v1

type ComponentEvents struct {
	Component string    `json:"component"`
	StartTime time.Time `json:"startTime"`
	EndTime   time.Time `json:"endTime"`
	Events    Events    `json:"events"`
}

type ComponentHealthStates struct {
	Component string       `json:"component"`
	States    HealthStates `json:"states"`
}

type ComponentInfo struct {
	Component string    `json:"component"`
	StartTime time.Time `json:"startTime"`
	EndTime   time.Time `json:"endTime"`
	Info      Info      `json:"info"`
}

type ComponentMetrics struct {
	Component string  `json:"component"`
	Metrics   Metrics `json:"metrics"`
}

type ComponentType string

type Event struct {
	// Component represents which component generated the event.
	Component string `json:"component,omitempty"`

	// Time represents when the event happened.
	Time metav1.Time `json:"time"`

	// Name represents the name of the event.
	Name string `json:"name,omitempty"`

	// Type represents the type of the event.
	Type EventType `json:"type,omitempty"`

	// Message represents the detailed message of the event.
	Message string `json:"message,omitempty"`
}

type EventType string

type Events []Event

type GPUdComponentEvents []ComponentEvents

type GPUdComponentHealthStates []ComponentHealthStates

type GPUdComponentInfos []ComponentInfo

type GPUdComponentMetrics []ComponentMetrics

type GossipRequest struct {
	MachineID   string       `json:"machineID"`
	MachineInfo *MachineInfo `json:"machineInfo,omitempty"`
}

type GossipResponse struct {
	Status string `json:"status,omitempty"`
	Error  string `json:"error,omitempty"`
}

type HealthState struct {
	// Time represents when the event happened.
	Time metav1.Time `json:"time"`

	// Component represents the component name.
	Component string `json:"component,omitempty"`
	// ComponentType represents the type of the component.
	// It is either "" (just 'component') or "custom-plugin".
	ComponentType ComponentType `json:"component_type,omitempty"`

	// Name is the name of the state,
	// can be different from the component name.
	Name string `json:"name,omitempty"`

	// RunMode is the run mode of the state.
	// It can be "manual" that requires manual trigger to run the check.
	// Or it can be empty that runs the check periodically.
	RunMode RunModeType `json:"run_mode,omitempty"`

	// Health represents the health level of the state,
	// including StateHealthy, StateUnhealthy and StateDegraded.
	// StateDegraded is similar to Unhealthy which also can trigger alerts
	// for users or operators, but what StateDegraded means is that the
	// issue detected does not affect users’ workload.
	Health HealthStateType `json:"health,omitempty"`

	// Reason represents what happened or detected by GPUd if it isn’t healthy.
	Reason string `json:"reason,omitempty"`

	// Error represents the detailed error information, which will be shown
	// as More Information to help analyze why it isn’t healthy.
	Error string `json:"error,omitempty"`

	// SuggestedActions represents the suggested actions to mitigate the issue.
	SuggestedActions *SuggestedActions `json:"suggested_actions,omitempty"`

	// ExtraInfo represents the extra information of the state.
	ExtraInfo map[string]string `json:"extra_info,omitempty"`

	// RawOutput represents the raw output of the health checker.
	// e.g., If a custom plugin runs a Python script, the raw output
	// is the stdout/stderr of the script.
	// The maximum length of the raw output is 4096 bytes.
	RawOutput string `json:"raw_output,omitempty"`
}

type HealthStateType string

type HealthStates []HealthState

type Info struct {
	States  HealthStates `json:"states"`
	Events  Events       `json:"events"`
	Metrics Metrics      `json:"metrics"`
}

type JoinRequest struct {
	ID                 string `json:"id"`
	ClusterName        string `json:"cluster_name,omitempty"`
	PublicIP           string `json:"public_ip,omitempty"`
	PrivateIP          string `json:"private_ip,omitempty"`
	Provider           string `json:"provider,omitempty"`
	ProviderInstanceID string `json:"provider_instance_id,omitempty"`
	ProviderGPUShape   string `json:"provider_gpu_shape,omitempty"`
	TotalCPU           int64  `json:"total_cpu,omitempty"`
	NodeGroup          string `json:"node_group,omitempty"`
	ExtraInfo          string `json:"extra_info,omitempty"`
	Region             string `json:"region,omitempty"`
}

