autoscaler

package
v0.12.0 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Jan 21, 2020 License: Apache-2.0 Imports: 39 Imported by: 61

README

Scaling documentation has moved to the docs folder.

Documentation

Overview

Package autoscaler calculates the number of pods necessary for the desired level of concurrency per pod (stableConcurrencyPerPod). It operates in two modes, stable mode and panic mode.

Stable mode calculates the average concurrency observed over the last 60 seconds and adjusts the observed pod count to achieve the target value. Current observed pod count is the number of unique pod names which show up in the last 60 seconds.

Panic mode calculates the average concurrency observed over the last 6 seconds and adjusts the observed pod count to achieve the stable target value. Panic mode is engaged when the observed 6 second average concurrency reaches 2x the target stable concurrency. Panic mode will last at least 60 seconds--longer if the 2x threshold is repeatedly breached. During panic mode the number of pods is never decreased in order to prevent flapping.

Package autoscaler supports both single-tenant (one autoscaler per revision) and multitenant (one autoscaler for all revisions) autoscalers; config/controller.yaml determines which kind of autoscaler is used.

Index

Constants

View Source 
const (

	// BucketSize is the size of the buckets of stats we create.
	// NB: if this is more than 1s, we need to average values in the
	// metrics buckets.
	BucketSize = scrapeTickInterval
)
View Source 
const (
	// ConfigName is the name of the config map of the autoscaler.
	ConfigName = "config-autoscaler"
)

Variables

View Source 
var (
	// ErrNoData denotes that the collector could not calculate data.
	ErrNoData = errors.New("no data available")

	// ErrNotScraping denotes that the collector is not collecting metrics for the given resource.
	ErrNotScraping = errors.New("the requested resource is not being scraped")
)
View Source 
var (
	// ErrFailedGetEndpoints specifies the error returned by scraper when it fails to
	// get endpoints.
	ErrFailedGetEndpoints = errors.New("failed to get endpoints")

	// ErrDidNotReceiveStat specifies the error returned by scraper when it does not receive
	// stat from an unscraped pod
	ErrDidNotReceiveStat = errors.New("did not receive stat from an unscraped pod")
)

Functions

This section is empty.

Types

type Autoscaler 

type Autoscaler struct {
	// contains filtered or unexported fields
}

Autoscaler stores current state of an instance of an autoscaler.

func New 

func New(
	namespace string,
	revision string,
	metricClient MetricClient,
	lister corev1listers.EndpointsLister,
	deciderSpec *DeciderSpec,
	reporter StatsReporter) (*Autoscaler, error)

New creates a new instance of autoscaler

func (*Autoscaler) Scale 

func (a *Autoscaler) Scale(ctx context.Context, now time.Time) (desiredPodCount int32, excessBC int32, validScale bool)

Scale calculates the desired scale based on current statistics given the current time. desiredPodCount is the calculated pod count the autoscaler would like to set. validScale signifies whether the desiredPodCount should be applied or not.

func (*Autoscaler) Update added in v0.3.0 

func (a *Autoscaler) Update(deciderSpec *DeciderSpec) error

Update reconfigures the UniScaler according to the DeciderSpec.

type Collector added in v0.8.0 

type Collector interface {
	// CreateOrUpdate either creates a collection for the given metric or update it, should
	// it already exist.
	CreateOrUpdate(*av1alpha1.Metric) error
	// Record allows stats to be captured that came from outside the Collector.
	Record(key types.NamespacedName, stat Stat)
	// Delete deletes a Metric and halts collection.
	Delete(string, string) error
}

Collector starts and stops metric collection for a given entity.

type Config 

type Config struct {
	// Feature flags.
	EnableScaleToZero bool

	// Enable connection-aware pod scaledown
	EnableGracefulScaledown bool

	// Target concurrency knobs for different container concurrency configurations.
	ContainerConcurrencyTargetFraction float64
	ContainerConcurrencyTargetDefault  float64
	// TargetUtilization is used for the metrics other than concurrency. This is not
	// configurable now. Customers can override it by specifying
	// autoscaling.knative.dev/targetUtilizationPercentage in Revision annotation.
	// TODO(yanweiguo): Expose this to config-autoscaler configmap and eventually
	// deprecate ContainerConcurrencyTargetFraction.
	TargetUtilization float64
	// RPSTargetDefault is the default target value for requests per second.
	RPSTargetDefault float64
	// NB: most of our computations are in floats, so this is float to avoid casting.
	TargetBurstCapacity float64

