sigs.k8s.io/kueue@v0.6.2/pkg/workload/workload.go

/*
Copyright 2022 The Kubernetes 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 workload

import (
	"context"
	"fmt"
	"maps"
	"strings"

	corev1 "k8s.io/api/core/v1"
	apimeta "k8s.io/apimachinery/pkg/api/meta"
	"k8s.io/apimachinery/pkg/api/resource"
	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
	"k8s.io/utils/ptr"
	"sigs.k8s.io/controller-runtime/pkg/client"
	"sigs.k8s.io/controller-runtime/pkg/controller/controllerutil"

	config "sigs.k8s.io/kueue/apis/config/v1beta1"
	kueue "sigs.k8s.io/kueue/apis/kueue/v1beta1"
	"sigs.k8s.io/kueue/pkg/constants"
	"sigs.k8s.io/kueue/pkg/util/api"
	"sigs.k8s.io/kueue/pkg/util/limitrange"
)

var (
	admissionManagedConditions = []string{kueue.WorkloadQuotaReserved, kueue.WorkloadEvicted, kueue.WorkloadAdmitted}
)

type AssigmentClusterQueueState struct {
	LastTriedFlavorIdx     []map[corev1.ResourceName]int
	CohortGeneration       int64
	ClusterQueueGeneration int64
}

func (s *AssigmentClusterQueueState) Clone() *AssigmentClusterQueueState {
	c := AssigmentClusterQueueState{
		LastTriedFlavorIdx:     make([]map[corev1.ResourceName]int, len(s.LastTriedFlavorIdx)),
		CohortGeneration:       s.CohortGeneration,
		ClusterQueueGeneration: s.ClusterQueueGeneration,
	}
	for ps, flavorIdx := range s.LastTriedFlavorIdx {
		c.LastTriedFlavorIdx[ps] = maps.Clone(flavorIdx)
	}
	return &c
}

// PendingFlavors returns whether there are pending flavors to try
// after the last attempt.
func (s *AssigmentClusterQueueState) PendingFlavors() bool {
	if s == nil {
		// This is only reached in unit tests.
		return false
	}
	for _, podSetIdxs := range s.LastTriedFlavorIdx {
		for _, idx := range podSetIdxs {
			if idx != -1 {
				return true
			}
		}
	}
	return false
}

// Info holds a Workload object and some pre-processing.
type Info struct {
	Obj *kueue.Workload
	// list of total resources requested by the podsets.
	TotalRequests []PodSetResources
	// Populated from the queue during admission or from the admission field if
	// already admitted.
	ClusterQueue   string
	LastAssignment *AssigmentClusterQueueState
}

type PodSetResources struct {
	Name     string
	Requests Requests
	Count    int32
	Flavors  map[corev1.ResourceName]kueue.ResourceFlavorReference
}

func (psr *PodSetResources) ScaledTo(newCount int32) *PodSetResources {
	ret := &PodSetResources{
		Name:     psr.Name,
		Requests: maps.Clone(psr.Requests),
		Count:    psr.Count,
		Flavors:  maps.Clone(psr.Flavors),
	}
	ret.Requests.scaleDown(int64(ret.Count))
	ret.Requests.scaleUp(int64(newCount))
	ret.Count = newCount
	return ret
}

func NewInfo(w *kueue.Workload) *Info {
	info := &Info{
		Obj: w,
	}
	if w.Status.Admission != nil {
		info.ClusterQueue = string(w.Status.Admission.ClusterQueue)
		info.TotalRequests = totalRequestsFromAdmission(w)
	} else {
		info.TotalRequests = totalRequestsFromPodSets(w)
	}
	return info
}

func (i *Info) Update(wl *kueue.Workload) {
	i.Obj = wl
}

func (i *Info) CanBePartiallyAdmitted() bool {
	return CanBePartiallyAdmitted(i.Obj)
}

func CanBePartiallyAdmitted(wl *kueue.Workload) bool {
	ps := wl.Spec.PodSets
	for psi := range ps {
		if ps[psi].Count > ptr.Deref(ps[psi].MinCount, ps[psi].Count) {
			return true
		}
	}
	return false
}

func Key(w *kueue.Workload) string {
	return fmt.Sprintf("%s/%s", w.Namespace, w.Name)
}

func QueueKey(w *kueue.Workload) string {
	return fmt.Sprintf("%s/%s", w.Namespace, w.Spec.QueueName)
}

