karmada/pkg/controllers/federatedhpa/replica_calculator.go

/*
Copyright 2023 The Karmada 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 federatedhpa

import (
	"context"
	"fmt"
	"math"
	"time"

	autoscalingv2 "k8s.io/api/autoscaling/v2"
	corev1 "k8s.io/api/core/v1"
	"k8s.io/apimachinery/pkg/labels"
	"k8s.io/apimachinery/pkg/util/sets"

	metricsclient "github.com/karmada-io/karmada/pkg/controllers/federatedhpa/metrics"
	"github.com/karmada-io/karmada/pkg/util/helper"
)

// This file is basically lifted from https://github.com/kubernetes/kubernetes/blob/release-1.27/pkg/controller/podautoscaler/replica_calculator.go.
// The main difference is:
// 1. ReplicaCalculator no longer has podLister built in. PodList is calculated in the outer controller.
// 2. ReplicaCalculator needs to import a calibration value to calibrate the calculation results
// when they are determined by the global number of ready Pods or metrics.

// ReplicaCalculator bundles all needed information to calculate the target amount of replicas
type ReplicaCalculator struct {
	metricsClient                 metricsclient.QueryClient
	tolerance                     float64
	cpuInitializationPeriod       time.Duration
	delayOfInitialReadinessStatus time.Duration
}

// NewReplicaCalculator creates a new ReplicaCalculator and passes all necessary information to the new instance
func NewReplicaCalculator(metricsClient metricsclient.QueryClient, tolerance float64, cpuInitializationPeriod, delayOfInitialReadinessStatus time.Duration) *ReplicaCalculator {
	return &ReplicaCalculator{
		metricsClient:                 metricsClient,
		tolerance:                     tolerance,
		cpuInitializationPeriod:       cpuInitializationPeriod,
		delayOfInitialReadinessStatus: delayOfInitialReadinessStatus,
	}
}

// GetResourceReplicas calculates the desired replica count based on a target resource utilization percentage
// of the given resource for pods matching the given selector in the given namespace, and the current replica count
//
//nolint:gocyclo
func (c *ReplicaCalculator) GetResourceReplicas(ctx context.Context, currentReplicas int32, targetUtilization int32, resource corev1.ResourceName, namespace string, selector labels.Selector, container string, podList []*corev1.Pod, calibration float64) (replicaCount int32, utilization int32, rawUtilization int64, timestamp time.Time, err error) {
	metrics, timestamp, err := c.metricsClient.GetResourceMetric(ctx, resource, namespace, selector, container)
	if err != nil {
		return 0, 0, 0, time.Time{}, fmt.Errorf("unable to get metrics for resource %s: %v", resource, err)
	}

	itemsLen := len(podList)
	if itemsLen == 0 {
		return 0, 0, 0, time.Time{}, fmt.Errorf("no pods returned by selector while calculating replica count")
	}

	readyPodCount, unreadyPods, missingPods, ignoredPods := groupPods(podList, metrics, resource, c.cpuInitializationPeriod, c.delayOfInitialReadinessStatus)
	removeMetricsForPods(metrics, ignoredPods)
	removeMetricsForPods(metrics, unreadyPods)
	requests, err := calculatePodRequests(podList, container, resource)
	if err != nil {
		return 0, 0, 0, time.Time{}, err
	}

	if len(metrics) == 0 {
		return 0, 0, 0, time.Time{}, fmt.Errorf("did not receive metrics for any ready pods")
	}

	usageRatio, utilization, rawUtilization, err := metricsclient.GetResourceUtilizationRatio(metrics, requests, targetUtilization)
	if err != nil {
		return 0, 0, 0, time.Time{}, err
	}

	scaleUpWithUnready := len(unreadyPods) > 0 && usageRatio > 1.0
	if !scaleUpWithUnready && len(missingPods) == 0 {
		if math.Abs(1.0-usageRatio) <= c.tolerance {
			// return the current replicas if the change would be too small
			return currentReplicas, utilization, rawUtilization, timestamp, nil
		}

		// if we don't have any unready or missing pods, we can calculate the new replica count now
		return int32(math.Ceil(usageRatio * float64(readyPodCount) / calibration)), utilization, rawUtilization, timestamp, nil
	}

	if len(missingPods) > 0 {
		if usageRatio < 1.0 {
			// on a scale-down, treat missing pods as using 100% (all) of the resource request
			// or the utilization target for targets higher than 100%
			fallbackUtilization := int64(max(100, targetUtilization))
			for podName := range missingPods {
				metrics[podName] = metricsclient.PodMetric{Value: requests[podName] * fallbackUtilization / 100}
			}
		} else if usageRatio > 1.0 {
			// on a scale-up, treat missing pods as using 0% of the resource request
			for podName := range missingPods {
				metrics[podName] = metricsclient.PodMetric{Value: 0}
			}
		}
	}

	if scaleUpWithUnready {
		// on a scale-up, treat unready pods as using 0% of the resource request
		for podName := range unreadyPods {
			metrics[podName] = metricsclient.PodMetric{Value: 0}
		}
	}