	// General autoscaler algorithm configuration.
	MaxScaleUpRate           float64
	MaxScaleDownRate         float64
	StableWindow             time.Duration
	PanicWindowPercentage    float64
	PanicThresholdPercentage float64
	TickInterval             time.Duration

	ScaleToZeroGracePeriod time.Duration
}

Config defines the tunable autoscaler parameters +k8s:deepcopy-gen=true

func NewConfigFromConfigMap 

func NewConfigFromConfigMap(configMap *corev1.ConfigMap) (*Config, error)

NewConfigFromConfigMap creates a Config from the supplied ConfigMap

func NewConfigFromMap 

func NewConfigFromMap(data map[string]string) (*Config, error)

NewConfigFromMap creates a Config from the supplied map

func (*Config) DeepCopy added in v0.2.0 

func (in *Config) DeepCopy() *Config

DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new Config.

func (*Config) DeepCopyInto added in v0.2.0 

func (in *Config) DeepCopyInto(out *Config)

DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.

type Decider added in v0.5.0 

type Decider struct {
	metav1.ObjectMeta
	Spec   DeciderSpec
	Status DeciderStatus
}

Decider is a resource which observes the request load of a Revision and recommends a number of replicas to run. +k8s:deepcopy-gen=true

func (*Decider) DeepCopy added in v0.5.0 

func (in *Decider) DeepCopy() *Decider

DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new Decider.

func (*Decider) DeepCopyInto added in v0.5.0 

func (in *Decider) DeepCopyInto(out *Decider)

DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.

type DeciderSpec added in v0.5.0 

type DeciderSpec struct {
	TickInterval     time.Duration
	MaxScaleUpRate   float64
	MaxScaleDownRate float64
	// The metric used for scaling, i.e. concurrency, rps.
	ScalingMetric string
	// The value of scaling metric per pod that we target to maintain.
	// TargetValue <= TotalValue.
	TargetValue float64
	// The total value of scaling metric that a pod can maintain.
	TotalValue float64
	// The burst capacity that user wants to maintain without queuing at the POD level.
	// Note, that queueing still might happen due to the non-ideal load balancing.
	TargetBurstCapacity float64
	PanicThreshold      float64
	// StableWindow is needed to determine when to exit panicmode.
	StableWindow time.Duration
	// The name of the k8s service for pod information.
	ServiceName string
}

DeciderSpec is the parameters in which the Revision should scaled.

type DeciderStatus added in v0.5.0 

type DeciderStatus struct {
	// DesiredScale is the target number of instances that autoscaler
	// this revision needs.
	DesiredScale int32

	// ExcessBurstCapacity is the difference between spare capacity
	// (how much more load the pods in the revision deployment can take before being
	// overloaded) and the configured target burst capacity.
	// If this number is negative: Activator will be threaded in
	// the request path by the PodAutoscaler controller.
	ExcessBurstCapacity int32
}

DeciderStatus is the current scale recommendation.

type MetricClient added in v0.6.0 

type MetricClient interface {
	// StableAndPanicConcurrency returns both the stable and the panic concurrency
	// for the given replica as of the given time.
	StableAndPanicConcurrency(key types.NamespacedName, now time.Time) (float64, float64, error)

	// StableAndPanicRPS returns both the stable and the panic RPS
	// for the given replica as of the given time.
	StableAndPanicRPS(key types.NamespacedName, now time.Time) (float64, float64, error)
}

MetricClient surfaces the metrics that can be obtained via the collector.

type MetricCollector added in v0.6.0 

type MetricCollector struct {
	// contains filtered or unexported fields
}

MetricCollector manages collection of metrics for many entities.

func NewMetricCollector added in v0.6.0 

func NewMetricCollector(statsScraperFactory StatsScraperFactory, logger *zap.SugaredLogger) *MetricCollector

NewMetricCollector creates a new metric collector.

func (*MetricCollector) CreateOrUpdate added in v0.8.0 

func (c *MetricCollector) CreateOrUpdate(metric *av1alpha1.Metric) error

CreateOrUpdate either creates a collection for the given metric or update it, should it already exist. Map access optimized via double-checked locking.

func (*MetricCollector) Delete added in v0.6.0 

func (c *MetricCollector) Delete(namespace, name string) error

Delete deletes a Metric and halts collection.

func (*MetricCollector) Record added in v0.6.0 

func (c *MetricCollector) Record(key types.NamespacedName, stat Stat)

Record records a stat that's been generated outside of the metric collector.

func (*MetricCollector) StableAndPanicConcurrency added in v0.6.0 

func (c *MetricCollector) StableAndPanicConcurrency(key types.NamespacedName, now time.Time) (float64, float64, error)

