Values[1].Value,
				Timestamp: ev.Timestamp,
			})
		}
	}
	return outVec
}

// === scalar(node VectorNode) Scalar ===
func scalarImpl(timestamp clientmodel.Timestamp, args []Node) interface{} {
	v := args[0].(VectorNode).Eval(timestamp)
	if len(v) != 1 {
		return clientmodel.SampleValue(math.NaN())
	}
	return clientmodel.SampleValue(v[0].Value)
}

// vectorElemBinop evaluates a binary operation between two vector elements.
func vectorElemBinop(op itemType, lhs, rhs clientmodel.SampleValue) (clientmodel.SampleValue, bool) {
	switch op {
	case itemADD:
		return lhs + rhs, true
	case itemSUB:
		return lhs - rhs, true
	case itemMUL:
		return lhs * rhs, true
	case itemDIV:
		return lhs / rhs, true
	case itemMOD:
		if rhs != 0 {
			return clientmodel.SampleValue(int(lhs) % int(rhs)), true
		}
		return clientmodel.SampleValue(math.NaN()), true
	case itemEQL:
		return lhs, lhs == rhs
	case itemNEQ:
		return lhs, lhs != rhs
	case itemGTR:
		return lhs, lhs > rhs
	case itemLSS:
		return lhs, lhs < rhs
	case itemGTE:
		return lhs, lhs >= rhs
	case itemLTE:
		return lhs, lhs <= rhs
	}
	panic(fmt.Errorf("operator %q not allowed for operations between vectors", op))
}

func (t *target) recordScrapeHealth(ingester extraction.Ingester, timestamp clientmodel.Timestamp, healthy bool, scrapeDuration time.Duration) {
	healthMetric := clientmodel.Metric{}
	durationMetric := clientmodel.Metric{}
	for label, value := range t.baseLabels {
		healthMetric[label] = value
		durationMetric[label] = value
	}
	healthMetric[clientmodel.MetricNameLabel] = clientmodel.LabelValue(scrapeHealthMetricName)
	durationMetric[clientmodel.MetricNameLabel] = clientmodel.LabelValue(scrapeDurationMetricName)
	healthMetric[InstanceLabel] = clientmodel.LabelValue(t.URL())
	durationMetric[InstanceLabel] = clientmodel.LabelValue(t.URL())

	healthValue := clientmodel.SampleValue(0)
	if healthy {
		healthValue = clientmodel.SampleValue(1)
	}

	healthSample := &clientmodel.Sample{
		Metric:    healthMetric,
		Timestamp: timestamp,
		Value:     healthValue,
	}
	durationSample := &clientmodel.Sample{
		Metric:    durationMetric,
		Timestamp: timestamp,
		Value:     clientmodel.SampleValue(float64(scrapeDuration) / float64(time.Second)),
	}

	ingester.Ingest(clientmodel.Samples{healthSample, durationSample})
}

// === scalar(node ExprVector) Scalar ===
func funcScalar(ev *evaluator, args Expressions) Value {
	v := ev.evalVector(args[0])
	if len(v) != 1 {
		return &Scalar{clientmodel.SampleValue(math.NaN()), ev.Timestamp}
	}
	return &Scalar{clientmodel.SampleValue(v[0].Value), ev.Timestamp}
}

// sampleValueAtIndex implements chunkIterator.
func (it *deltaEncodedChunkIterator) sampleValueAtIndex(idx int) clientmodel.SampleValue {
	offset := deltaHeaderBytes + idx*int(it.tBytes+it.vBytes) + int(it.tBytes)

