devicemanager

Documentation

Index

• Constants
• Variables
• func FilterActivePods(pods []*v1.Pod) []*v1.Pod
• func GPUAffinityFilter(nodeResources NodeResources, sharepod *faasv1.SharePod)
• func GPUAntiAffinityFilter(nodeResources NodeResources, sharepod *faasv1.SharePod)
• func GPUExclusionFilter(nodeResources NodeResources, sharepod *faasv1.SharePod)
• func IsPodActive(p *v1.Pod) bool
• func IsPodHot(p *v1.Pod) bool
• func PodKey(pod *v1.Pod) string
• func RandStr(length int) string
• func ScheduleAlgorithmBestFit(isGPUPod bool, gpu_request float64, gpu_mem int64, sharepod *faasv1.SharePod, ...) (schedNodeName string, schedGPUID string)
• func StartConfigManager(stopCh <-chan struct{}, kc kubernetes.Interface) error
• func UpdateNodeGPUInfo(nodeName string, uuid2mem *map[string]string)
• type ActivePodsWithRanks
  • func (s ActivePodsWithRanks) Len() int
  • func (s ActivePodsWithRanks) Less(i, j int) bool
  • func (s ActivePodsWithRanks) Swap(i, j int)
• type ClientStatus
• type Controller
  • func NewController(kubeclient kubernetes.Interface, kubeshareclient clientset.Interface, ...) *Controller
  • func (c *Controller) Run(threadiness int, stopCh <-chan struct{}) error
• type GPUInfo
• type GPUResource
  • func (this *GPUResource) DeepCopy() *GPUResource
• type NameList
• type NodeInfo
• type NodeResource
  • func (this *NodeResource) DeepCopy() *NodeResource
• type NodeResources
  • func (this *NodeResources) DeepCopy() *NodeResources
  • func (this *NodeResources) PrintMe()
• type NodeStatus
• type PodRequest
  • func FindInQueue(key string, pl *list.List) (*PodRequest, bool)
• type RealPodControl
• type SharepodProbes

Constants

const (
	// SuccessSynced is used as part of the Event 'reason' when a SharePod is synced
	SuccessSynced = "Synced"
	// ErrResourceExists is used as part of the Event 'reason' when a SharePod fails
	// to sync due to a Deployment of the same name already existing.
	ErrResourceExists = "ErrResourceExists"

	ErrValueError = "ErrValueError"

	// MessageResourceExists is the message used for Events when a resource
	// fails to sync due to a Deployment already existing
	MessageResourceExists = "Resource %q already exists and is not managed by SharePod"
	// MessageResourceSynced is the message used for an Event fired when a SharePod
	// is synced successfully
	MessageResourceSynced = "SharePod synced successfully"

	KubeShareLibraryPath = "/kubeshare/library"
	SchedulerIpPath      = KubeShareLibraryPath + "/schedulerIP.txt"
	PodManagerPortStart  = 50050

	KubeShareScheduleAffinity     = "kubeshare/sched_affinity"
	KubeShareScheduleAntiAffinity = "kubeshare/sched_anti-affinity"
	KubeShareScheduleExclusion    = "kubeshare/sched_exclusion"

	FaasShareWarm = "faas-share/warmpool"
)

const (
	LabelMinReplicas = "com.openfaas.scale.min"
)

Variables

var (
	KeyFunc = cache.DeletionHandlingMetaNamespaceKeyFunc
)

var (
	ResourceQuantity1 = resource.MustParse("1")
)

Functions

func FilterActivePods

func FilterActivePods(pods []*v1.Pod) []*v1.Pod

more percise filter mechanism for pre-warm pod

func GPUAffinityFilter

func GPUAffinityFilter(nodeResources NodeResources, sharepod *faasv1.SharePod)

func GPUAntiAffinityFilter

func GPUAntiAffinityFilter(nodeResources NodeResources, sharepod *faasv1.SharePod)

func GPUExclusionFilter

func GPUExclusionFilter(nodeResources NodeResources, sharepod *faasv1.SharePod)

func IsPodActive

func IsPodActive(p *v1.Pod) bool

func IsPodHot

func IsPodHot(p *v1.Pod) bool

func PodKey

func PodKey(pod *v1.Pod) string

PodKey returns a key unique to the given pod within a cluster. It's used so we consistently use the same key scheme in this module. It does exactly what cache.MetaNamespaceKeyFunc would have done except there's not possibility for error since we know the exact type.

func RandStr

func RandStr(length int) string

func ScheduleAlgorithmBestFit

func ScheduleAlgorithmBestFit(isGPUPod bool, gpu_request float64, gpu_mem int64, sharepod *faasv1.SharePod, nodeResources NodeResources) (schedNodeName string, schedGPUID string)

func StartConfigManager

func StartConfigManager(stopCh <-chan struct{}, kc kubernetes.Interface) error

func UpdateNodeGPUInfo

func UpdateNodeGPUInfo(nodeName string, uuid2mem *map[string]string)