StableAndPanicConcurrency returns both the stable and the panic concurrency. It may truncate metric buckets as a side-effect.

func (*MetricCollector) StableAndPanicRPS added in v0.9.0 

func (c *MetricCollector) StableAndPanicRPS(key types.NamespacedName, now time.Time) (float64, float64, error)

StableAndPanicRPS returns both the stable and the panic RPS. It may truncate metric buckets as a side-effect.

type MetricProvider added in v0.7.0 

type MetricProvider struct {
	// contains filtered or unexported fields
}

MetricProvider is a provider to back a custom-metrics API implementation.

func NewMetricProvider added in v0.7.0 

func NewMetricProvider(metricClient MetricClient) *MetricProvider

NewMetricProvider creates a new MetricProvider.

func (*MetricProvider) GetMetricByName added in v0.7.0 

func (p *MetricProvider) GetMetricByName(name types.NamespacedName, info provider.CustomMetricInfo,
	metricSelector labels.Selector) (*cmetrics.MetricValue, error)

GetMetricByName implements the interface.

func (*MetricProvider) GetMetricBySelector added in v0.7.0 

func (p *MetricProvider) GetMetricBySelector(string, labels.Selector, provider.CustomMetricInfo, labels.Selector) (*cmetrics.MetricValueList, error)

GetMetricBySelector implements the interface.

func (*MetricProvider) ListAllMetrics added in v0.7.0 

func (p *MetricProvider) ListAllMetrics() []provider.CustomMetricInfo

ListAllMetrics implements the interface.

type MultiScaler added in v0.2.0 

type MultiScaler struct {
	// contains filtered or unexported fields
}

MultiScaler maintains a collection of Uniscalers.

func NewMultiScaler added in v0.2.0 

func NewMultiScaler(
	stopCh <-chan struct{},
	uniScalerFactory UniScalerFactory,
	logger *zap.SugaredLogger) *MultiScaler

NewMultiScaler constructs a MultiScaler.

func (*MultiScaler) Create added in v0.2.0 

func (m *MultiScaler) Create(ctx context.Context, decider *Decider) (*Decider, error)

Create instantiates the desired Decider.

func (*MultiScaler) Delete added in v0.2.0 

func (m *MultiScaler) Delete(ctx context.Context, namespace, name string) error

Delete stops and removes a Decider.

func (*MultiScaler) Get added in v0.2.0 

func (m *MultiScaler) Get(ctx context.Context, namespace, name string) (*Decider, error)

Get returns the copy of the current Decider.

func (*MultiScaler) Inform added in v0.3.0 

func (m *MultiScaler) Inform(event types.NamespacedName) bool

Inform sends an update to the registered watcher function, if it is set.

func (*MultiScaler) Poke added in v0.6.0 

func (m *MultiScaler) Poke(key types.NamespacedName, stat Stat)

Poke checks if the autoscaler needs to be run immediately.

func (*MultiScaler) Update added in v0.3.0 

func (m *MultiScaler) Update(ctx context.Context, decider *Decider) (*Decider, error)

Update applied the desired DeciderSpec to a currently running Decider.

func (*MultiScaler) Watch added in v0.2.0 

func (m *MultiScaler) Watch(fn func(types.NamespacedName))

Watch registers a singleton function to call when DeciderStatus is updated.

type Reporter 

type Reporter struct {
	// contains filtered or unexported fields
}

Reporter holds cached metric objects to report autoscaler metrics

func NewStatsReporter 

func NewStatsReporter(ns, service, config, revision string) (*Reporter, error)

NewStatsReporter creates a reporter that collects and reports autoscaler metrics

func (*Reporter) ReportActualPodCount added in v0.4.0 

func (r *Reporter) ReportActualPodCount(v int64) error

ReportActualPodCount captures value v for actual pod count measure.

func (*Reporter) ReportDesiredPodCount added in v0.4.0 

func (r *Reporter) ReportDesiredPodCount(v int64) error

ReportDesiredPodCount captures value v for desired pod count measure.

func (*Reporter) ReportExcessBurstCapacity added in v0.8.0 

func (r *Reporter) ReportExcessBurstCapacity(v float64) error

ReportExcessBurstCapacity captures value v for excess target burst capacity.

func (*Reporter) ReportPanic added in v0.4.0 

func (r *Reporter) ReportPanic(v int64) error

ReportPanic captures value v for panic mode measure.

func (*Reporter) ReportPanicRPS added in v0.9.0 

func (r *Reporter) ReportPanicRPS(v float64) error

ReportPanicRPS captures value v for panic RPS measure.

