v1

package
v0.17.0-rc.1 Latest Latest
Warning

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

 Go to latest Published: Oct 28, 2024 License: Apache-2.0 Imports: 14 Imported by: 1

Documentation

Index

Constants

View Source 
const (
	ResourceNamePrefix              = "nvidia.com"
	DefaultSharedResourceNameSuffix = ".shared"
	MaxResourceNameLength           = 63
)

Constants related to resource names

View Source 
const (
	MigStrategyNone   = "none"
	MigStrategySingle = "single"
	MigStrategyMixed  = "mixed"
)

Constants representing the various MIG strategies

View Source 
const (
	DeviceListStrategyEnvVar         = "envvar"
	DeviceListStrategyVolumeMounts   = "volume-mounts"
	DeviceListStrategyCDIAnnotations = "cdi-annotations"
	DeviceListStrategyCDICRI         = "cdi-cri"
)

Constants to represent the various device list strategies

View Source 
const (
	DeviceIDStrategyUUID  = "uuid"
	DeviceIDStrategyIndex = "index"
)

Constants to represent the various device id strategies

View Source 
const (
	DefaultCDIAnnotationPrefix = cdiapi.AnnotationPrefix
	DefaultNvidiaCTKPath       = "/usr/bin/nvidia-ctk"
	DefaultContainerDriverRoot = "/driver-root"
)

Constants related to generating CDI specifications

View Source 
const (
	SharingStrategyMPS         = SharingStrategy("mps")
	SharingStrategyNone        = SharingStrategy("none")
	SharingStrategyTimeSlicing = SharingStrategy("time-slicing")
)
View Source 
const (
	ImexChannelEnvVar = "NVIDIA_IMEX_CHANNELS"
)
View Source 
const Version = "v1"

Version indicates the version of the 'Config' struct used to hold configuration information.

Variables

This section is empty.

Functions

func AssertChannelIDsValid added in v0.17.0 

func AssertChannelIDsValid(ids []int) error

AssertChannelIDsIsValid checks whether the specified list of channel IDs is valid.

func DisableResourceNamingInConfig added in v0.15.0 

func DisableResourceNamingInConfig(logger logger, config *Config)

DisableResourceNamingInConfig temporarily disable the resource renaming feature of the plugin. This may be reenabled in a future release.

Types

type CommandLineFlags 

type CommandLineFlags struct {
	MigStrategy             *string                 `json:"migStrategy"                yaml:"migStrategy"`
	FailOnInitError         *bool                   `json:"failOnInitError"            yaml:"failOnInitError"`
	MpsRoot                 *string                 `json:"mpsRoot,omitempty"          yaml:"mpsRoot,omitempty"`
	NvidiaDriverRoot        *string                 `json:"nvidiaDriverRoot,omitempty" yaml:"nvidiaDriverRoot,omitempty"`
	NvidiaDevRoot           *string                 `json:"nvidiaDevRoot,omitempty"    yaml:"nvidiaDevRoot,omitempty"`
	GDSEnabled              *bool                   `json:"gdsEnabled"                 yaml:"gdsEnabled"`
	MOFEDEnabled            *bool                   `json:"mofedEnabled"               yaml:"mofedEnabled"`
	UseNodeFeatureAPI       *bool                   `json:"useNodeFeatureAPI"          yaml:"useNodeFeatureAPI"`
	DeviceDiscoveryStrategy *string                 `json:"deviceDiscoveryStrategy"    yaml:"deviceDiscoveryStrategy"`
	Plugin                  *PluginCommandLineFlags `json:"plugin,omitempty"           yaml:"plugin,omitempty"`
	GFD                     *GFDCommandLineFlags    `json:"gfd,omitempty"              yaml:"gfd,omitempty"`
}

CommandLineFlags holds the list of command line flags used to configure the device plugin and GFD.

