scheduling

package
v1.0.1 Latest Latest
Warning

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

 Go to latest Published: Sep 5, 2024 License: Apache-2.0 Imports: 40 Imported by: 2

Documentation

Index

Constants

View Source 
const (
	ControllerLabel = "controller"
)

Variables

View Source 
var (
	SchedulingDurationSeconds = prometheus.NewHistogramVec(
		prometheus.HistogramOpts{
			Namespace: metrics.Namespace,
			Subsystem: schedulerSubsystem,
			Name:      "scheduling_duration_seconds",
			Help:      "Duration of scheduling simulations used for deprovisioning and provisioning in seconds.",
			Buckets:   metrics.DurationBuckets(),
		},
		[]string{
			ControllerLabel,
		},
	)
	QueueDepth = prometheus.NewGaugeVec(
		prometheus.GaugeOpts{
			Namespace: metrics.Namespace,
			Subsystem: schedulerSubsystem,
			Name:      "queue_depth",
			Help:      "The number of pods currently waiting to be scheduled.",
		},
		[]string{
			ControllerLabel,
			schedulingIDLabel,
		},
	)
)
View Source 
var MaxInstanceTypes = 60

MaxInstanceTypes is a constant that restricts the number of instance types to be sent for launch. Note that this is intentionally changed to var just to help in testing the code.

View Source 
var PodNominationRateLimiter = flowcontrol.NewTokenBucketRateLimiter(5, 10)

PodNominationRateLimiter is a pointer so it rate-limits across events

Functions

func IgnoredForTopology 

func IgnoredForTopology(p *corev1.Pod) bool

func InstanceTypeList 

func InstanceTypeList(instanceTypeOptions []*cloudprovider.InstanceType) string

func NominatePodEvent 

func NominatePodEvent(pod *corev1.Pod, node *corev1.Node, nodeClaim *v1.NodeClaim) events.Event

func PodFailedToScheduleEvent 

func PodFailedToScheduleEvent(pod *corev1.Pod, err error) events.Event

func TopologyListOptions 

func TopologyListOptions(namespace string, labelSelector *metav1.LabelSelector) *client.ListOptions

Types

type ExistingNode 

type ExistingNode struct {
	*state.StateNode

	Pods []*v1.Pod
	// contains filtered or unexported fields
}

func NewExistingNode 

func NewExistingNode(n *state.StateNode, topology *Topology, daemonResources v1.ResourceList) *ExistingNode

func (*ExistingNode) Add 

func (n *ExistingNode) Add(ctx context.Context, kubeClient client.Client, pod *v1.Pod) error

type NodeClaim 

type NodeClaim struct {
	NodeClaimTemplate

	Pods []*v1.Pod
	// contains filtered or unexported fields
}

NodeClaim is a set of constraints, compatible pods, and possible instance types that could fulfill these constraints. This will be turned into one or more actual node instances within the cluster after bin packing.

func NewNodeClaim 

func NewNodeClaim(nodeClaimTemplate *NodeClaimTemplate, topology *Topology, daemonResources v1.ResourceList, instanceTypes []*cloudprovider.InstanceType) *NodeClaim

func (*NodeClaim) Add 

func (n *NodeClaim) Add(pod *v1.Pod) error

func (*NodeClaim) FinalizeScheduling 

func (n *NodeClaim) FinalizeScheduling()

FinalizeScheduling is called once all scheduling has completed and allows the node to perform any cleanup necessary before its requirements are used for instance launching

func (*NodeClaim) RemoveInstanceTypeOptionsByPriceAndMinValues added in v1.0.0 

func (n *NodeClaim) RemoveInstanceTypeOptionsByPriceAndMinValues(reqs scheduling.Requirements, maxPrice float64) (*NodeClaim, error)

type NodeClaimTemplate 

type NodeClaimTemplate struct {
	v1.NodeClaim

	NodePoolName        string
	InstanceTypeOptions cloudprovider.InstanceTypes
	Requirements        scheduling.Requirements
}

NodeClaimTemplate encapsulates the fields required to create a node and mirrors the fields in NodePool. These structs are maintained separately in order for fields like Requirements to be able to be stored more efficiently.

