appconfig

Documentation

Index

• Constants
• type Config
• type DeviceOptions
• type DumpConfig
• type KubernetesGPUIDType

Constants

const (
	GPUUID     KubernetesGPUIDType = "uid"
	DeviceName KubernetesGPUIDType = "device-name"

	NvidiaResourceName      = "nvidia.com/gpu"
	NvidiaMigResourcePrefix = "nvidia.com/mig-"
	MIG_UUID_PREFIX         = "MIG-"
)

Types

type Config

type Config struct {
	CollectorsFile                   string
	Address                          string
	CollectInterval                  int
	Kubernetes                       bool
	KubernetesEnablePodLabels        bool
	KubernetesEnablePodUID           bool
	KubernetesGPUIdType              KubernetesGPUIDType
	KubernetesPodLabelAllowlistRegex []string // Regex patterns for filtering pod labels
	KubernetesPodLabelCacheSize      int      // Maximum number of label keys to cache (<=0 means default size)
	CollectDCP                       bool
	UseOldNamespace                  bool
	UseRemoteHE                      bool
	RemoteHEInfo                     string
	GPUDeviceOptions                 DeviceOptions
	SwitchDeviceOptions              DeviceOptions
	CPUDeviceOptions                 DeviceOptions
	NoHostname                       bool
	UseFakeGPUs                      bool
	ConfigMapData                    string
	MetricGroups                     []dcgm.MetricGroup
	WebSystemdSocket                 bool
	WebConfigFile                    string
	XIDCountWindowSize               int
	ReplaceBlanksInModelName         bool
	Debug                            bool
	ClockEventsCountWindowSize       int
	EnableDCGMLog                    bool
	DCGMLogLevel                     string
	PodResourcesKubeletSocket        string
	HPCJobMappingDir                 string
	NvidiaResourceNames              []string
	KubernetesVirtualGPUs            bool
	DumpConfig                       DumpConfig // Configuration for file-based dumps
	KubernetesEnableDRA              bool
	DisableStartupValidate           bool
	EnableGPUBindUnbindWatch         bool          // Enable GPU bind/unbind event monitoring
	GPUBindUnbindPollInterval        time.Duration // Poll interval for GPU bind/unbind events
	EnablePprof                      bool          // Enable /debug/pprof/ HTTP endpoints
}

type DeviceOptions

type DeviceOptions struct {
	Flex       bool  // If true, then monitor all GPUs if MIG mode is disabled or all GPU instances if MIG is enabled.
	MajorRange []int // The indices of each GPU/NvSwitch to monitor, or -1 to monitor all
	MinorRange []int // The indices of each GPUInstance/NvLink to monitor, or -1 to monitor all
}

type DumpConfig

type DumpConfig struct {
	Enabled     bool   `yaml:"enabled" json:"enabled"`         // Enable file-based dumps
	Directory   string `yaml:"directory" json:"directory"`     // Directory to store dump files
	Retention   int    `yaml:"retention" json:"retention"`     // Retention period in hours (0 = no cleanup)
	Compression bool   `yaml:"compression" json:"compression"` // Use gzip compression for dump files
}

DumpConfig controls file-based debugging dumps

type KubernetesGPUIDType

type KubernetesGPUIDType string