type Config 

type Config struct {
	Version   string    `json:"version"             yaml:"version"`
	Flags     Flags     `json:"flags,omitempty"     yaml:"flags,omitempty"`
	Resources Resources `json:"resources,omitempty" yaml:"resources,omitempty"`
	Sharing   Sharing   `json:"sharing,omitempty"   yaml:"sharing,omitempty"`
	Imex      Imex      `json:"imex,omitempty"      yaml:"imex,omitempty"`
}

Config is a versioned struct used to hold configuration information.

func NewConfig 

func NewConfig(c *cli.Context, flags []cli.Flag) (*Config, error)

NewConfig builds out a Config struct from a config file (or command line flags). The data stored in the config will be populated in order of precedence from (1) command line, (2) environment variable, (3) config file.

type DeviceListStrategies added in v0.14.0 

type DeviceListStrategies map[string]bool

DeviceListStrategies defines which strategies are enabled and should be used when passing the device list to the container runtime.

func NewDeviceListStrategies added in v0.14.0 

func NewDeviceListStrategies(strategies []string) (DeviceListStrategies, error)

NewDeviceListStrategies constructs a new DeviceListStrategy

func (DeviceListStrategies) AllCDIEnabled added in v0.16.0 

func (s DeviceListStrategies) AllCDIEnabled() bool

AllCDIEnabled returns whether all strategies being used require CDI.

func (DeviceListStrategies) AnyCDIEnabled added in v0.16.0 

func (s DeviceListStrategies) AnyCDIEnabled() bool

AnyCDIEnabled returns whether any of the strategies being used require CDI.

func (DeviceListStrategies) Includes added in v0.14.0 

func (s DeviceListStrategies) Includes(strategy string) bool

Includes returns whether the given strategy is present in the set of strategies.

type Duration 

type Duration time.Duration

Duration wraps a time.Duration function with custom JSON marshaling/unmarshaling

func (Duration) MarshalJSON 

func (d Duration) MarshalJSON() ([]byte, error)

MarshalJSON marshals 'Duration' to its raw bytes representation

func (*Duration) UnmarshalJSON 

func (d *Duration) UnmarshalJSON(b []byte) error

UnmarshalJSON unmarshals raw bytes into a 'Duration' type.

type Flags 

type Flags struct {
	CommandLineFlags
}

Flags holds the full list of flags used to configure the device plugin and GFD.

func (*Flags) UpdateFromCLIFlags 

func (f *Flags) UpdateFromCLIFlags(c *cli.Context, flags []cli.Flag)

UpdateFromCLIFlags updates Flags from settings in the cli Flags if they are set.

type GFDCommandLineFlags 

type GFDCommandLineFlags struct {
	Oneshot         *bool     `json:"oneshot"         yaml:"oneshot"`
	NoTimestamp     *bool     `json:"noTimestamp"     yaml:"noTimestamp"`
	SleepInterval   *Duration `json:"sleepInterval"   yaml:"sleepInterval"`
	OutputFile      *string   `json:"outputFile"      yaml:"outputFile"`
	MachineTypeFile *string   `json:"machineTypeFile" yaml:"machineTypeFile"`
}

GFDCommandLineFlags holds the list of command line flags specific to GFD.

type Imex added in v0.17.0 

type Imex struct {
	// ChannelIDs defines a list of channel IDs to inject into containers that request NVIDIA devices.
	// If a channel ID is specified and the associated channel device node exists, the corresponding
	// channel will be added to the ContainerAllocateResponse for containers with access to NVIDIA
	// devices.
	ChannelIDs []int `json:"channelIDs,omitempty" yaml:"channelIDs,omitempty"`
	// Required specifies whether the requested IMEX channel IDs are required or not.
	// If a channel is required, it is expected to exist as the device plugin starts.
	// If it is not required its injection is skipped if the device nodes do not exist or if its
	// existence cannot be queried.
	Required bool `json:"required,omitempty" yaml:"required,omitempty"`
}