func NewNodeClaimTemplate 

func NewNodeClaimTemplate(nodePool *v1.NodePool) *NodeClaimTemplate

func (*NodeClaimTemplate) ToNodeClaim 

func (i *NodeClaimTemplate) ToNodeClaim(nodePool *v1.NodePool) *v1.NodeClaim

type Preferences 

type Preferences struct {
	// ToleratePreferNoSchedule controls if preference relaxation adds a toleration for PreferNoSchedule taints.  This only
	// helps if there is a corresponding taint, so we don't always add it.
	ToleratePreferNoSchedule bool
}

func (*Preferences) Relax 

func (p *Preferences) Relax(ctx context.Context, pod *v1.Pod) bool

type Queue 

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

Queue is a queue of pods that is scheduled. It's used to attempt to schedule pods as long as we are making progress in scheduling. This is sometimes required to maintain zonal topology spreads with constrained pods, and can satisfy pod affinities that occur in a batch of pods if there are enough constraints provided.

func NewQueue 

func NewQueue(pods ...*v1.Pod) *Queue

NewQueue constructs a new queue given the input pods, sorting them to optimize for bin-packing into nodes.

func (*Queue) List 

func (q *Queue) List() []*v1.Pod

func (*Queue) Pop 

func (q *Queue) Pop() (*v1.Pod, bool)

Pop returns the next pod or false if no longer making progress

func (*Queue) Push 

func (q *Queue) Push(pod *v1.Pod, relaxed bool)

Push a pod onto the queue, counting each time a pod is immediately requeued. This is used to detect staleness.

type Results 

type Results struct {
	NewNodeClaims []*NodeClaim
	ExistingNodes []*ExistingNode
	PodErrors     map[*corev1.Pod]error
}

Results contains the results of the scheduling operation

func (Results) AllNonPendingPodsScheduled 

func (r Results) AllNonPendingPodsScheduled() bool

AllNonPendingPodsScheduled returns true if all pods scheduled. We don't care if a pod was pending before consolidation and will still be pending after. It may be a pod that we can't schedule at all and don't want it to block consolidation.

func (Results) NonPendingPodSchedulingErrors 

func (r Results) NonPendingPodSchedulingErrors() string

NonPendingPodSchedulingErrors creates a string that describes why pods wouldn't schedule that is suitable for presentation

func (Results) Record added in v0.35.0 

func (r Results) Record(ctx context.Context, recorder events.Recorder, cluster *state.Cluster)

Record sends eventing and log messages back for the results that were produced from a scheduling run It also nominates nodes in the cluster state based on the scheduling run to signal to other components leveraging the cluster state that a previous scheduling run that was recorded is relying on these nodes

func (Results) TruncateInstanceTypes added in v0.35.0 

func (r Results) TruncateInstanceTypes(maxInstanceTypes int) Results

TruncateInstanceTypes filters the result based on the maximum number of instanceTypes that needs to be considered. This filters all instance types generated in NewNodeClaims in the Results

type Scheduler 

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

func NewScheduler 

func NewScheduler(kubeClient client.Client, nodePools []*v1.NodePool,
	cluster *state.Cluster, stateNodes []*state.StateNode, topology *Topology,
	instanceTypes map[string][]*cloudprovider.InstanceType, daemonSetPods []*corev1.Pod,
	recorder events.Recorder) *Scheduler

func (*Scheduler) Solve 

func (s *Scheduler) Solve(ctx context.Context, pods []*corev1.Pod) Results

type Topology 

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

func NewTopology 

func NewTopology(ctx context.Context, kubeClient client.Client, cluster *state.Cluster, domains map[string]sets.Set[string], pods []*corev1.Pod) (*Topology, error)

func (*Topology) AddRequirements 

func (t *Topology) AddRequirements(podRequirements, nodeRequirements scheduling.Requirements, p *corev1.Pod, compatabilityOptions ...option.Function[scheduling.CompatibilityOptions]) (scheduling.Requirements, error)

AddRequirements tightens the input requirements by adding additional requirements that are being enforced by topology spreads affinities, anti-affinities or inverse anti-affinities. The nodeHostname is the hostname that we are currently considering placing the pod on. It returns these newly tightened requirements, or an error in the case of a set of requirements that cannot be satisfied.

func (*Topology) Record 

func (t *Topology) Record(p *corev1.Pod, requirements scheduling.Requirements, compatabilityOptions ...option.Function[scheduling.CompatibilityOptions])

Record records the topology changes given that pod p schedule on a node with the given requirements

func (*Topology) Register 

func (t *Topology) Register(topologyKey string, domain string)

Register is used to register a domain as available across topologies for the given topology key.

