API Reference¶

Packages¶

mokka.nvidia.com/v1alpha1

mokka.nvidia.com/v1alpha1¶

Package v1alpha1 contains API types for the mokka.nvidia.com group.

Resource Types¶

SGPUInventory
SGPURack
SGPURackProfile
SGPURuntimePolicy

CapabilityAttribute¶

CapabilityAttribute is a typed variant; set exactly one field.

Appears in: - GPUCapabilities

Field	Description	Validation
`bool` boolean	Bool stores a boolean capability value.	Optional: {}
`int` integer	Int stores an integer capability value.	Optional: {}
`string` string	String stores a string capability value.	Optional: {}
`strings` string array	Strings stores a set of string capability values.	Optional: {}

ClockRates¶

ClockRates is a graphics/SM/memory/video clock tuple in MHz.

Appears in: - GPUClocks

Field	Description	Validation
`graphics` integer	Graphics is the graphics clock rate in MHz.	Optional: {}
`sm` integer	SM is the streaming-multiprocessor clock rate in MHz.	Optional: {}
`memory` integer	Memory is the memory clock rate in MHz.	Optional: {}
`video` integer	Video is the video clock rate in MHz.	Optional: {}

ClocksTelemetry¶

ClocksTelemetry reports current clock rates.

Appears in: - RuntimeTelemetry

Field	Description	Validation
`graphicsMHz` integer	GraphicsMHz is the simulated graphics clock rate in MHz.	Minimum: 0 Optional: {}
`smMHz` integer	SMMHz is the simulated streaming-multiprocessor clock rate in MHz.	Minimum: 0 Optional: {}
`memoryMHz` integer	MemoryMHz is the simulated memory clock rate in MHz.	Minimum: 0 Optional: {}
`videoMHz` integer	VideoMHz is the simulated video clock rate in MHz.	Minimum: 0 Optional: {}

ComputeCapability¶

ComputeCapability is the CUDA compute capability.

Appears in: - GPUModel

Field	Description	Default	Validation
`major` integer	Major is the major CUDA compute capability version.
`minor` integer	Minor is the minor CUDA compute capability version.

DeviceState¶

Underlying type: string

DeviceState is a simulated GPU health state.

Validation: - Enum: [Healthy Degraded Failed]

Appears in: - RuntimeState

Field	Description
`Healthy`
`Degraded`
`Failed`

FabricDomain¶

FabricDomain is the fabric compute domain.

Appears in: - GPUFabric

Field	Description	Default	Validation
`scope` string	Scope is the level at which GPUs share a fabric domain.		Enum: [Node Rack Cluster] Optional: {}
`gpuCount` integer	GPUCount is the number of GPUs in the domain.		Optional: {}

FabricSwitches¶

FabricSwitches is fabric switches visible per node.

Appears in: - GPUFabric

Field	Description	Default	Validation
`visiblePerNode` integer	VisiblePerNode is the number of fabric switches visible from each Node.		Optional: {}

GPUBoard¶

GPUBoard is board-level identity.

Appears in: - GPUModel

Field	Description	Default	Validation
`partNumber` string	PartNumber is the board part number reported for the GPU.		Optional: {}

GPUCapabilities¶

GPUCapabilities is GPU feature toggles and extensible attributes.

Appears in: - SGPUGPUs

Field	Description	Default	Validation
`mig` MIGCapability	MIG describes Multi-Instance GPU support.		Optional: {}
`attributes` object (keys:string, values:CapabilityAttribute)	Extensible capability flags keyed by qualified name.		Optional: {}

GPUClocks¶

GPUClocks is clock ceilings and per-memory-clock schedules.

Appears in: - SGPUGPUs

Field	Description	Default	Validation
`maximumMHz` ClockRates	MaximumMHz gives the maximum graphics, SM, memory, and video clock rates.		Optional: {}
`supported` SupportedClocks array	Supported lists graphics clock rates available at each memory clock rate.		Optional: {}

GPUCores¶

GPUCores is programmable core counts.

Appears in: - GPUModel

Field	Description	Default	Validation
`cuda` integer	CUDA is the number of CUDA cores per GPU.		Optional: {}

GPUFabric¶

GPUFabric is NVLink/NVSwitch fabric.

Appears in: - SGPUTopology

Field	Description	Validation
`type` string	Type names the GPU interconnect fabric.	Optional: {}
`generation` integer	Generation is the fabric generation.	Optional: {}
`linksPerGPU` integer	LinksPerGPU is the number of fabric links available per GPU.	Optional: {}
`bandwidthPerLinkMBps` integer	BandwidthPerLinkMBps is the bandwidth of each link in megabytes per second.	Optional: {}
`c2cSupported` boolean	C2CSupported indicates whether coherent CPU-to-GPU links are supported.	Optional: {}
`domain` FabricDomain	Domain describes the scope and size of the GPU fabric domain.	Optional: {}
`switches` FabricSwitches	Switches describes the fabric switches visible to each Node.	Optional: {}

GPUFirmware¶

GPUFirmware is firmware versions.

