scheduler.go

/*
Copyright 2022 Ciena Corporation.

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 scheduler

import (
	"context"
	"errors"
	"sync"
	"time"

	constraint_policy_client "github.com/ciena/turnbuckle/internal/pkg/constraint-policy-client"
	"github.com/go-logr/logr"
	v1 "k8s.io/api/core/v1"
	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
	ktypes "k8s.io/apimachinery/pkg/types"
	"k8s.io/apimachinery/pkg/util/wait"
	"k8s.io/client-go/kubernetes"
	"k8s.io/client-go/util/workqueue"
	"k8s.io/kubernetes/pkg/scheduler/framework"
)

// ConstraintPolicyScheduler defines the runtime information used by the
// constraint policy scheduler.
type ConstraintPolicyScheduler struct {
	options               ConstraintPolicySchedulerOptions
	log                   logr.Logger
	defaultPlanner        *ConstraintPolicySchedulerPlanner
	pluginFrameworkHandle framework.Handle
	quit                  chan struct{}
	podRequeueQueue       workqueue.RateLimitingInterface
	podRequeueMap         map[ktypes.NamespacedName]struct{}
	constraintPolicyMutex sync.Mutex
	podRequeueMutex       sync.Mutex
}

// ConstraintPolicySchedulerOptions defines the configuration options for the
// constraint policy scheduler.
type ConstraintPolicySchedulerOptions struct {
	Debug                bool
	NumRetriesOnFailure  int
	MinDelayOnFailure    time.Duration
	MaxDelayOnFailure    time.Duration
	FallbackOnNoOffers   bool
	RetryOnNoOffers      bool
	RequeuePeriod        time.Duration
	PlannerNodeQueueSize uint
	CallTimeout          time.Duration
	UpdateWorkerPeriod   time.Duration
}

// NewScheduler create a new instance of the schedule logic with the
// given set or parameters.
func NewScheduler(options ConstraintPolicySchedulerOptions,
	clientset kubernetes.Interface,
	pluginFrameworkHandle framework.Handle,
	constraintPolicyClient constraint_policy_client.ConstraintPolicyClient,
	log logr.Logger) *ConstraintPolicyScheduler {
	var addPodCallback, deletePodCallback func(pod *v1.Pod)

	constraintPolicyScheduler := &ConstraintPolicyScheduler{}

	deletePodCallback = func(pod *v1.Pod) {
		constraintPolicyScheduler.handlePodDelete(pod)
	}

	defaultPlanner := NewPlanner(
		ConstraintPolicySchedulerPlannerOptions{
			CallTimeout:        options.CallTimeout,
			UpdateWorkerPeriod: options.UpdateWorkerPeriod,
			NodeQueueSize:      options.PlannerNodeQueueSize,
			AddPodCallback:     addPodCallback,
			DeletePodCallback:  deletePodCallback,
		},
		clientset, constraintPolicyClient, log.WithName("default-planner"))

	constraintPolicyScheduler.options = options
	constraintPolicyScheduler.pluginFrameworkHandle = pluginFrameworkHandle
	constraintPolicyScheduler.defaultPlanner = defaultPlanner
	constraintPolicyScheduler.log = log
	constraintPolicyScheduler.quit = make(chan struct{})
	constraintPolicyScheduler.podRequeueQueue = workqueue.
		NewRateLimitingQueue(workqueue.NewItemExponentialFailureRateLimiter(
			options.MinDelayOnFailure,
			options.MaxDelayOnFailure))
	constraintPolicyScheduler.podRequeueMap = make(map[ktypes.NamespacedName]struct{})

	return constraintPolicyScheduler
}

func (s *ConstraintPolicyScheduler) handlePodDelete(pod *v1.Pod) {
	s.podRequeueMutex.Lock()
	defer s.podRequeueMutex.Unlock()
	delete(s.podRequeueMap, ktypes.NamespacedName{Name: pod.Name, Namespace: pod.Namespace})
}

// Stop halts the constraint policy scheduler.
func (s *ConstraintPolicyScheduler) Stop() {
	s.podRequeueQueue.ShutDown()
	close(s.quit)
}

// Start invokes the constraint policy scheduler as a go routine.
func (s *ConstraintPolicyScheduler) Start() {
	go s.listenForPodRequeueEvents()
}

func (s *ConstraintPolicyScheduler) processRequeueEvents() bool {
	// get blocks till there is an item
	item, quit := s.podRequeueQueue.Get()
	if quit {
		return false
	}

	defer s.podRequeueQueue.Done(item)

	s.processRequeue(context.Background(), item)

	return true
}

func (s *ConstraintPolicyScheduler) processRequeue(ctx context.Context, item interface{}) bool {
	data, ok := item.(*v1.Pod)
	if !ok {
		s.log.V(1).Info("pod-requeue-item-not-a-pod")
		s.podRequeueQueue.Forget(item)

		return false
	}
	// if the item is not in the requeue map, forget it
	s.podRequeueMutex.Lock()
	defer s.podRequeueMutex.Unlock()

	forgetItem := true

	defer func() {
		if forgetItem {
			s.podRequeueQueue.Forget(item)
			delete(s.podRequeueMap, ktypes.NamespacedName{Name: data.Name, Namespace: data.Namespace})
		}
	}()