func (*Topology) Update 

func (t *Topology) Update(ctx context.Context, p *corev1.Pod) error

Update unregisters the pod as the owner of all affinities and then creates any new topologies based on the pod spec registered the pod as the owner of all associated affinities, new or old. This allows Update() to be called after relaxation of a preference to properly break the topology <-> owner relationship so that the preferred topology will no longer influence scheduling.

type TopologyGroup 

type TopologyGroup struct {
	// Hashed Fields
	Key  string
	Type TopologyType
	// contains filtered or unexported fields
}

TopologyGroup is used to track pod counts that match a selector by the topology domain (e.g. SELECT COUNT(*) FROM pods GROUP BY(topology_ke

func NewTopologyGroup 

func NewTopologyGroup(topologyType TopologyType, topologyKey string, pod *v1.Pod, namespaces sets.Set[string], labelSelector *metav1.LabelSelector, maxSkew int32, minDomains *int32, domains sets.Set[string]) *TopologyGroup

func (*TopologyGroup) AddOwner 

func (t *TopologyGroup) AddOwner(key types.UID)

func (*TopologyGroup) Counts 

func (t *TopologyGroup) Counts(pod *v1.Pod, requirements scheduling.Requirements, compatabilityOptions ...option.Function[scheduling.CompatibilityOptions]) bool

Counts returns true if the pod would count for the topology, given that it schedule to a node with the provided requirements

func (*TopologyGroup) Get 

func (t *TopologyGroup) Get(pod *v1.Pod, podDomains, nodeDomains *scheduling.Requirement) *scheduling.Requirement

func (*TopologyGroup) Hash 

func (t *TopologyGroup) Hash() uint64

Hash is used so we can track single topologies that affect multiple groups of pods. If a deployment has 100x pods with self anti-affinity, we track that as a single topology with 100 owners instead of 100x topologies.

func (*TopologyGroup) IsOwnedBy 

func (t *TopologyGroup) IsOwnedBy(key types.UID) bool

func (*TopologyGroup) Record 

func (t *TopologyGroup) Record(domains ...string)

func (*TopologyGroup) Register 

func (t *TopologyGroup) Register(domains ...string)

Register ensures that the topology is aware of the given domain names.

func (*TopologyGroup) RemoveOwner 

func (t *TopologyGroup) RemoveOwner(key types.UID)

type TopologyNodeFilter 

type TopologyNodeFilter []scheduling.Requirements

TopologyNodeFilter is used to determine if a given actual node or scheduling node matches the pod's node selectors and required node affinity terms. This is used with topology spread constraints to determine if the node should be included for topology counting purposes. This is only used with topology spread constraints as affinities/anti-affinities always count across all nodes. A nil or zero-value TopologyNodeFilter behaves well and the filter returns true for all nodes.

func MakeTopologyNodeFilter 

func MakeTopologyNodeFilter(p *v1.Pod) TopologyNodeFilter

func (TopologyNodeFilter) Matches 

func (t TopologyNodeFilter) Matches(node *v1.Node) bool

Matches returns true if the TopologyNodeFilter doesn't prohibit node from the participating in the topology

func (TopologyNodeFilter) MatchesRequirements 

func (t TopologyNodeFilter) MatchesRequirements(requirements scheduling.Requirements, compatabilityOptions ...option.Function[scheduling.CompatibilityOptions]) bool

MatchesRequirements returns true if the TopologyNodeFilter doesn't prohibit a node with the requirements from participating in the topology. This method allows checking the requirements from a scheduling.NodeClaim to see if the node we will soon create participates in this topology.

type TopologyType 

type TopologyType byte
const (
	TopologyTypeSpread TopologyType = iota
	TopologyTypePodAffinity
	TopologyTypePodAntiAffinity
)

func (TopologyType) String 

func (t TopologyType) String() string

type VolumeTopology 

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

func NewVolumeTopology 

func NewVolumeTopology(kubeClient client.Client) *VolumeTopology

func (*VolumeTopology) Inject 

func (v *VolumeTopology) Inject(ctx context.Context, pod *v1.Pod) error

func (*VolumeTopology) ValidatePersistentVolumeClaims 

func (v *VolumeTopology) ValidatePersistentVolumeClaims(ctx context.Context, pod *v1.Pod) error

ValidatePersistentVolumeClaims returns an error if the pod doesn't appear to be valid with respect to PVCs (e.g. the PVC is not found or references an unknown storage class).

Source Files

View all Source files

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.