Types

type ActivePodsWithRanks

type ActivePodsWithRanks struct {
	Pods []*v1.Pod

	// Rank is a ranking of pods.  This ranking is used during sorting when
	// comparing two pods that are both scheduled, in the same phase, and
	// having the same ready status.
	Rank []int
	// Now is a reference timestamp for doing logarithmic timestamp comparisons.
	// If zero, comparison happens without scaling.
	Now metav1.Time
}

ActivePodsWithRanks is a sortable list of pods and a list of corresponding ranks which will be considered during sorting. The two lists must have equal length. After sorting, the pods will be ordered as follows, applying each rule in turn until one matches:

1. If only one of the pods is assigned to a node, the pod that is not assigned comes before the pod that is.
2. If the pods' phases differ, a pending pod comes before a pod whose phase is unknown, and a pod whose phase is unknown comes before a running pod.
3. If exactly one of the pods is ready, the pod that is not ready comes before the ready pod.
4. If controller.kubernetes.io/pod-deletion-cost annotation is set, then the pod with the lower value will come first.
5. If the pods' ranks differ, the pod with greater rank comes before the pod with lower rank.
6. If both pods are ready but have not been ready for the same amount of time, the pod that has been ready for a shorter amount of time comes before the pod that has been ready for longer.
7. If one pod has a container that has restarted more than any container in the other pod, the pod with the container with more restarts comes before the other pod.
8. If the pods' creation times differ, the pod that was created more recently comes before the older pod.

In 6 and 8, times are compared in a logarithmic scale. This allows a level of randomness among equivalent Pods when sorting. If two pods have the same logarithmic rank, they are sorted by UUID to provide a pseudorandom order.

If none of these rules matches, the second pod comes before the first pod.

The intention of this ordering is to put pods that should be preferred for deletion first in the list.

func (ActivePodsWithRanks) Len

func (s ActivePodsWithRanks) Len() int

func (ActivePodsWithRanks) Less

func (s ActivePodsWithRanks) Less(i, j int) bool

Less compares two pods with corresponding ranks and returns true if the first one should be preferred for deletion.

func (ActivePodsWithRanks) Swap

func (s ActivePodsWithRanks) Swap(i, j int)

type ClientStatus

type ClientStatus string

const (
	ClientReady    ClientStatus = "Ready"
	ClientNotReady ClientStatus = "NotReady"
)

type Controller

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

func NewController

func NewController(
	kubeclient kubernetes.Interface,
	kubeshareclient clientset.Interface,
	nodeInformer coreinformers.NodeInformer,
	podInformer coreinformers.PodInformer,

	kubeshareInformer informers.SharePodInformer) *Controller

NewController returns a new sample controller

func (*Controller) Run

func (c *Controller) Run(threadiness int, stopCh <-chan struct{}) error

Try out without stopCh, rather with context

type GPUInfo

type GPUInfo struct {
	UUID    string
	Usage   float64
	Mem     int64
	PodList *list.List
}

type GPUResource

type GPUResource struct {
	GPUFreeReq int64
	// GPUFreeMem in bytes
	GPUFreeMem int64

	GPUAffinityTags     []string
	GPUAntiAffinityTags []string
	// len(GPUExclusionTags) should be only one
	GPUExclusionTags []string
}

func (*GPUResource) DeepCopy

func (this *GPUResource) DeepCopy() *GPUResource

type NameList

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

TODO here to extend to new Struct for multil access for different sharepod

type NodeInfo

type NodeInfo struct {
	// GPUID -> GPU
	GPUID2GPU map[string]*GPUInfo
	// UUID -> Port (string)
	UUID2Port map[string]string

	// port in use
	PodManagerPortBitmap *bitmap.RRBitmap
	PodIP                string
}

type NodeResource

type NodeResource struct {
	CpuTotal int64
	MemTotal int64
	GpuTotal int
	// GpuMemTotal in bytes
	GpuMemTotal int64
	CpuFree     int64
	MemFree     int64
	/* Available GPU calculate */
	// Total GPU count - Pods using nvidia.com/gpu
	GpuFreeCount int
	// GPUs available usage (1.0 - SharePod usage)
	// GPUID to integer index mapping
	GpuFree map[string]*GPUResource
}

func (*NodeResource) DeepCopy

func (this *NodeResource) DeepCopy() *NodeResource

type NodeResources

type NodeResources map[string]*NodeResource

NodeResources: Available resources in cluster to schedule Training Jobs

func (*NodeResources) DeepCopy

func (this *NodeResources) DeepCopy() *NodeResources

func (*NodeResources) PrintMe

func (this *NodeResources) PrintMe()

type NodeStatus

type NodeStatus struct {
	Conn          net.Conn
	LastHeartbeat time.Time
	ClientStatus  ClientStatus
}

type PodRequest

type PodRequest struct {
	Key            string
	Request        float64
	Limit          float64
	Memory         int64
	PodManagerPort int
}

func FindInQueue

func FindInQueue(key string, pl *list.List) (*PodRequest, bool)

type RealPodControl

type RealPodControl struct {
	KubeClient clientset.Interface
	Recorder   record.EventRecorder
}

type SharepodProbes

type SharepodProbes struct {
	Liveness  *v1.Probe
	Readiness *v1.Probe
}