stats_reporter.go

/*
Copyright 2018 The Knative Authors

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/

package autoscaler

import (
	"context"
	"errors"

	"knative.dev/pkg/metrics"
	"knative.dev/pkg/metrics/metricskey"

	"go.opencensus.io/stats"
	"go.opencensus.io/stats/view"
	"go.opencensus.io/tag"
)

var (
	desiredPodCountM = stats.Int64(
		"desired_pods",
		"Number of pods autoscaler wants to allocate",
		stats.UnitDimensionless)
	requestedPodCountM = stats.Int64(
		"requested_pods",
		"Number of pods autoscaler requested from Kubernetes",
		stats.UnitDimensionless)
	actualPodCountM = stats.Int64(
		"actual_pods",
		"Number of pods that are allocated currently",
		stats.UnitDimensionless)
	excessBurstCapacityM = stats.Float64(
		"excess_burst_capacity",
		"Excess burst capacity overserved over the stable window",
		stats.UnitDimensionless)
	stableRequestConcurrencyM = stats.Float64(
		"stable_request_concurrency",
		"Average of requests count per observed pod over the stable window",
		stats.UnitDimensionless)
	panicRequestConcurrencyM = stats.Float64(
		"panic_request_concurrency",
		"Average of requests count per observed pod over the panic window",
		stats.UnitDimensionless)
	targetRequestConcurrencyM = stats.Float64(
		"target_concurrency_per_pod",
		"The desired number of concurrent requests for each pod",
		stats.UnitDimensionless)
	panicM = stats.Int64(
		"panic_mode",
		"1 if autoscaler is in panic mode, 0 otherwise",
		stats.UnitDimensionless)
	namespaceTagKey tag.Key
	configTagKey    tag.Key
	revisionTagKey  tag.Key
	serviceTagKey   tag.Key
)

func init() {
	register()
}

func register() {
	var err error
	// Create the tag keys that will be used to add tags to our measurements.
	// Tag keys must conform to the restrictions described in
	// go.opencensus.io/tag/validate.go. Currently those restrictions are:
	// - length between 1 and 255 inclusive
	// - characters are printable US-ASCII
	namespaceTagKey, err = tag.NewKey(metricskey.LabelNamespaceName)
	if err != nil {
		panic(err)
	}
	serviceTagKey, err = tag.NewKey(metricskey.LabelServiceName)
	if err != nil {
		panic(err)
	}
	configTagKey, err = tag.NewKey(metricskey.LabelConfigurationName)
	if err != nil {
		panic(err)
	}
	revisionTagKey, err = tag.NewKey(metricskey.LabelRevisionName)
	if err != nil {
		panic(err)
	}

	// Create views to see our measurements. This can return an error if
	// a previously-registered view has the same name with a different value.
	// View name defaults to the measure name if unspecified.
	err = view.Register(
		&view.View{
			Description: "Number of pods autoscaler wants to allocate",
			Measure:     desiredPodCountM,
			Aggregation: view.LastValue(),
			TagKeys:     []tag.Key{namespaceTagKey, serviceTagKey, configTagKey, revisionTagKey},
		},
		&view.View{
			Description: "Number of pods autoscaler requested from Kubernetes",
			Measure:     requestedPodCountM,
			Aggregation: view.LastValue(),
			TagKeys:     []tag.Key{namespaceTagKey, serviceTagKey, configTagKey, revisionTagKey},
		},
		&view.View{
			Description: "Number of pods that are allocated currently",
			Measure:     actualPodCountM,
			Aggregation: view.LastValue(),
			TagKeys:     []tag.Key{namespaceTagKey, serviceTagKey, configTagKey, revisionTagKey},
		},
		&view.View{
			Description: "Average of requests count over the stable window",
			Measure:     stableRequestConcurrencyM,
			Aggregation: view.LastValue(),
			TagKeys:     []tag.Key{namespaceTagKey, serviceTagKey, configTagKey, revisionTagKey},
		},
		&view.View{
			Description: "Current excess burst capacity over average request count over the stable window",
			Measure:     excessBurstCapacityM,
			Aggregation: view.LastValue(),
			TagKeys:     []tag.Key{namespaceTagKey, serviceTagKey, configTagKey, revisionTagKey},
		},
		&view.View{
			Description: "Average of requests count over the panic window",
			Measure:     panicRequestConcurrencyM,
			Aggregation: view.LastValue(),
			TagKeys:     []tag.Key{namespaceTagKey, serviceTagKey, configTagKey, revisionTagKey},
		},
		&view.View{
			Description: "The desired number of concurrent requests for each pod",
			Measure:     targetRequestConcurrencyM,
			Aggregation: view.LastValue(),
			TagKeys:     []tag.Key{namespaceTagKey, serviceTagKey, configTagKey, revisionTagKey},
		},
		&view.View{
			Description: "1 if autoscaler is in panic mode, 0 otherwise",
			Measure:     panicM,
			Aggregation: view.LastValue(),
			TagKeys:     []tag.Key{namespaceTagKey, serviceTagKey, configTagKey, revisionTagKey},
		},
	)
	if err != nil {
		panic(err)
	}
}