	// re-run the utilization calculation with our new numbers
	newUsageRatio, _, _, err := metricsclient.GetResourceUtilizationRatio(metrics, requests, targetUtilization)
	if err != nil {
		return 0, utilization, rawUtilization, time.Time{}, err
	}

	if math.Abs(1.0-newUsageRatio) <= c.tolerance || (usageRatio < 1.0 && newUsageRatio > 1.0) || (usageRatio > 1.0 && newUsageRatio < 1.0) {
		// return the current replicas if the change would be too small,
		// or if the new usage ratio would cause a change in scale direction
		return currentReplicas, utilization, rawUtilization, timestamp, nil
	}

	newReplicas := int32(math.Ceil(newUsageRatio * float64(len(metrics)) / calibration))
	if (newUsageRatio < 1.0 && newReplicas > currentReplicas) || (newUsageRatio > 1.0 && newReplicas < currentReplicas) {
		// return the current replicas if the change of metrics length would cause a change in scale direction
		return currentReplicas, utilization, rawUtilization, timestamp, nil
	}

	// return the result, where the number of replicas considered is
	// however many replicas factored into our calculation
	return newReplicas, utilization, rawUtilization, timestamp, nil
}

// GetRawResourceReplicas calculates the desired replica count based on a target resource utilization (as a raw milli-value)
// for pods matching the given selector in the given namespace, and the current replica count
func (c *ReplicaCalculator) GetRawResourceReplicas(ctx context.Context, currentReplicas int32, targetUsage int64, resource corev1.ResourceName, namespace string, selector labels.Selector, container string, podList []*corev1.Pod, calibration float64) (replicaCount int32, usage int64, timestamp time.Time, err error) {
	metrics, timestamp, err := c.metricsClient.GetResourceMetric(ctx, resource, namespace, selector, container)
	if err != nil {
		return 0, 0, time.Time{}, fmt.Errorf("unable to get metrics for resource %s: %v", resource, err)
	}

	replicaCount, usage, err = c.calcPlainMetricReplicas(metrics, currentReplicas, targetUsage, resource, podList, calibration)
	return replicaCount, usage, timestamp, err
}

// GetMetricReplicas calculates the desired replica count based on a target metric usage
// (as a milli-value) for pods matching the given selector in the given namespace, and the
// current replica count
func (c *ReplicaCalculator) GetMetricReplicas(currentReplicas int32, targetUsage int64, metricName string, namespace string, selector labels.Selector, metricSelector labels.Selector, podList []*corev1.Pod, calibration float64) (replicaCount int32, usage int64, timestamp time.Time, err error) {
	metrics, timestamp, err := c.metricsClient.GetRawMetric(metricName, namespace, selector, metricSelector)
	if err != nil {
		return 0, 0, time.Time{}, fmt.Errorf("unable to get metric %s: %v", metricName, err)
	}

	replicaCount, usage, err = c.calcPlainMetricReplicas(metrics, currentReplicas, targetUsage, corev1.ResourceName(""), podList, calibration)
	return replicaCount, usage, timestamp, err
}

// calcPlainMetricReplicas calculates the desired replicas for plain (i.e. non-utilization percentage) metrics.
//
//nolint:gocyclo
func (c *ReplicaCalculator) calcPlainMetricReplicas(metrics metricsclient.PodMetricsInfo, currentReplicas int32, targetUsage int64, resource corev1.ResourceName, podList []*corev1.Pod, calibration float64) (replicaCount int32, usage int64, err error) {
	if len(podList) == 0 {
		return 0, 0, fmt.Errorf("no pods returned by selector while calculating replica count")
	}

	readyPodCount, unreadyPods, missingPods, ignoredPods := groupPods(podList, metrics, resource, c.cpuInitializationPeriod, c.delayOfInitialReadinessStatus)
	removeMetricsForPods(metrics, ignoredPods)
	removeMetricsForPods(metrics, unreadyPods)

	if len(metrics) == 0 {
		return 0, 0, fmt.Errorf("did not receive metrics for any ready pods")
	}

	usageRatio, usage := metricsclient.GetMetricUsageRatio(metrics, targetUsage)

	scaleUpWithUnready := len(unreadyPods) > 0 && usageRatio > 1.0

	if !scaleUpWithUnready && len(missingPods) == 0 {
		if math.Abs(1.0-usageRatio) <= c.tolerance {
			// return the current replicas if the change would be too small
			return currentReplicas, usage, nil
		}