type JoinResponse struct {
	Error  string `json:"error"`
	Status string `json:"status"`
}

type LoginRequest struct {
	Token              string            `json:"token"`
	MachineID          string            `json:"machineID"`
	NodeGroup          string            `json:"nodeGroup,omitempty"`
	Network            *MachineNetwork   `json:"network,omitempty"`
	Location           *MachineLocation  `json:"location,omitempty"`
	Provider           string            `json:"provider"`
	ProviderInstanceID string            `json:"providerInstanceID"`
	MachineInfo        *MachineInfo      `json:"machineInfo,omitempty"`
	Resources          map[string]string `json:"resources,omitempty"`
}

type LoginResponse struct {
	// Status is the status of the login request.
	Status string `json:"status,omitempty"`

	// Error is the error message for the login request.
	// The error is only returned if the machine ID retrieval fails.
	Error string `json:"error,omitempty"`

	// MachineID is the ID of the machine generated by the login control plane.
	// Only set upon successful login.
	MachineID string `json:"machineID"`

	// Token is the token used to report data from the machine.
	Token string `json:"token,omitempty"`

	// ValidationResults is the validation results for the login request.
	// The validation results are done by the control plane.
	ValidationResults []ValidationResult `json:"validationResults,omitempty"`
}

type MachineCPUInfo struct {
	Type         string `json:"type,omitempty"`
	Manufacturer string `json:"manufacturer,omitempty"`
	Architecture string `json:"architecture,omitempty"`
	LogicalCores int64  `json:"logicalCores,omitempty"`
}

type MachineDiskDevice struct {
	Name       string   `json:"name,omitempty"`
	Type       string   `json:"type,omitempty"`
	Size       int64    `json:"size,omitempty"`
	Used       int64    `json:"used,omitempty"`
	Rota       bool     `json:"rota,omitempty"`
	Serial     string   `json:"serial,omitempty"`
	WWN        string   `json:"wwn,omitempty"`
	Vendor     string   `json:"vendor,omitempty"`
	Model      string   `json:"model,omitempty"`
	Rev        string   `json:"rev,omitempty"`
	MountPoint string   `json:"mountPoint,omitempty"`
	FSType     string   `json:"fsType,omitempty"`
	PartUUID   string   `json:"partUUID,omitempty"`
	Parents    []string `json:"parents,omitempty"`
	Children   []string `json:"children,omitempty"`
}

type MachineDiskInfo struct {
	BlockDevices []MachineDiskDevice `json:"blockDevices,omitempty"`
	// ContainerRootDisk is the disk device name that mounts the container root (such as "/var/lib/kubelet" mount point).
	ContainerRootDisk string `json:"containerRootDisk,omitempty"`
}

type MachineGPUInfo struct {
	// Product may be "NVIDIA-Graphics-Device" for NVIDIA GB200.
	Product string `json:"product,omitempty"`

	// Manufacturer is "NVIDIA" for NVIDIA GPUs (same as Brand).
	Manufacturer string `json:"manufacturer,omitempty"`

	// Architecture is "blackwell" for NVIDIA GB200.
	Architecture string `json:"architecture,omitempty"`

	Memory string `json:"memory,omitempty"`

	// GPUs is the GPU info of the machine.
	GPUs []MachineGPUInstance `json:"gpus,omitempty"`
}

type MachineGPUInstance struct {
	// UUID is the GPU UUID from the nvml API.
	// e.g., "GPU-46a3bbe2-3e87-3dde-b464-a03eba0c21d7"
	UUID string `json:"uuid,omitempty"`

	// BusID is the GPU bus ID from the nvml API.
	//  e.g., "0000:0f:00.0"
	BusID string `json:"busID,omitempty"`

