slurm

Documentation

Overview

Package job contains code for accessing compute resources via Slurm.

Package slurm contains code for accessing compute resources via Slurm.

Package job contains code for accessing compute resources via Slurm.

Index

• Variables
• func AllocatableResources(ctx context.Context) corev1.ResourceList
• func CancelJob(args string) (string, error)
• func ConnectionOK() bool
• func GetJobID(pod *corev1.Pod) string
• func HasJobID(pod *corev1.Pod) bool
• func SetContainerStatusID(status *corev1.ContainerStatus, typedValue string)
• func SetPodID(pod *corev1.Pod, idType JobIDType, value string)
• func SubmitJob(scriptFile string) (string, error)
• func TotalResources() corev1.ResourceList
• type JobIDType
• type NodeInfo
  • func (i NodeInfo) ResourceList() corev1.ResourceList
• type Stats

Variables

var ErrInvalidJob = errors.New("invalid job id")

var ErrRety = errors.New("retry later")

var ExcludeNodes = "--exclude="

ExcludeNodes EXISTS ONLY FOR DEBUGGING PURPOSES of Inotify on NFS.

var NewUserEnv = "--get-user-env=10L"

NewUserEnv is used to generate the /run/user/ folder required by cgroups. The optional mode value control the "su" options. With a mode value of "S", "su" is executed without the "-" option. With a mode value of "L", "su" is executed with the "-" option, replicating the login environment.

var Signal = "--signal=TERM"

var SignalChildren = "--full"

SignalChildren signals the submission script and the pause environment. SignalChildren signals the batch script and its children processes (pause).

var SignalParentOnly = "--batch"

SignalParentOnly signals only the submission script. https://slurm.schedmd.com/scancel.html#OPT_batch

var Slurm struct {
	SubmitCmd string
	CancelCmd string
	StatsCmd  string
}

Slurm represents a SLURM installation.

Functions

func AllocatableResources

func AllocatableResources(ctx context.Context) corev1.ResourceList

func CancelJob

func CancelJob(args string) (string, error)

func ConnectionOK

func ConnectionOK() bool

ConnectionOK return true if HPK maintains connection with the Slurm manager. Otherwise, it returns false.

func GetJobID

func GetJobID(pod *corev1.Pod) string

func HasJobID

func HasJobID(pod *corev1.Pod) bool

func SetContainerStatusID

func SetContainerStatusID(status *corev1.ContainerStatus, typedValue string)

func SetPodID

func SetPodID(pod *corev1.Pod, idType JobIDType, value string)

func SubmitJob

func SubmitJob(scriptFile string) (string, error)

func TotalResources

func TotalResources() corev1.ResourceList

Types

type JobIDType

type JobIDType string

const (
	JobIDTypeInstance JobIDType = "instance://"

	JobIDTypeProcess JobIDType = "pid://"

	JobIDTypeSlurm JobIDType = "slurm://"

	JobIDTypeEmpty JobIDType = "Empty"
)

type NodeInfo

type NodeInfo struct {
	Architecture  string `json:"architecture"`
	KernelVersion string `json:"operating_system"`

	Name     string `json:"name"`
	CPUs     uint64 `json:"cpus"`
	CPUCores uint64 `json:"cores"`

	EphemeralStorage uint64 `json:"temporary_disk"`

	// FreeMemory ... reported in MegaBytes
	//[TODO: temporarily changed it to int64 due to sometimes slurm declares freememory as "-2"]
	FreeMemory int64    `json:"free_memory"`
	Partitions []string `json:"partitions"`
}

func (NodeInfo) ResourceList

func (i NodeInfo) ResourceList() corev1.ResourceList

ResourceList converts the Slurm-reported stats into Kubernetes-Stats.

type Stats

type Stats struct {
	Nodes []NodeInfo `json:"nodes"`
}