Imex stores the configuration options for fabric-attached devices.

type PluginCommandLineFlags 

type PluginCommandLineFlags struct {
	PassDeviceSpecs     *bool                   `json:"passDeviceSpecs"     yaml:"passDeviceSpecs"`
	DeviceListStrategy  *deviceListStrategyFlag `json:"deviceListStrategy"  yaml:"deviceListStrategy"`
	DeviceIDStrategy    *string                 `json:"deviceIDStrategy"    yaml:"deviceIDStrategy"`
	CDIAnnotationPrefix *string                 `json:"cdiAnnotationPrefix" yaml:"cdiAnnotationPrefix"`
	NvidiaCTKPath       *string                 `json:"nvidiaCTKPath"       yaml:"nvidiaCTKPath"`
	ContainerDriverRoot *string                 `json:"containerDriverRoot" yaml:"containerDriverRoot"`
}

PluginCommandLineFlags holds the list of command line flags specific to the device plugin.

type ReplicatedDeviceRef 

type ReplicatedDeviceRef string

ReplicatedDeviceRef can either be a full GPU index, a MIG index, or a UUID (full GPU or MIG)

func (ReplicatedDeviceRef) IsGPUIndex 

func (d ReplicatedDeviceRef) IsGPUIndex() bool

IsGPUIndex checks if a ReplicatedDeviceRef is a full GPU index

func (ReplicatedDeviceRef) IsGpuUUID 

func (d ReplicatedDeviceRef) IsGpuUUID() bool

IsGpuUUID checks if a ReplicatedDeviceRef is a GPU UUID A GPU UUID must be of the form GPU-b1028956-cfa2-0990-bf4a-5da9abb51763

func (ReplicatedDeviceRef) IsMigIndex 

func (d ReplicatedDeviceRef) IsMigIndex() bool

IsMigIndex checks if a ReplicatedDeviceRef is a MIG index

func (ReplicatedDeviceRef) IsMigUUID 

func (d ReplicatedDeviceRef) IsMigUUID() bool

IsMigUUID checks if a ReplicatedDeviceRef is a MIG UUID A MIG UUID can be of one of two forms:

  • MIG-b1028956-cfa2-0990-bf4a-5da9abb51763
  • MIG-GPU-b1028956-cfa2-0990-bf4a-5da9abb51763/3/0

func (ReplicatedDeviceRef) IsUUID 

func (d ReplicatedDeviceRef) IsUUID() bool

IsUUID checks if a ReplicatedDeviceRef is a UUID

type ReplicatedDevices 

type ReplicatedDevices struct {
	All   bool
	Count int
	List  []ReplicatedDeviceRef
}

ReplicatedDevices encapsulates the set of devices that should be replicated for a given resource. This struct should be treated as a 'union' and only one of the fields in this struct should be set at any given time.

func (*ReplicatedDevices) MarshalJSON 

func (s *ReplicatedDevices) MarshalJSON() ([]byte, error)

MarshalJSON marshals ReplicatedDevices to its raw bytes representation

func (*ReplicatedDevices) UnmarshalJSON 

func (s *ReplicatedDevices) UnmarshalJSON(b []byte) error

UnmarshalJSON unmarshals raw bytes into a 'ReplicatedDevices' struct.

type ReplicatedResource 

type ReplicatedResource struct {
	Name     ResourceName      `json:"name"             yaml:"name"`
	Rename   ResourceName      `json:"rename,omitempty" yaml:"rename,omitempty"`
	Devices  ReplicatedDevices `json:"devices"          yaml:"devices,flow"`
	Replicas int               `json:"replicas"         yaml:"replicas"`
}

ReplicatedResource represents a resource to be replicated.

func (*ReplicatedResource) UnmarshalJSON 

func (s *ReplicatedResource) UnmarshalJSON(b []byte) error

UnmarshalJSON unmarshals raw bytes into a 'ReplicatedResource' struct.