	SN      string `json:"sn,omitempty"`
	MinorID string `json:"minorID,omitempty"`
	BoardID uint32 `json:"boardID,omitempty"`
}

type MachineInfo struct {
	// GPUdVersion represents the current version of GPUd
	GPUdVersion string `json:"gpudVersion,omitempty"`
	// GPUDriverVersion represents the current version of GPU driver installed
	GPUDriverVersion string `json:"gpuDriverVersion,omitempty"`
	// CUDAVersion represents the current version of cuda library.
	CUDAVersion string `json:"cudaVersion,omitempty"`
	// ContainerRuntime Version reported by the node through runtime remote API (e.g. containerd://1.4.2).
	ContainerRuntimeVersion string `json:"containerRuntimeVersion,omitempty"`
	// Kernel Version reported by the node from 'uname -r' (e.g. 3.16.0-0.bpo.4-amd64).
	KernelVersion string `json:"kernelVersion,omitempty"`
	// OS Image reported by the node from /etc/os-release (e.g. Debian GNU/Linux 7 (wheezy)).
	OSImage string `json:"osImage,omitempty"`
	// The Operating System reported by the node
	OperatingSystem string `json:"operatingSystem,omitempty"`
	// SystemUUID comes from https://github.com/google/cadvisor/blob/master/utils/sysfs/sysfs.go#L442
	SystemUUID string `json:"systemUUID,omitempty"`
	// MachineID is collected by GPUd. It comes from /etc/machine-id or /var/lib/dbus/machine-id
	MachineID string `json:"machineID,omitempty"`
	// BootID is collected by GPUd.
	BootID string `json:"bootID,omitempty"`
	// Hostname is the current host of machine
	Hostname string `json:"hostname,omitempty"`
	// Uptime represents when the machine up
	Uptime metav1.Time `json:"uptime,omitempty"`

	// CPUInfo is the CPU info of the machine.
	CPUInfo *MachineCPUInfo `json:"cpuInfo,omitempty"`
	// MemoryInfo is the memory info of the machine.
	MemoryInfo *MachineMemoryInfo `json:"memoryInfo,omitempty"`
	// GPUInfo is the GPU info of the machine.
	GPUInfo *MachineGPUInfo `json:"gpuInfo,omitempty"`
	// DiskInfo is the Disk info of the machine.
	DiskInfo *MachineDiskInfo `json:"diskInfo,omitempty"`
	// NICInfo is the network info of the machine.
	NICInfo *MachineNICInfo `json:"nicInfo,omitempty"`
}

type MachineLocation struct {
	Region string `json:"region,omitempty"`
	Zone   string `json:"zone,omitempty"`
}

type MachineMemoryInfo struct {
	TotalBytes uint64 `json:"totalBytes"`
}

type MachineNICInfo struct {
	// PrivateIPInterfaces is the private network interface info of the machine.
	PrivateIPInterfaces []MachineNetworkInterface `json:"privateIPInterfaces,omitempty"`
}

type MachineNetwork struct {
	// PublicIP is the public IP address of the machine.
	PublicIP string `json:"publicIP,omitempty"`
	// PrivateIP is the first private IP in IPv4 family,
	// detected from the local host.
	// May be overridden by the user with the private IP address.
	PrivateIP string `json:"privateIP,omitempty"`
}

type MachineNetworkInterface struct {
	// Interface is the network interface name of the machine.
	Interface string `json:"interface,omitempty"`

	// MAC is the MAC address of the machine.
	MAC string `json:"mac,omitempty"`

	// IP is the string representation of the netip.Addr of the machine.
	IP string `json:"ip,omitempty"`

	// Addr is the netip.Addr of the machine.
	Addr netip.Addr `json:"-"`
}

type Metric struct {
	UnixSeconds int64             `json:"unix_seconds"`
	Name        string            `json:"name"`
	Labels      map[string]string `json:"labels,omitempty"`
	Value       float64           `json:"value"`
}

type Metrics []Metric

type NotificationRequest struct {
	ID   string           `json:"id"`
	Type NotificationType `json:"type"`
}

type NotificationResponse struct {
	Error  string `json:"error"`
	Status string `json:"status"`
}

type NotificationType string

type PackagePhase string

type PackageStatus struct {
	Name           string       `json:"name"`
	Phase          PackagePhase `json:"phase"`
	Status         string       `json:"status"`
	CurrentVersion string       `json:"current_version"`
}

type RepairActionType string

type RunModeType string

type SuggestedActions struct {
	// Description describes the issue in detail.
	Description string `json:"description"`

	// A list of repair actions to mitigate the issue.
	RepairActions []RepairActionType `json:"repair_actions"`
}

type ValidationResult struct {
	Name       string `json:"name"`
	Valid      bool   `json:"valid"`
	Reason     string `json:"reason"`
	Suggestion string `json:"suggestion"`
}