func reclaimableCounts(wl *kueue.Workload) map[string]int32 {
	ret := make(map[string]int32, len(wl.Status.ReclaimablePods))
	for i := range wl.Status.ReclaimablePods {
		reclaimInfo := &wl.Status.ReclaimablePods[i]
		ret[reclaimInfo.Name] = reclaimInfo.Count
	}
	return ret
}

func podSetsCounts(wl *kueue.Workload) map[string]int32 {

	ret := make(map[string]int32, len(wl.Spec.PodSets))
	for i := range wl.Spec.PodSets {
		ps := &wl.Spec.PodSets[i]
		ret[ps.Name] = ps.Count
	}
	return ret
}

func podSetsCountsAfterReclaim(wl *kueue.Workload) map[string]int32 {
	totalCounts := podSetsCounts(wl)
	reclaimCounts := reclaimableCounts(wl)
	for podSetName := range totalCounts {
		if rc, found := reclaimCounts[podSetName]; found {
			totalCounts[podSetName] -= rc
		}
	}
	return totalCounts
}

func totalRequestsFromPodSets(wl *kueue.Workload) []PodSetResources {
	if len(wl.Spec.PodSets) == 0 {
		return nil
	}
	res := make([]PodSetResources, 0, len(wl.Spec.PodSets))
	currentCounts := podSetsCountsAfterReclaim(wl)
	for _, ps := range wl.Spec.PodSets {
		count := currentCounts[ps.Name]
		setRes := PodSetResources{
			Name:  ps.Name,
			Count: count,
		}
		setRes.Requests = newRequests(limitrange.TotalRequests(&ps.Template.Spec))
		setRes.Requests.scaleUp(int64(count))
		res = append(res, setRes)
	}
	return res
}

func totalRequestsFromAdmission(wl *kueue.Workload) []PodSetResources {
	if wl.Status.Admission == nil {
		return nil
	}
	res := make([]PodSetResources, 0, len(wl.Spec.PodSets))
	currentCounts := podSetsCountsAfterReclaim(wl)
	totalCounts := podSetsCounts(wl)
	for _, psa := range wl.Status.Admission.PodSetAssignments {
		setRes := PodSetResources{
			Name:     psa.Name,
			Flavors:  psa.Flavors,
			Count:    ptr.Deref(psa.Count, totalCounts[psa.Name]),
			Requests: newRequests(psa.ResourceUsage),
		}

		if count := currentCounts[psa.Name]; count != setRes.Count {
			setRes.Requests.scaleDown(int64(setRes.Count))
			setRes.Requests.scaleUp(int64(count))
			setRes.Count = count
		}

		res = append(res, setRes)
	}
	return res
}

// The following resources calculations are inspired on
// https://github.com/kubernetes/kubernetes/blob/master/pkg/scheduler/framework/types.go

// Requests maps ResourceName to flavor to value; for CPU it is tracked in MilliCPU.
type Requests map[corev1.ResourceName]int64

func newRequests(rl corev1.ResourceList) Requests {
	r := Requests{}
	for name, quant := range rl {
		r[name] = ResourceValue(name, quant)
	}
	return r
}

func (r Requests) ToResourceList() corev1.ResourceList {
	ret := make(corev1.ResourceList, len(r))
	for k, v := range r {
		ret[k] = ResourceQuantity(k, v)
	}
	return ret
}

// ResourceValue returns the integer value for the resource name.
// It's milli-units for CPU and absolute units for everything else.
func ResourceValue(name corev1.ResourceName, q resource.Quantity) int64 {
	if name == corev1.ResourceCPU {
		return q.MilliValue()
	}
	return q.Value()
}

func ResourceQuantity(name corev1.ResourceName, v int64) resource.Quantity {
	switch name {
	case corev1.ResourceCPU:
		return *resource.NewMilliQuantity(v, resource.DecimalSI)
	case corev1.ResourceMemory, corev1.ResourceEphemeralStorage:
		return *resource.NewQuantity(v, resource.BinarySI)
	default:
		if strings.HasPrefix(string(name), corev1.ResourceHugePagesPrefix) {
			return *resource.NewQuantity(v, resource.BinarySI)
		}
		return *resource.NewQuantity(v, resource.DecimalSI)
	}
}

func (r Requests) scaleUp(f int64) {
	for name := range r {
		r[name] *= f
	}
}

func (r Requests) scaleDown(f int64) {
	for name := range r {
		r[name] /= f
	}
}