		// if we don't have any unready or missing pods, we can calculate the new replica count now
		return int32(math.Ceil(usageRatio * float64(readyPodCount) / calibration)), usage, nil
	}

	if len(missingPods) > 0 {
		if usageRatio < 1.0 {
			// on a scale-down, treat missing pods as using 100% of the resource request
			for podName := range missingPods {
				metrics[podName] = metricsclient.PodMetric{Value: targetUsage}
			}
		} else if usageRatio > 1.0 {
			// on a scale-up, treat missing pods as using 0% of the resource request
			for podName := range missingPods {
				metrics[podName] = metricsclient.PodMetric{Value: 0}
			}
		}
	}

	if scaleUpWithUnready {
		// on a scale-up, treat unready pods as using 0% of the resource request
		for podName := range unreadyPods {
			metrics[podName] = metricsclient.PodMetric{Value: 0}
		}
	}

	// re-run the utilization calculation with our new numbers
	newUsageRatio, _ := metricsclient.GetMetricUsageRatio(metrics, targetUsage)

	if math.Abs(1.0-newUsageRatio) <= c.tolerance || (usageRatio < 1.0 && newUsageRatio > 1.0) || (usageRatio > 1.0 && newUsageRatio < 1.0) {
		// return the current replicas if the change would be too small,
		// or if the new usage ratio would cause a change in scale direction
		return currentReplicas, usage, nil
	}

	newReplicas := int32(math.Ceil(newUsageRatio * float64(len(metrics)) / calibration))
	if (newUsageRatio < 1.0 && newReplicas > currentReplicas) || (newUsageRatio > 1.0 && newReplicas < currentReplicas) {
		// return the current replicas if the change of metrics length would cause a change in scale direction
		return currentReplicas, usage, nil
	}

	// return the result, where the number of replicas considered is
	// however many replicas factored into our calculation
	return newReplicas, usage, nil
}

// GetObjectMetricReplicas calculates the desired replica count based on a target metric usage (as a milli-value)
// for the given object in the given namespace, and the current replica count.
func (c *ReplicaCalculator) GetObjectMetricReplicas(currentReplicas int32, targetUsage int64, metricName string, namespace string, objectRef *autoscalingv2.CrossVersionObjectReference, metricSelector labels.Selector, podList []*corev1.Pod, calibration float64) (replicaCount int32, usage int64, timestamp time.Time, err error) {
	usage, _, err = c.metricsClient.GetObjectMetric(metricName, namespace, objectRef, metricSelector)
	if err != nil {
		return 0, 0, time.Time{}, fmt.Errorf("unable to get metric %s: %v on %s %s/%s", metricName, objectRef.Kind, namespace, objectRef.Name, err)
	}

	usageRatio := float64(usage) / float64(targetUsage)
	replicaCount, timestamp, err = c.getUsageRatioReplicaCount(currentReplicas, usageRatio, podList, calibration)
	return replicaCount, usage, timestamp, err
}

// GetObjectPerPodMetricReplicas calculates the desired replica count based on a target metric usage (as a milli-value)
// for the given object in the given namespace, and the current replica count.
func (c *ReplicaCalculator) GetObjectPerPodMetricReplicas(statusReplicas int32, targetAverageUsage int64, metricName string, namespace string, objectRef *autoscalingv2.CrossVersionObjectReference, metricSelector labels.Selector, calibration float64) (replicaCount int32, usage int64, timestamp time.Time, err error) {
	// The usage here refers to the total value of all metrics from the pods.
	usage, timestamp, err = c.metricsClient.GetObjectMetric(metricName, namespace, objectRef, metricSelector)
	if err != nil {
		return 0, 0, time.Time{}, fmt.Errorf("unable to get metric %s: %v on %s %s/%s", metricName, objectRef.Kind, namespace, objectRef.Name, err)
	}

	replicaCount = statusReplicas
	usageRatio := float64(usage) / (float64(targetAverageUsage) * float64(replicaCount))
	if math.Abs(1.0-usageRatio) > c.tolerance {
		// update number of replicas if change is large enough
		replicaCount = int32(math.Ceil(float64(usage) / float64(targetAverageUsage) / calibration))
	}
	usage = int64(math.Ceil(float64(usage) / float64(statusReplicas)))
	return int32(math.Ceil(float64(replicaCount) / calibration)), usage, timestamp, nil
}

