dcgmexporter

type ComputeInstanceInfo struct {
	InstanceInfo dcgm.MigEntityInfo
	ProfileName  string
	EntityId     uint
}

type Config struct {
	CollectorsFile      string
	Address             string
	CollectInterval     int
	Kubernetes          bool
	KubernetesGPUIdType KubernetesGPUIDType
	CollectDCP          bool
	UseOldNamespace     bool
	UseRemoteHE         bool
	RemoteHEInfo        string
	GPUDevices          DeviceOptions
	SwitchDevices       DeviceOptions
	NoHostname          bool
	UseFakeGpus         bool
	ConfigMapData       string
	MetricGroups        []dcgm.MetricGroup
}

type Counter struct {
	FieldID   dcgm.Short
	FieldName string
	PromType  string
	Help      string
}

type DCGMCollector struct {
	Counters        []Counter
	DeviceFields    []dcgm.Short
	Cleanups        []func()
	UseOldNamespace bool
	SysInfo         SystemInfo
	Hostname        string
}

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 GpuInfo struct {
	DeviceInfo   dcgm.Device
	GpuInstances []GpuInstanceInfo
	MigEnabled   bool
}

type GpuInstanceInfo struct {
	Info             dcgm.MigEntityInfo
	ProfileName      string
	EntityId         uint
	ComputeInstances []ComputeInstanceInfo
}

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

type KubernetesGPUIDType string

type Metric struct {
	Counter *Counter
	Value   string

	GPU          string
	GPUUUID      string
	GPUDevice    string
	GPUModelName string

	UUID string

	MigProfile    string
	GPUInstanceID string
	Hostname      string

	Labels     *map[string]string
	Attributes map[string]string
}

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

type MetricsServer struct {
	sync.Mutex
	// contains filtered or unexported fields
}

type MonitoringInfo struct {
	Entity       dcgm.GroupEntityPair
	DeviceInfo   dcgm.Device
	InstanceInfo *GpuInstanceInfo
	ParentId     uint
}

type PodInfo struct {
	Name      string
	Namespace string
	Container string
}

type PodMapper struct {
	Config *Config
}

type SwitchInfo struct {
	EntityId uint
	NvLinks  []dcgm.NvLinkStatus
}

type SystemInfo struct {
	GpuCount uint
	Gpus     [dcgm.MAX_NUM_DEVICES]GpuInfo

	InfoType dcgm.Field_Entity_Group
	Switches []SwitchInfo
	// contains filtered or unexported fields
}

type Transform interface {
	Process(metrics [][]Metric, sysInfo SystemInfo) error
	Name() string
}