type ReplicatedResources added in v0.15.0 

type ReplicatedResources struct {
	RenameByDefault            bool                 `json:"renameByDefault,omitempty"            yaml:"renameByDefault,omitempty"`
	FailRequestsGreaterThanOne bool                 `json:"failRequestsGreaterThanOne,omitempty" yaml:"failRequestsGreaterThanOne,omitempty"`
	Resources                  []ReplicatedResource `json:"resources,omitempty"                  yaml:"resources,omitempty"`
}

ReplicatedResources defines generic options for replicating devices.

func (*ReplicatedResources) UnmarshalJSON added in v0.15.0 

func (s *ReplicatedResources) UnmarshalJSON(b []byte) error

UnmarshalJSON unmarshals raw bytes into a 'ReplicatedResources' struct.

type Resource 

type Resource struct {
	Pattern ResourcePattern `json:"pattern" yaml:"pattern"`
	Name    ResourceName    `json:"name"    yaml:"name"`
}

Resource pairs a pattern matcher with a resource name.

func NewResource 

func NewResource(pattern, name string) (*Resource, error)

NewResource builds a resource from a name and pattern

func (*Resource) UnmarshalJSON 

func (r *Resource) UnmarshalJSON(b []byte) error

UnmarshalJSON unmarshals raw bytes into a 'Resource' struct.

type ResourceName 

type ResourceName string

ResourceName represents a valid resource name in Kubernetes

func NewResourceName 

func NewResourceName(n string) (ResourceName, error)

NewResourceName builds a resource name from the standard prefix and a name. An error is returned if the format is incorrect.

func (ResourceName) DefaultSharedRename 

func (r ResourceName) DefaultSharedRename() ResourceName

DefaultSharedRename returns the default renaming to apply when this resource is shared

func (ResourceName) Split 

func (r ResourceName) Split() (string, string)

Split splits a full resource name into prefix and name

func (*ResourceName) UnmarshalJSON 

func (r *ResourceName) UnmarshalJSON(b []byte) error

UnmarshalJSON unmarshals raw bytes into a 'ResourceName' type.

type ResourcePattern 

type ResourcePattern string

ResourcePattern is used to match a resource name to a specific pattern

func (ResourcePattern) Matches 

func (p ResourcePattern) Matches(s string) bool

Matches checks if the provided string matches the ResourcePattern or not.

type Resources 

type Resources struct {
	GPUs []Resource `json:"gpus"           yaml:"gpus"`
	MIGs []Resource `json:"mig,omitempty"  yaml:"mig,omitempty"`
}

Resources lists full GPUs and MIG devices separately.

func (*Resources) AddGPUResource 

func (r *Resources) AddGPUResource(pattern, name string) error

AddGPUResource adds a GPU resource to the list of GPU resources.

func (*Resources) AddMIGResource 

func (r *Resources) AddMIGResource(pattern, name string) error

AddMIGResource adds a MIG resource to the list of MIG resources.

type Sharing 

type Sharing struct {
	// TimeSlicing defines the set of replicas to be made for timeSlicing available resources.
	TimeSlicing ReplicatedResources `json:"timeSlicing,omitempty" yaml:"timeSlicing,omitempty"`
	// MPS defines the set of replicas to be shared using MPS
	MPS *ReplicatedResources `json:"mps,omitempty"         yaml:"mps,omitempty"`
}

Sharing encapsulates the set of sharing strategies that are supported.

func (*Sharing) ReplicatedResources added in v0.15.0 

func (s *Sharing) ReplicatedResources() *ReplicatedResources

ReplicatedResources returns the resources associated with the active sharing strategy.

func (*Sharing) SharingStrategy added in v0.15.0 

func (s *Sharing) SharingStrategy() SharingStrategy

SharingStrategy returns the active sharing strategy.

type SharingStrategy added in v0.15.0 

type SharingStrategy string

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.