	if it.isInt {
		switch it.vBytes {
		case d0:
			return it.baseV
		case d1:
			return it.baseV + clientmodel.SampleValue(int8(it.c[offset]))
		case d2:
			return it.baseV + clientmodel.SampleValue(int16(binary.LittleEndian.Uint16(it.c[offset:])))
		case d4:
			return it.baseV + clientmodel.SampleValue(int32(binary.LittleEndian.Uint32(it.c[offset:])))
		// No d8 for ints.
		default:
			panic("Invalid number of bytes for integer delta")
		}
	} else {
		switch it.vBytes {
		case d4:
			return it.baseV + clientmodel.SampleValue(math.Float32frombits(binary.LittleEndian.Uint32(it.c[offset:])))
		case d8:
			// Take absolute value for d8.
			return clientmodel.SampleValue(math.Float64frombits(binary.LittleEndian.Uint64(it.c[offset:])))
		default:
			panic("Invalid number of bytes for floating point delta")
		}
	}
}

func AppendRepeatingValuesTests(p metric.Persistence, t test.Tester) {
	m := clientmodel.Metric{
		clientmodel.MetricNameLabel: "errors_total",
		"controller":                "foo",
		"operation":                 "bar",
	}

	increments := 10
	repetitions := 500

	for i := 0; i < increments; i++ {
		for j := 0; j < repetitions; j++ {
			time := clientmodel.Timestamp(0).Add(time.Duration(i) * time.Hour).Add(time.Duration(j) * time.Second)
			testAppendSamples(p, &clientmodel.Sample{
				Value:     clientmodel.SampleValue(i),
				Timestamp: time,
				Metric:    m,
			}, t)
		}
	}

	v, ok := p.(metric.View)
	if !ok {
		// It's purely a benchmark for a Persistence that is not viewable.
		return
	}

	matchers := labelMatchersFromLabelSet(clientmodel.LabelSet{
		clientmodel.MetricNameLabel: "errors_total",
		"controller":                "foo",
		"operation":                 "bar",
	})

	for i := 0; i < increments; i++ {
		for j := 0; j < repetitions; j++ {
			fingerprints, err := p.GetFingerprintsForLabelMatchers(matchers)
			if err != nil {
				t.Fatal(err)
			}
			if len(fingerprints) != 1 {
				t.Fatalf("expected %d fingerprints, got %d", 1, len(fingerprints))
			}

			time := clientmodel.Timestamp(0).Add(time.Duration(i) * time.Hour).Add(time.Duration(j) * time.Second)
			samples := v.GetValueAtTime(fingerprints[0], time)
			if len(samples) == 0 {
				t.Fatal("expected at least one sample.")
			}

			expected := clientmodel.SampleValue(i)

			for _, sample := range samples {
				if sample.Value != expected {
					t.Fatalf("expected %v value, got %v", expected, sample.Value)
				}
			}
		}
	}
}

// === delta(matrix MatrixNode, isCounter=0 ScalarNode) Vector ===
func deltaImpl(timestamp clientmodel.Timestamp, args []Node) interface{} {
	matrixNode := args[0].(MatrixNode)
	isCounter := len(args) >= 2 && args[1].(ScalarNode).Eval(timestamp) > 0
	resultVector := Vector{}

	// If we treat these metrics as counters, we need to fetch all values
	// in the interval to find breaks in the timeseries' monotonicity.
	// I.e. if a counter resets, we want to ignore that reset.
	var matrixValue Matrix
	if isCounter {
		matrixValue = matrixNode.Eval(timestamp)
	} else {
		matrixValue = matrixNode.EvalBoundaries(timestamp)
	}
	for _, samples := range matrixValue {
		// No sense in trying to compute a delta without at least two points. Drop
		// this vector element.
		if len(samples.Values) < 2 {
			continue
		}

		counterCorrection := clientmodel.SampleValue(0)
		lastValue := clientmodel.SampleValue(0)
		for _, sample := range samples.Values {
			currentValue := sample.Value
			if isCounter && currentValue < lastValue {
				counterCorrection += lastValue - currentValue
			}
			lastValue = currentValue
		}
		resultValue := lastValue - samples.Values[0].Value + counterCorrection