	pod, err := s.defaultPlanner.
		GetClientset().CoreV1().Pods(data.Namespace).
		Get(ctx, data.Name, metav1.GetOptions{})
	if err != nil {
		s.log.Error(err, "pod-requeue-pod-get-failure", "pod", data.Name)

		return false
	}

	if !s.podCheckScheduler(pod) {
		s.log.V(1).Info("pod-requeue-scheduler-mismatch", "pod", pod.Name)

		return false
	}

	if pod.Spec.NodeName != "" {
		s.log.V(1).Info("pod-requeue-node-assigned", "pod", pod.Name, "node", pod.Spec.NodeName)

		return false
	}

	if pod.Status.Phase != v1.PodPending {
		s.log.V(1).Info("pod-requeue-status-not-pending", "pod", pod.Name)

		return false
	}

	if _, ok := s.podRequeueMap[ktypes.NamespacedName{Name: pod.Name, Namespace: pod.Namespace}]; !ok {
		s.log.V(1).Info("pod-requeue-delete", "pod-probably-deleted", pod.Name)

		return false
	}

	numRequeues := s.podRequeueQueue.NumRequeues(item)
	if s.options.NumRetriesOnFailure <= 0 || numRequeues <= s.options.NumRetriesOnFailure {
		forgetItem = false
		// add back the pod to the podqueue
		s.log.V(1).Info("pod-requeue-add", "pod", pod.Name, "num-requeues", numRequeues)

		return true
	}

	s.log.V(1).Info("pod-requeue-requeues-exceeded", "pod", pod.Name)

	return false
}

func (s *ConstraintPolicyScheduler) listenForPodRequeueEvents() {
	defer s.podRequeueQueue.ShutDown()

	go wait.Until(s.requeueWorker, s.options.RequeuePeriod, s.quit)

	<-s.quit
}

func (s *ConstraintPolicyScheduler) requeueWorker() {
	for s.processRequeueEvents() {
	}
}

// requeue pod into the worker queue.
func (s *ConstraintPolicyScheduler) requeue(ctx context.Context, pod *v1.Pod) bool {
	s.podRequeueMutex.Lock()

	if _, ok := s.podRequeueMap[ktypes.NamespacedName{Name: pod.Name, Namespace: pod.Namespace}]; !ok {
		s.podRequeueMap[ktypes.NamespacedName{Name: pod.Name, Namespace: pod.Namespace}] = struct{}{}
	}

	s.log.V(1).Info("pod-requeue", "pod", pod.Name, "namespace", pod.Namespace)
	s.podRequeueQueue.AddRateLimited(pod)
	s.podRequeueMutex.Unlock()

	// retry from a rate limited queue
	// blocks till item is available
	item, quit := s.podRequeueQueue.Get()
	if quit {
		return false
	}

	defer s.podRequeueQueue.Done(item)

	return s.processRequeue(ctx, item)
}

func (s *ConstraintPolicyScheduler) requeueFixup(pod *v1.Pod) {
	s.podRequeueMutex.Lock()
	defer s.podRequeueMutex.Unlock()
	delete(s.podRequeueMap, ktypes.NamespacedName{Name: pod.Name, Namespace: pod.Namespace})
	s.podRequeueQueue.Forget(pod)
}

func (s *ConstraintPolicyScheduler) podCheckScheduler(pod *v1.Pod) bool {
	return true
}

func (s *ConstraintPolicyScheduler) findFit(
	ctx context.Context,
	pod *v1.Pod,
	eligibleNodes []*v1.Node) (*v1.Node, error) {
	var nodeInstance *v1.Node

	var err error

	for {
		nodeInstance, err = s.defaultPlanner.FindBestNode(ctx, pod, eligibleNodes)
		if errors.Is(err, ErrNoOffers) {
			s.log.V(1).Info("no-offers-found-for-pod", "pod", pod.Name)

			if s.options.FallbackOnNoOffers {
				s.log.V(1).Info("random-assignment", "pod", pod.Name)

				return s.defaultPlanner.FindFitRandom(pod, eligibleNodes)
			}

			if s.options.RetryOnNoOffers {
				s.log.V(1).Info("requeuing-for-retry", "pod", pod.Name)

				if !s.requeue(ctx, pod) {
					return nil, err
				}

				continue
			}

			return nil, err
		}

		if err != nil {
			s.log.Error(err, "nodes-not-found", "pod", pod.Name)
			s.log.V(1).Info("requeuing-for-retry", "pod", pod.Name)

			if !s.requeue(ctx, pod) {
				return nil, err
			}

			continue
		}

		break
	}

	// if the pod was requeued, then cleanup the entry from requeue map
	s.requeueFixup(pod)
	s.log.V(1).Info("found-matching", "node", nodeInstance.Name, "pod", pod.Name)

	return nodeInstance, nil
}

// FindBestNode finds the best node for the pod.
func (s *ConstraintPolicyScheduler) FindBestNode(ctx context.Context,
	pod *v1.Pod,
	feasibleNodes []*v1.Node) (*v1.Node, error,
) {
	s.log.V(1).Info("find-best-node", "pod", pod.Name)

	s.constraintPolicyMutex.Lock()
	defer s.constraintPolicyMutex.Unlock()

	node, err := s.findFit(ctx, pod, feasibleNodes)
	if err != nil {
		return nil, err
	}

	if node == nil {
		s.log.V(1).Info("scheduler-plugin", "pod-waiting-for-planner-assignment", pod.Name)

		return nil, ErrNoNodesFound
	}

	return node, nil
}