// UpdateStatus updates the condition of a workload with ssa,
// fieldManager being set to managerPrefix + "-" + conditionType
func UpdateStatus(ctx context.Context,
	c client.Client,
	wl *kueue.Workload,
	conditionType string,
	conditionStatus metav1.ConditionStatus,
	reason, message string,
	managerPrefix string) error {
	now := metav1.Now()
	condition := metav1.Condition{
		Type:               conditionType,
		Status:             conditionStatus,
		LastTransitionTime: now,
		Reason:             reason,
		Message:            api.TruncateConditionMessage(message),
	}

	newWl := BaseSSAWorkload(wl)
	newWl.Status.Conditions = []metav1.Condition{condition}
	return c.Status().Patch(ctx, newWl, client.Apply, client.FieldOwner(managerPrefix+"-"+condition.Type))
}

// UnsetQuotaReservationWithCondition sets the QuotaReserved condition to false and clears
// the admission.
// Returns whether any change was done.
func UnsetQuotaReservationWithCondition(wl *kueue.Workload, reason, message string) bool {
	condition := metav1.Condition{
		Type:               kueue.WorkloadQuotaReserved,
		Status:             metav1.ConditionFalse,
		LastTransitionTime: metav1.Now(),
		Reason:             reason,
		Message:            api.TruncateConditionMessage(message),
	}
	changed := apimeta.SetStatusCondition(&wl.Status.Conditions, condition)
	if wl.Status.Admission != nil {
		wl.Status.Admission = nil
		changed = true
	}

	// Reset the admitted condition if necessary.
	if SyncAdmittedCondition(wl) {
		changed = true
	}
	return changed
}

// BaseSSAWorkload creates a new object based on the input workload that
// only contains the fields necessary to identify the original object.
// The object can be used in as a base for Server-Side-Apply.
func BaseSSAWorkload(w *kueue.Workload) *kueue.Workload {
	wlCopy := &kueue.Workload{
		ObjectMeta: metav1.ObjectMeta{
			UID:        w.UID,
			Name:       w.Name,
			Namespace:  w.Namespace,
			Generation: w.Generation, // Produce a conflict if there was a change in the spec.
		},
		TypeMeta: w.TypeMeta,
	}
	if wlCopy.APIVersion == "" {
		wlCopy.APIVersion = kueue.GroupVersion.String()
	}
	if wlCopy.Kind == "" {
		wlCopy.Kind = "Workload"
	}
	return wlCopy
}

// SetQuotaReservation applies the provided admission to the workload.
// The WorkloadAdmitted and WorkloadEvicted are added or updated if necessary.
func SetQuotaReservation(w *kueue.Workload, admission *kueue.Admission) {
	w.Status.Admission = admission
	admittedCond := metav1.Condition{
		Type:               kueue.WorkloadQuotaReserved,
		Status:             metav1.ConditionTrue,
		LastTransitionTime: metav1.Now(),
		Reason:             "QuotaReserved",
		Message:            fmt.Sprintf("Quota reserved in ClusterQueue %s", w.Status.Admission.ClusterQueue),
	}
	apimeta.SetStatusCondition(&w.Status.Conditions, admittedCond)

	//reset Evicted condition if present.
	if evictedCond := apimeta.FindStatusCondition(w.Status.Conditions, kueue.WorkloadEvicted); evictedCond != nil {
		evictedCond.Status = metav1.ConditionFalse
		evictedCond.LastTransitionTime = metav1.Now()
	}
}

func SetEvictedCondition(w *kueue.Workload, reason string, message string) {
	condition := metav1.Condition{
		Type:               kueue.WorkloadEvicted,
		Status:             metav1.ConditionTrue,
		LastTransitionTime: metav1.Now(),
		Reason:             reason,
		Message:            message,
	}
	apimeta.SetStatusCondition(&w.Status.Conditions, condition)
}

// admissionPatch creates a new object based on the input workload that contains
// the admission and related conditions. The object can be used in Server-Side-Apply.
func admissionPatch(w *kueue.Workload) *kueue.Workload {
	wlCopy := BaseSSAWorkload(w)

	wlCopy.Status.Admission = w.Status.Admission.DeepCopy()
	wlCopy.Status.RequeueState = w.Status.RequeueState.DeepCopy()
	for _, conditionName := range admissionManagedConditions {
		if existing := apimeta.FindStatusCondition(w.Status.Conditions, conditionName); existing != nil {
			wlCopy.Status.Conditions = append(wlCopy.Status.Conditions, *existing.DeepCopy())
		}
	}
	return wlCopy
}