// StatsReporter defines the interface for sending autoscaler metrics
type StatsReporter interface {
	ReportDesiredPodCount(v int64) error
	ReportRequestedPodCount(v int64) error
	ReportActualPodCount(v int64) error
	ReportStableRequestConcurrency(v float64) error
	ReportPanicRequestConcurrency(v float64) error
	ReportTargetRequestConcurrency(v float64) error
	ReportExcessBurstCapacity(v float64) error
	ReportPanic(v int64) error
}

// Reporter holds cached metric objects to report autoscaler metrics
type Reporter struct {
	ctx         context.Context
	initialized bool
}

func valueOrUnknown(v string) string {
	if v != "" {
		return v
	}
	return metricskey.ValueUnknown
}

// NewStatsReporter creates a reporter that collects and reports autoscaler metrics
func NewStatsReporter(podNamespace string, service string, config string, revision string) (*Reporter, error) {
	r := &Reporter{}

	// Our tags are static. So, we can get away with creating a single context
	// and reuse it for reporting all of our metrics. Note that service names
	// can be an empty string, so it needs a special treatment.
	ctx, err := tag.New(
		context.Background(),
		tag.Insert(namespaceTagKey, podNamespace),
		tag.Insert(serviceTagKey, valueOrUnknown(service)),
		tag.Insert(configTagKey, config),
		tag.Insert(revisionTagKey, revision))
	if err != nil {
		return nil, err
	}

	r.ctx = ctx
	r.initialized = true
	return r, nil
}

// ReportDesiredPodCount captures value v for desired pod count measure.
func (r *Reporter) ReportDesiredPodCount(v int64) error {
	return r.report(desiredPodCountM.M(v))
}

// ReportRequestedPodCount captures value v for requested pod count measure.
func (r *Reporter) ReportRequestedPodCount(v int64) error {
	return r.report(requestedPodCountM.M(v))
}

// ReportActualPodCount captures value v for actual pod count measure.
func (r *Reporter) ReportActualPodCount(v int64) error {
	return r.report(actualPodCountM.M(v))
}

// ReportExcessBurstCapacity captures value v for excess target burst capacity.
func (r *Reporter) ReportExcessBurstCapacity(v float64) error {
	return r.report(excessBurstCapacityM.M(v))
}

// ReportStableRequestConcurrency captures value v for stable request concurrency measure.
func (r *Reporter) ReportStableRequestConcurrency(v float64) error {
	return r.report(stableRequestConcurrencyM.M(v))
}

// ReportPanicRequestConcurrency captures value v for panic request concurrency measure.
func (r *Reporter) ReportPanicRequestConcurrency(v float64) error {
	return r.report(panicRequestConcurrencyM.M(v))
}

// ReportTargetRequestConcurrency captures value v for target request concurrency measure.
func (r *Reporter) ReportTargetRequestConcurrency(v float64) error {
	return r.report(targetRequestConcurrencyM.M(v))
}

// ReportPanic captures value v for panic mode measure.
func (r *Reporter) ReportPanic(v int64) error {
	return r.report(panicM.M(v))
}

func (r *Reporter) report(m stats.Measurement) error {
	if !r.initialized {
		return errors.New("StatsReporter is not initialized yet")
	}

	metrics.Record(r.ctx, m)
	return nil
}