func (*Reporter) ReportPanicRequestConcurrency added in v0.4.0 

func (r *Reporter) ReportPanicRequestConcurrency(v float64) error

ReportPanicRequestConcurrency captures value v for panic request concurrency measure.

func (*Reporter) ReportRequestedPodCount added in v0.4.0 

func (r *Reporter) ReportRequestedPodCount(v int64) error

ReportRequestedPodCount captures value v for requested pod count measure.

func (*Reporter) ReportStableRPS added in v0.9.0 

func (r *Reporter) ReportStableRPS(v float64) error

ReportStableRPS captures value v for stable RPS measure.

func (*Reporter) ReportStableRequestConcurrency added in v0.4.0 

func (r *Reporter) ReportStableRequestConcurrency(v float64) error

ReportStableRequestConcurrency captures value v for stable request concurrency measure.

func (*Reporter) ReportTargetRPS added in v0.9.0 

func (r *Reporter) ReportTargetRPS(v float64) error

ReportTargetRPS captures value v for target requests-per-second measure.

func (*Reporter) ReportTargetRequestConcurrency added in v0.4.0 

func (r *Reporter) ReportTargetRequestConcurrency(v float64) error

ReportTargetRequestConcurrency captures value v for target request concurrency measure.

type ServiceScraper added in v0.4.0 

type ServiceScraper struct {
	// contains filtered or unexported fields
}

ServiceScraper scrapes Revision metrics via a K8S service by sampling. Which pod to be picked up to serve the request is decided by K8S. Please see https://kubernetes.io/docs/concepts/services-networking/network-policies/ for details.

func NewServiceScraper added in v0.4.0 

func NewServiceScraper(metric *av1alpha1.Metric, counter resources.ReadyPodCounter) (*ServiceScraper, error)

NewServiceScraper creates a new StatsScraper for the Revision which the given Metric is responsible for.

func (*ServiceScraper) Scrape added in v0.4.0 

func (s *ServiceScraper) Scrape() (Stat, error)

Scrape calls the destination service then sends it to the given stats channel.

type Stat 

type Stat struct {
	// The time the data point was received by autoscaler.
	Time time.Time

	// The unique identity of this pod.  Used to count how many pods
	// are contributing to the metrics.
	PodName string

	// Average number of requests currently being handled by this pod.
	AverageConcurrentRequests float64

	// Part of AverageConcurrentRequests, for requests going through a proxy.
	AverageProxiedConcurrentRequests float64

	// Number of requests received since last Stat (approximately requests per second).
	RequestCount float64

	// Part of RequestCount, for requests going through a proxy.
	ProxiedRequestCount float64

	// Process uptime in seconds.
	ProcessUptime float64
}

Stat defines a single measurement at a point in time

type StatMessage 

type StatMessage struct {
	Key  types.NamespacedName
	Stat Stat
}

StatMessage wraps a Stat with identifying information so it can be routed to the correct receiver.

type StatsReporter 

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
	ReportStableRPS(v float64) error
	ReportPanicRPS(v float64) error
	ReportTargetRPS(v float64) error
	ReportExcessBurstCapacity(v float64) error
	ReportPanic(v int64) error
}

StatsReporter defines the interface for sending autoscaler metrics

type StatsScraper added in v0.4.0 

type StatsScraper interface {
	// Scrape scrapes the Revision queue metric endpoint.
	Scrape() (Stat, error)
}

StatsScraper defines the interface for collecting Revision metrics

type StatsScraperFactory added in v0.5.0 

type StatsScraperFactory func(*av1alpha1.Metric) (StatsScraper, error)

StatsScraperFactory creates a StatsScraper for a given Metric.

type UniScaler added in v0.2.0 

type UniScaler interface {
	// Scale either proposes a number of replicas and available excess burst capacity,
	// or skips proposing. The proposal is requested at the given time.
	// The returned boolean is true if and only if a proposal was returned.
	Scale(context.Context, time.Time) (int32, int32, bool)

	// Update reconfigures the UniScaler according to the DeciderSpec.
	Update(*DeciderSpec) error
}

UniScaler records statistics for a particular Decider and proposes the scale for the Decider's target based on those statistics.

type UniScalerFactory added in v0.2.0 

type UniScalerFactory func(*Decider) (UniScaler, error)

UniScalerFactory creates a UniScaler for a given PA using the given dynamic configuration.

Source Files

View all Source files

Directories

Path Synopsis
Package statserver provides a WebSocket server which receives autoscaler statistics, typically from queue proxy sidecar containers, and sends them to a channel.
 Package statserver provides a WebSocket server which receives autoscaler statistics, typically from queue proxy sidecar containers, and sends them to a channel.

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.