Beispiel #1
0
// eval evaluates the rule expression and then creates pending alerts and fires
// or removes previously pending alerts accordingly.
func (rule *AlertingRule) eval(timestamp clientmodel.Timestamp, engine *promql.Engine) (promql.Vector, error) {
	query, err := engine.NewInstantQuery(rule.vector.String(), timestamp)
	if err != nil {
		return nil, err
	}
	exprResult, err := query.Exec().Vector()
	if err != nil {
		return nil, err
	}

	rule.mutex.Lock()
	defer rule.mutex.Unlock()

	// Create pending alerts for any new vector elements in the alert expression
	// or update the expression value for existing elements.
	resultFPs := map[clientmodel.Fingerprint]struct{}{}
	for _, sample := range exprResult {
		fp := sample.Metric.Metric.Fingerprint()
		resultFPs[fp] = struct{}{}

		if alert, ok := rule.activeAlerts[fp]; !ok {
			labels := clientmodel.LabelSet{}
			labels.MergeFromMetric(sample.Metric.Metric)
			labels = labels.Merge(rule.labels)
			if _, ok := labels[clientmodel.MetricNameLabel]; ok {
				delete(labels, clientmodel.MetricNameLabel)
			}
			rule.activeAlerts[fp] = &Alert{
				Name:        rule.name,
				Labels:      labels,
				State:       StatePending,
				ActiveSince: timestamp,
				Value:       sample.Value,
			}
		} else {
			alert.Value = sample.Value
		}
	}

	vector := promql.Vector{}

	// Check if any pending alerts should be removed or fire now. Write out alert timeseries.
	for fp, activeAlert := range rule.activeAlerts {
		if _, ok := resultFPs[fp]; !ok {
			vector = append(vector, activeAlert.sample(timestamp, 0))
			delete(rule.activeAlerts, fp)
			continue
		}

		if activeAlert.State == StatePending && timestamp.Sub(activeAlert.ActiveSince) >= rule.holdDuration {
			vector = append(vector, activeAlert.sample(timestamp, 0))
			activeAlert.State = StateFiring
		}

		vector = append(vector, activeAlert.sample(timestamp, 1))
	}

	return vector, nil
}
Beispiel #2
0
// Eval evaluates the rule and then overrides the metric names and labels accordingly.
func (rule RecordingRule) Eval(ctx context.Context, timestamp model.Time, engine *promql.Engine, _ string) (model.Vector, error) {
	query, err := engine.NewInstantQuery(rule.vector.String(), timestamp)
	if err != nil {
		return nil, err
	}

	var (
		result = query.Exec(ctx)
		vector model.Vector
	)
	if result.Err != nil {
		return nil, err
	}

	switch result.Value.(type) {
	case model.Vector:
		vector, err = result.Vector()
		if err != nil {
			return nil, err
		}
	case *model.Scalar:
		scalar, err := result.Scalar()
		if err != nil {
			return nil, err
		}
		vector = model.Vector{&model.Sample{
			Value:     scalar.Value,
			Timestamp: scalar.Timestamp,
			Metric:    model.Metric{},
		}}
	default:
		return nil, fmt.Errorf("rule result is not a vector or scalar")
	}

	// Override the metric name and labels.
	for _, sample := range vector {
		sample.Metric[model.MetricNameLabel] = model.LabelValue(rule.name)

		for label, value := range rule.labels {
			if value == "" {
				delete(sample.Metric, label)
			} else {
				sample.Metric[label] = value
			}
		}
	}

	return vector, nil
}
Beispiel #3
0
func query(q string, timestamp clientmodel.Timestamp, queryEngine *promql.Engine) (queryResult, error) {
	query, err := queryEngine.NewInstantQuery(q, timestamp)
	if err != nil {
		return nil, err
	}
	res := query.Exec()
	if res.Err != nil {
		return nil, res.Err
	}
	var vector promql.Vector

	switch v := res.Value.(type) {
	case promql.Matrix:
		return nil, errors.New("matrix return values not supported")
	case promql.Vector:
		vector = v
	case *promql.Scalar:
		vector = promql.Vector{&promql.Sample{
			Value:     v.Value,
			Timestamp: v.Timestamp,
		}}
	case *promql.String:
		vector = promql.Vector{&promql.Sample{
			Metric: clientmodel.COWMetric{
				Metric: clientmodel.Metric{"__value__": clientmodel.LabelValue(v.Value)},
				Copied: true,
			},
			Timestamp: v.Timestamp,
		}}
	default:
		panic("template.query: unhandled result value type")
	}

