gpuschedulerplugin

Documentation

Index

• Constants
• Variables
• func AddResourcesToNodeTreeCache(nodeName string, nodeResources types.ResourceList)
• func ConvertToBestGPURequests(podInfo *types.PodInfo) bool
• func RemoveNodeFromNodeTreeCache(nodeName string)
• func SetGPUReqs(cont *types.ContainerInfo)
• func TranslateGPUContainerResources(alloc types.ResourceList, cont types.ContainerInfo) types.ResourceList
• func TranslateGPUResources(neededGPUs int64, nodeResources types.ResourceList, ...) types.ResourceList
• func TranslatePodGPUResources(nodeInfo *types.NodeInfo, podInfo *types.PodInfo) (error, bool)
• type NvidiaGPUScheduler
  • func (ns *NvidiaGPUScheduler) AddNode(nodeName string, nodeInfo *types.NodeInfo)
  • func (ns *NvidiaGPUScheduler) GetName() string
  • func (ns *NvidiaGPUScheduler) PodAllocate(nodeInfo *types.NodeInfo, podInfo *types.PodInfo) error
  • func (ns *NvidiaGPUScheduler) PodFitsDevice(nodeInfo *types.NodeInfo, podInfo *types.PodInfo, fillAllocateFrom bool) (bool, []devicescheduler.PredicateFailureReason, float64)
  • func (ns *NvidiaGPUScheduler) RemoveNode(nodeName string)
  • func (ns *NvidiaGPUScheduler) ReturnPodResources(nodeInfo *types.NodeInfo, podInfo *types.PodInfo) error
  • func (ns *NvidiaGPUScheduler) TakePodResources(nodeInfo *types.NodeInfo, podInfo *types.PodInfo) error
  • func (ns *NvidiaGPUScheduler) UsingGroupScheduler() bool

Constants

const (
	// auto topology generation "0" means default (everything in its own group)
	GPUTopologyGeneration types.ResourceName = "gpu/gpu-generate-topology"
)

Variables

var NodeCacheMap = make(map[*sctypes.SortedTreeNode]treeInfo)

var NodeLocationMap = make(map[string]*sctypes.SortedTreeNode)

Functions

func AddResourcesToNodeTreeCache

func AddResourcesToNodeTreeCache(nodeName string, nodeResources types.ResourceList)

func ConvertToBestGPURequests

func ConvertToBestGPURequests(podInfo *types.PodInfo) bool

find total GPUs needed

func RemoveNodeFromNodeTreeCache

func RemoveNodeFromNodeTreeCache(nodeName string)

func SetGPUReqs

func SetGPUReqs(cont *types.ContainerInfo)

func TranslateGPUContainerResources

func TranslateGPUContainerResources(alloc types.ResourceList, cont types.ContainerInfo) types.ResourceList

func TranslateGPUResources

func TranslateGPUResources(neededGPUs int64, nodeResources types.ResourceList, containerRequests types.ResourceList) types.ResourceList

TranslateGPUResources translates GPU resources to max level advertised by the node

func TranslatePodGPUResources

func TranslatePodGPUResources(nodeInfo *types.NodeInfo, podInfo *types.PodInfo) (error, bool)

Types

type NvidiaGPUScheduler

type NvidiaGPUScheduler struct {
}

func (*NvidiaGPUScheduler) AddNode

func (ns *NvidiaGPUScheduler) AddNode(nodeName string, nodeInfo *types.NodeInfo)

force translation to two levels

func (*NvidiaGPUScheduler) GetName

func (ns *NvidiaGPUScheduler) GetName() string

func (*NvidiaGPUScheduler) PodAllocate

func (ns *NvidiaGPUScheduler) PodAllocate(nodeInfo *types.NodeInfo, podInfo *types.PodInfo) error

func (*NvidiaGPUScheduler) PodFitsDevice

func (ns *NvidiaGPUScheduler) PodFitsDevice(nodeInfo *types.NodeInfo, podInfo *types.PodInfo, fillAllocateFrom bool) (bool, []devicescheduler.PredicateFailureReason, float64)

func (*NvidiaGPUScheduler) RemoveNode

func (ns *NvidiaGPUScheduler) RemoveNode(nodeName string)

func (*NvidiaGPUScheduler) ReturnPodResources

func (ns *NvidiaGPUScheduler) ReturnPodResources(nodeInfo *types.NodeInfo, podInfo *types.PodInfo) error

func (*NvidiaGPUScheduler) TakePodResources

func (ns *NvidiaGPUScheduler) TakePodResources(nodeInfo *types.NodeInfo, podInfo *types.PodInfo) error

func (*NvidiaGPUScheduler) UsingGroupScheduler

func (ns *NvidiaGPUScheduler) UsingGroupScheduler() bool