		targetInterval := args[0].(*MatrixSelector).interval
		sampledInterval := samples.Values[len(samples.Values)-1].Timestamp.Sub(samples.Values[0].Timestamp)
		if sampledInterval == 0 {
			// Only found one sample. Cannot compute a rate from this.
			continue
		}
		// Correct for differences in target vs. actual delta interval.
		//
		// Above, we didn't actually calculate the delta for the specified target
		// interval, but for an interval between the first and last found samples
		// under the target interval, which will usually have less time between
		// them. Depending on how many samples are found under a target interval,
		// the delta results are distorted and temporal aliasing occurs (ugly
		// bumps). This effect is corrected for below.
		intervalCorrection := clientmodel.SampleValue(targetInterval) / clientmodel.SampleValue(sampledInterval)
		resultValue *= intervalCorrection

		resultSample := &Sample{
			Metric:    samples.Metric,
			Value:     resultValue,
			Timestamp: timestamp,
		}
		resultSample.Metric.Delete(clientmodel.MetricNameLabel)
		resultVector = append(resultVector, resultSample)
	}
	return resultVector
}

func evalVectorBinop(opType BinOpType,
	lhs clientmodel.SampleValue,
	rhs clientmodel.SampleValue) (clientmodel.SampleValue, bool) {
	switch opType {
	case ADD:
		return lhs + rhs, true
	case SUB:
		return lhs - rhs, true
	case MUL:
		return lhs * rhs, true
	case DIV:
		if rhs != 0 {
			return lhs / rhs, true
		}
		return clientmodel.SampleValue(math.Inf(int(rhs))), true
	case MOD:
		if rhs != 0 {
			return clientmodel.SampleValue(int(lhs) % int(rhs)), true
		}
		return clientmodel.SampleValue(math.Inf(int(rhs))), true
	case EQ:
		if lhs == rhs {
			return lhs, true
		}
		return 0, false
	case NE:
		if lhs != rhs {
			return lhs, true
		}
		return 0, false
	case GT:
		if lhs > rhs {
			return lhs, true
		}
		return 0, false
	case LT:
		if lhs < rhs {
			return lhs, true
		}
		return 0, false
	case GE:
		if lhs >= rhs {
			return lhs, true
		}
		return 0, false
	case LE:
		if lhs <= rhs {
			return lhs, true
		}
		return 0, false
	case AND:
		return lhs, true
	case OR:
		return lhs, true // TODO: implement OR
	}
	panic("Not all enum values enumerated in switch")
}

// interpolateSamples interpolates a value at a target time between two
// provided sample pairs.
func interpolateSamples(first, second *metric.SamplePair, timestamp clientmodel.Timestamp) *metric.SamplePair {
	dv := second.Value - first.Value
	dt := second.Timestamp.Sub(first.Timestamp)

	dDt := dv / clientmodel.SampleValue(dt)
	offset := clientmodel.SampleValue(timestamp.Sub(first.Timestamp))

	return &metric.SamplePair{
		Value:     first.Value + (offset * dDt),
		Timestamp: timestamp,
	}
}

func evalScalarBinop(opType BinOpType,
	lhs clientmodel.SampleValue,
	rhs clientmodel.SampleValue) clientmodel.SampleValue {
	switch opType {
	case ADD:
		return lhs + rhs
	case SUB:
		return lhs - rhs
	case MUL:
		return lhs * rhs
	case DIV:
		if rhs != 0 {
			return lhs / rhs
		}
		return clientmodel.SampleValue(math.Inf(int(rhs)))
	case MOD:
		if rhs != 0 {
			return clientmodel.SampleValue(int(lhs) % int(rhs))
		}
		return clientmodel.SampleValue(math.Inf(int(rhs)))
	case EQ:
		if lhs == rhs {
			return 1
		}
		return 0
	case NE:
		if lhs != rhs {
			return 1
		}
		return 0
	case GT:
		if lhs > rhs {
			return 1
		}
		return 0
	case LT:
		if lhs < rhs {
			return 1
		}
		return 0
	case GE:
		if lhs >= rhs {
			return 1
		}
		return 0
	case LE:
		if lhs <= rhs {
			return 1
		}
		return 0
	}
	panic("Not all enum values enumerated in switch")
}