// ApplyAdmissionStatus updated all the admission related status fields of a workload with SSA.
// if strict is true, resourceVersion will be part of the patch, make this call fail if Workload
// was changed.
func ApplyAdmissionStatus(ctx context.Context, c client.Client, w *kueue.Workload, strict bool) error {
	patch := admissionPatch(w)
	if strict {
		patch.ResourceVersion = w.ResourceVersion
	}
	return c.Status().Patch(ctx, patch, client.Apply, client.FieldOwner(constants.AdmissionName))
}

type Ordering struct {
	PodsReadyRequeuingTimestamp config.RequeuingTimestamp
}

// GetQueueOrderTimestamp return the timestamp to be used by the scheduler. It could
// be the workload creation time or the last time a PodsReady timeout has occurred.
func (o Ordering) GetQueueOrderTimestamp(w *kueue.Workload) *metav1.Time {
	if o.PodsReadyRequeuingTimestamp == config.EvictionTimestamp {
		if evictedCond, evictedByTimout := IsEvictedByPodsReadyTimeout(w); evictedByTimout {
			return &evictedCond.LastTransitionTime
		}
	}
	return &w.CreationTimestamp
}

// HasQuotaReservation checks if workload is admitted based on conditions
func HasQuotaReservation(w *kueue.Workload) bool {
	return apimeta.IsStatusConditionTrue(w.Status.Conditions, kueue.WorkloadQuotaReserved)
}

// UpdateReclaimablePods updates the ReclaimablePods list for the workload wit SSA.
func UpdateReclaimablePods(ctx context.Context, c client.Client, w *kueue.Workload, reclaimablePods []kueue.ReclaimablePod) error {
	patch := BaseSSAWorkload(w)
	patch.Status.ReclaimablePods = reclaimablePods
	return c.Status().Patch(ctx, patch, client.Apply, client.FieldOwner(constants.ReclaimablePodsMgr))
}

// ReclaimablePodsAreEqual checks if two Reclaimable pods are semantically equal
// having the same length and all keys have the same value.
func ReclaimablePodsAreEqual(a, b []kueue.ReclaimablePod) bool {
	if len(a) != len(b) {
		return false
	}

	mb := make(map[string]int32, len(b))
	for i := range b {
		mb[b[i].Name] = b[i].Count
	}

	for i := range a {
		if bCount, found := mb[a[i].Name]; !found || bCount != a[i].Count {
			return false
		}
	}
	return true
}

// HasRequeueState returns true if the workload has re-queue state.
func HasRequeueState(w *kueue.Workload) bool {
	return w.Status.RequeueState != nil
}

// IsAdmitted returns true if the workload is admitted.
func IsAdmitted(w *kueue.Workload) bool {
	return apimeta.IsStatusConditionTrue(w.Status.Conditions, kueue.WorkloadAdmitted)
}

// IsFinished returns true if the workload is finished.
func IsFinished(w *kueue.Workload) bool {
	return apimeta.IsStatusConditionTrue(w.Status.Conditions, kueue.WorkloadFinished)
}

// IsEvictedByDeactivation returns true if the workload is evicted by deactivation.
func IsEvictedByDeactivation(w *kueue.Workload) bool {
	cond := apimeta.FindStatusCondition(w.Status.Conditions, kueue.WorkloadEvicted)
	return cond != nil && cond.Status == metav1.ConditionTrue && cond.Reason == kueue.WorkloadEvictedByDeactivation
}

func IsEvictedByPodsReadyTimeout(w *kueue.Workload) (*metav1.Condition, bool) {
	cond := apimeta.FindStatusCondition(w.Status.Conditions, kueue.WorkloadEvicted)
	if cond == nil || cond.Status != metav1.ConditionTrue || cond.Reason != kueue.WorkloadEvictedByPodsReadyTimeout {
		return nil, false
	}
	return cond, true
}

func RemoveFinalizer(ctx context.Context, c client.Client, wl *kueue.Workload) error {
	if controllerutil.RemoveFinalizer(wl, kueue.ResourceInUseFinalizerName) {
		return c.Update(ctx, wl)
	}
	return nil
}