	// promql.Vector is hard to work with in templates, so convert to
	// base data types.
	var result = make(queryResult, len(vector))
	for n, v := range vector {
		s := sample{
			Value:  float64(v.Value),
			Labels: make(map[string]string),
		}
		for label, value := range v.Metric.Metric {
			s.Labels[string(label)] = string(value)
		}
		result[n] = &s
	}
	return result, nil
}
Beispiel #4
0
// eval evaluates the rule and then overrides the metric names and labels accordingly.
func (rule RecordingRule) eval(timestamp model.Time, engine *promql.Engine) (promql.Vector, error) {
	query, err := engine.NewInstantQuery(rule.vector.String(), timestamp)
	if err != nil {
		return nil, err
	}

	result := query.Exec()
	var vector promql.Vector
	switch result.Value.(type) {
	case promql.Vector:
		vector, err = result.Vector()
		if err != nil {
			return nil, err
		}
	case *promql.Scalar:
		scalar, err := result.Scalar()
		if err != nil {
			return nil, err
		}
		vector = promql.Vector{&promql.Sample{Value: scalar.Value, Timestamp: scalar.Timestamp}}
	default:
		return nil, fmt.Errorf("rule result is not a vector or scalar")
	}

	// Override the metric name and labels.
	for _, sample := range vector {
		sample.Metric.Set(model.MetricNameLabel, model.LabelValue(rule.name))
		for label, value := range rule.labels {
			if value == "" {
				sample.Metric.Del(label)
			} else {
				sample.Metric.Set(label, value)
			}
		}
	}

	return vector, nil
}
Beispiel #5
0
// eval evaluates the rule and then overrides the metric names and labels accordingly.
func (rule RecordingRule) eval(timestamp clientmodel.Timestamp, engine *promql.Engine) (promql.Vector, error) {
	query, err := engine.NewInstantQuery(rule.vector.String(), timestamp)
	if err != nil {
		return nil, err
	}
	vector, err := query.Exec().Vector()
	if err != nil {
		return nil, err
	}

	// Override the metric name and labels.
	for _, sample := range vector {
		sample.Metric.Set(clientmodel.MetricNameLabel, clientmodel.LabelValue(rule.name))
		for label, value := range rule.labels {
			if value == "" {
				sample.Metric.Delete(label)
			} else {
				sample.Metric.Set(label, value)
			}
		}
	}

	return vector, nil
}
Beispiel #6
0
// eval evaluates the rule expression and then creates pending alerts and fires
// or removes previously pending alerts accordingly.
func (r *AlertingRule) eval(ts model.Time, engine *promql.Engine) (model.Vector, error) {
	query, err := engine.NewInstantQuery(r.vector.String(), ts)
	if err != nil {
		return nil, err
	}
	res, err := query.Exec().Vector()
	if err != nil {
		return nil, err
	}

	r.mtx.Lock()
	defer r.mtx.Unlock()

	// Create pending alerts for any new vector elements in the alert expression
	// or update the expression value for existing elements.
	resultFPs := map[model.Fingerprint]struct{}{}

	for _, smpl := range res {
		fp := smpl.Metric.Fingerprint()
		resultFPs[fp] = struct{}{}

		if alert, ok := r.active[fp]; ok {
			alert.Value = smpl.Value
			continue
		}

		delete(smpl.Metric, model.MetricNameLabel)

		r.active[fp] = &Alert{
			Labels:   model.LabelSet(smpl.Metric),
			ActiveAt: ts,
			State:    StatePending,
			Value:    smpl.Value,
		}
	}

	var vec model.Vector
	// Check if any pending alerts should be removed or fire now. Write out alert timeseries.
	for fp, a := range r.active {
		if _, ok := resultFPs[fp]; !ok {
			if a.State != StateInactive {
				vec = append(vec, r.sample(a, ts, false))
			}
			// If the alert was previously firing, keep it around for a given
			// retention time so it is reported as resolved to the AlertManager.
			if a.State == StatePending || (a.ResolvedAt != 0 && ts.Sub(a.ResolvedAt) > resolvedRetention) {
				delete(r.active, fp)
			}
			if a.State != StateInactive {
				a.State = StateInactive
				a.ResolvedAt = ts
			}
			continue
		}