func (c *deltaEncodedChunk) valueAtIndex(idx int) *metric.SamplePair {
	offset := deltaHeaderBytes + idx*c.sampleSize()

	var ts clientmodel.Timestamp
	switch c.timeBytes() {
	case d1:
		ts = c.baseTime() + clientmodel.Timestamp(uint8(c.buf[offset]))
	case d2:
		ts = c.baseTime() + clientmodel.Timestamp(binary.LittleEndian.Uint16(c.buf[offset:]))
	case d4:
		ts = c.baseTime() + clientmodel.Timestamp(binary.LittleEndian.Uint32(c.buf[offset:]))
	case d8:
		// Take absolute value for d8.
		ts = clientmodel.Timestamp(binary.LittleEndian.Uint64(c.buf[offset:]))
	default:
		panic("Invalid number of bytes for time delta")
	}

	offset += int(c.timeBytes())

	var v clientmodel.SampleValue
	if c.isInt() {
		switch c.valueBytes() {
		case d0:
			v = c.baseValue()
		case d1:
			v = c.baseValue() + clientmodel.SampleValue(int8(c.buf[offset]))
		case d2:
			v = c.baseValue() + clientmodel.SampleValue(int16(binary.LittleEndian.Uint16(c.buf[offset:])))
		case d4:
			v = c.baseValue() + clientmodel.SampleValue(int32(binary.LittleEndian.Uint32(c.buf[offset:])))
		// No d8 for ints.
		default:
			panic("Invalid number of bytes for integer delta")
		}
	} else {
		switch c.valueBytes() {
		case d4:
			v = c.baseValue() + clientmodel.SampleValue(math.Float32frombits(binary.LittleEndian.Uint32(c.buf[offset:])))
		case d8:
			// Take absolute value for d8.
			v = clientmodel.SampleValue(math.Float64frombits(binary.LittleEndian.Uint64(c.buf[offset:])))
		default:
			panic("Invalid number of bytes for floating point delta")
		}
	}
	return &metric.SamplePair{
		Timestamp: ts,
		Value:     v,
	}
}

// sampleValueAtIndex implements chunkIterator.
func (it *doubleDeltaEncodedChunkIterator) sampleValueAtIndex(idx int) clientmodel.SampleValue {
	if idx == 0 {
		return it.baseV
	}
	if idx == 1 {
		// If value bytes are at d8, the value is saved directly rather
		// than as a difference.
		if it.vBytes == d8 {
			return it.baseΔV
		}
		return it.baseV + it.baseΔV
	}

	offset := doubleDeltaHeaderBytes + (idx-2)*int(it.tBytes+it.vBytes) + int(it.tBytes)

	if it.isInt {
		switch it.vBytes {
		case d0:
			return it.baseV +
				clientmodel.SampleValue(idx)*it.baseΔV
		case d1:
			return it.baseV +
				clientmodel.SampleValue(idx)*it.baseΔV +
				clientmodel.SampleValue(int8(it.c[offset]))
		case d2:
			return it.baseV +
				clientmodel.SampleValue(idx)*it.baseΔV +
				clientmodel.SampleValue(int16(binary.LittleEndian.Uint16(it.c[offset:])))
		case d4:
			return it.baseV +
				clientmodel.SampleValue(idx)*it.baseΔV +
				clientmodel.SampleValue(int32(binary.LittleEndian.Uint32(it.c[offset:])))
		// No d8 for ints.
		default:
			panic("Invalid number of bytes for integer delta")
		}
	} else {
		switch it.vBytes {
		case d4:
			return it.baseV +
				clientmodel.SampleValue(idx)*it.baseΔV +
				clientmodel.SampleValue(math.Float32frombits(binary.LittleEndian.Uint32(it.c[offset:])))
		case d8:
			// Take absolute value for d8.
			return clientmodel.SampleValue(math.Float64frombits(binary.LittleEndian.Uint64(it.c[offset:])))
		default:
			panic("Invalid number of bytes for floating point delta")
		}
	}
}