// getUsageRatioReplicaCount calculates the desired replica count based on usageRatio and ready pods count.
// For currentReplicas=0 doesn't take into account ready pods count and tolerance to support scaling to zero pods.
func (c *ReplicaCalculator) getUsageRatioReplicaCount(currentReplicas int32, usageRatio float64, podList []*corev1.Pod, calibration float64) (replicaCount int32, timestamp time.Time, err error) {
	if currentReplicas != 0 {
		if math.Abs(1.0-usageRatio) <= c.tolerance {
			// return the current replicas if the change would be too small
			return currentReplicas, timestamp, nil
		}
		readyPodCount := int64(0)
		readyPodCount, err = c.getReadyPodsCount(podList)
		if err != nil {
			return 0, time.Time{}, fmt.Errorf("unable to calculate ready pods: %s", err)
		}
		replicaCount = int32(math.Ceil(usageRatio * float64(readyPodCount) / calibration))
	} else {
		// Scale to zero or n pods depending on usageRatio
		replicaCount = int32(math.Ceil(usageRatio))
	}

	return replicaCount, timestamp, err
}

// @TODO(mattjmcnaughton) Many different functions in this module use variations
// of this function. Make this function generic, so we don't repeat the same
// logic in multiple places.
func (c *ReplicaCalculator) getReadyPodsCount(podList []*corev1.Pod) (int64, error) {
	if len(podList) == 0 {
		return 0, fmt.Errorf("no pods returned by selector while calculating replica count")
	}

	readyPodCount := 0

	for _, pod := range podList {
		if pod.Status.Phase == corev1.PodRunning && helper.IsPodReady(pod) {
			readyPodCount++
		}
	}

	return int64(readyPodCount), nil
}

func groupPods(pods []*corev1.Pod, metrics metricsclient.PodMetricsInfo, resource corev1.ResourceName, cpuInitializationPeriod, delayOfInitialReadinessStatus time.Duration) (readyPodCount int, unreadyPods, missingPods, ignoredPods sets.Set[string]) {
	missingPods = sets.New[string]()
	unreadyPods = sets.New[string]()
	ignoredPods = sets.New[string]()
	for _, pod := range pods {
		if pod.DeletionTimestamp != nil || pod.Status.Phase == corev1.PodFailed {
			ignoredPods.Insert(pod.Name)
			continue
		}
		// Pending pods are ignored.
		if pod.Status.Phase == corev1.PodPending {
			unreadyPods.Insert(pod.Name)
			continue
		}
		// Pods missing metrics.
		metric, found := metrics[pod.Name]
		if !found {
			missingPods.Insert(pod.Name)
			continue
		}
		// Unready pods are ignored.
		if resource == corev1.ResourceCPU {
			var unready bool
			_, condition := helper.GetPodCondition(&pod.Status, corev1.PodReady)
			if condition == nil || pod.Status.StartTime == nil {
				unready = true
			} else {
				// Pod still within possible initialisation period.
				if pod.Status.StartTime.Add(cpuInitializationPeriod).After(time.Now()) {
					// Ignore sample if pod is unready or one window of metric wasn't collected since last state transition.
					unready = condition.Status == corev1.ConditionFalse || metric.Timestamp.Before(condition.LastTransitionTime.Time.Add(metric.Window))
				} else {
					// Ignore metric if pod is unready and it has never been ready.
					unready = condition.Status == corev1.ConditionFalse && pod.Status.StartTime.Add(delayOfInitialReadinessStatus).After(condition.LastTransitionTime.Time)
				}
			}
			if unready {
				unreadyPods.Insert(pod.Name)
				continue
			}
		}
		readyPodCount++
	}
	return
}

func calculatePodRequests(pods []*corev1.Pod, container string, resource corev1.ResourceName) (map[string]int64, error) {
	requests := make(map[string]int64, len(pods))
	for _, pod := range pods {
		podSum := int64(0)
		for _, c := range pod.Spec.Containers {
			if container == "" || container == c.Name {
				if containerRequest, ok := c.Resources.Requests[resource]; ok {
					podSum += containerRequest.MilliValue()
				} else {
					return nil, fmt.Errorf("missing request for %s in container %s of Pod %s", resource, c.Name, pod.ObjectMeta.Name)
				}
			}
		}
		requests[pod.Name] = podSum
	}
	return requests, nil
}

func removeMetricsForPods(metrics metricsclient.PodMetricsInfo, pods sets.Set[string]) {
	for _, pod := range pods.UnsortedList() {
		delete(metrics, pod)
	}
}