		if a.State == StatePending && ts.Sub(a.ActiveAt) >= r.holdDuration {
			vec = append(vec, r.sample(a, ts, false))
			a.State = StateFiring
		}

		vec = append(vec, r.sample(a, ts, true))
	}

	return vec, nil
}
Beispiel #7
0
// eval evaluates the rule expression and then creates pending alerts and fires
// or removes previously pending alerts accordingly.
func (r *AlertingRule) eval(ts model.Time, engine *promql.Engine, externalURLPath string) (model.Vector, error) {
	query, err := engine.NewInstantQuery(r.vector.String(), ts)
	if err != nil {
		return nil, err
	}
	res, err := query.Exec().Vector()
	if err != nil {
		return nil, err
	}

	r.mtx.Lock()
	defer r.mtx.Unlock()

	// Create pending alerts for any new vector elements in the alert expression
	// or update the expression value for existing elements.
	resultFPs := map[model.Fingerprint]struct{}{}

	for _, smpl := range res {
		// Provide the alert information to the template.
		l := make(map[string]string, len(smpl.Metric))
		for k, v := range smpl.Metric {
			l[string(k)] = string(v)
		}

		tmplData := struct {
			Labels map[string]string
			Value  float64
		}{
			Labels: l,
			Value:  float64(smpl.Value),
		}
		// Inject some convenience variables that are easier to remember for users
		// who are not used to Go's templating system.
		defs := "{{$labels := .Labels}}{{$value := .Value}}"

		expand := func(text model.LabelValue) model.LabelValue {
			tmpl := template.NewTemplateExpander(
				defs+string(text),
				"__alert_"+r.Name(),
				tmplData,
				ts,
				engine,
				externalURLPath,
			)
			result, err := tmpl.Expand()
			if err != nil {
				result = fmt.Sprintf("<error expanding template: %s>", err)
				log.Warnf("Error expanding alert template %v with data '%v': %s", r.Name(), tmplData, err)
			}
			return model.LabelValue(result)
		}

		labels := make(model.LabelSet, len(smpl.Metric)+len(r.labels)+1)
		for ln, lv := range smpl.Metric {
			labels[ln] = lv
		}
		for ln, lv := range r.labels {
			labels[ln] = expand(lv)
		}
		labels[model.AlertNameLabel] = model.LabelValue(r.Name())

		annotations := make(model.LabelSet, len(r.annotations))
		for an, av := range r.annotations {
			annotations[an] = expand(av)
		}
		fp := smpl.Metric.Fingerprint()
		resultFPs[fp] = struct{}{}

		if alert, ok := r.active[fp]; ok && alert.State != StateInactive {
			alert.Value = smpl.Value
			continue
		}

		delete(smpl.Metric, model.MetricNameLabel)

		r.active[fp] = &Alert{
			Labels:      labels,
			Annotations: annotations,
			ActiveAt:    ts,
			State:       StatePending,
			Value:       smpl.Value,
		}
	}

	var vec model.Vector
	// Check if any pending alerts should be removed or fire now. Write out alert timeseries.
	for fp, a := range r.active {
		if _, ok := resultFPs[fp]; !ok {
			if a.State != StateInactive {
				vec = append(vec, r.sample(a, ts, false))
			}
			// If the alert was previously firing, keep it around for a given
			// retention time so it is reported as resolved to the AlertManager.
			if a.State == StatePending || (a.ResolvedAt != 0 && ts.Sub(a.ResolvedAt) > resolvedRetention) {
				delete(r.active, fp)
			}
			if a.State != StateInactive {
				a.State = StateInactive
				a.ResolvedAt = ts
			}
			continue
		}

		if a.State == StatePending && ts.Sub(a.ActiveAt) >= r.holdDuration {
			vec = append(vec, r.sample(a, ts, false))
			a.State = StateFiring
		}

		vec = append(vec, r.sample(a, ts, true))
	}

	return vec, nil
}