Appears in: - GPUModel

Field	Description	Validation
`vbiosVersion` string	VBIOSVersion is the video BIOS version.	Optional: {}
`gspVersion` string	GSPVersion is the GPU System Processor firmware version.	Optional: {}
`infoROM` InfoROM	InfoROM contains versions of the GPU information-ROM objects.	Optional: {}

GPUMemory¶

GPUMemory is on-device memory.

Appears in: - SGPUGPUs

Field	Description	Validation
`capacity` Quantity	Capacity is the total device memory capacity.
`reserved` Quantity	Reserved is the amount of memory reserved from workloads.	Optional: {}
`bar1Capacity` Quantity	Bar1Capacity is the device BAR1 aperture size.	Optional: {}
`busWidthBits` integer	BusWidthBits is the memory bus width in bits.	Optional: {}

GPUModel¶

GPUModel is the vendor/product identity.

Appears in: - SGPUGPUs

Field	Description	Validation
`vendor` string	Vendor is the GPU manufacturer name.	Optional: {}
`product` string	Product is the vendor's product identifier.	Optional: {}
`productName` string	ProductName is the human-readable product name.	Optional: {}
`architecture` string	Architecture is the GPU architecture name.	Optional: {}
`computeCapability` ComputeCapability	ComputeCapability is the CUDA compute capability version.	Optional: {}
`cores` GPUCores	Cores contains programmable core counts.	Optional: {}
`board` GPUBoard	Board contains board-level identity.	Optional: {}
`firmware` GPUFirmware	Firmware contains firmware and information-ROM versions.	Optional: {}

GPUPCI¶

GPUPCI is PCI identity and link characteristics.

Appears in: - SGPUGPUs

Field	Description	Validation
`vendorID` string	VendorID is the PCI vendor identifier in hexadecimal.	Optional: {}
`deviceID` string	DeviceID is the PCI device identifier in hexadecimal.	Optional: {}
`subsystemVendorID` string	SubsystemVendorID is the PCI subsystem vendor identifier.	Optional: {}
`subsystemDeviceID` string	SubsystemDeviceID is the PCI subsystem device identifier.	Optional: {}
`maxLink` PCILink	MaxLink is the maximum supported PCIe generation and width.	Optional: {}

GPUPower¶

GPUPower is the power envelope.

Appears in: - SGPUGPUs

Field	Description	Default	Validation
`managementSupported` boolean	ManagementSupported indicates whether power management is available.		Optional: {}
`limitsMilliWatts` PowerLimits	LimitsMilliWatts gives the minimum, default, and maximum power limits.		Optional: {}

GPUSlot¶

GPUSlot pins one GPU to a PCI address + NUMA/root-complex/host CPU.

Appears in: - SGPUTopology

Field	Description	Validation
`index` integer	Index is the zero-based GPU index on the Node.	Maximum: 63 Minimum: 0
`pciAddress` string	PCIAddress is the PCI bus address of this GPU.	Pattern: `^[0-9a-f]\{4\}:[0-9a-f]\{2\}:[0-9a-f]\{2\}\.[0-7]$` Required: {}
`rootComplex` string	RootComplex identifies the PCI root complex connected to this GPU.	Pattern: `^pci[0-9a-f]\{4\}:[0-9a-f]\{2\}$` Required: {}
`numaNode` integer	NumaNode is the host NUMA node associated with this GPU.	Optional: {}
`hostProcessorIndex` integer	HostProcessorIndex is the host CPU index associated with this GPU.	Optional: {}

GPUThermal¶

GPUThermal is thermal thresholds.

Appears in: - SGPUGPUs

Field	Description	Validation
`targetCelsius` integer	TargetCelsius is the target GPU temperature in degrees Celsius.	Optional: {}
`maxOperatingCelsius` integer	MaxOperatingCelsius is the maximum rated operating temperature.	Optional: {}
`slowdownThresholdCelsius` integer	SlowdownThresholdCelsius is the temperature at which thermal slowdown begins.	Optional: {}
`shutdownThresholdCelsius` integer	ShutdownThresholdCelsius is the temperature at which shutdown is triggered.	Optional: {}

HostCPU¶

HostCPU is host CPU identity.

Appears in: - SGPUHost

Field	Description	Validation
`vendor` string	Vendor is the processor manufacturer.	Optional: {}
`product` string	Product is the processor product identifier.	Optional: {}
`architecture` string	Architecture is the host processor architecture.	Optional: {}
`cores` integer	Cores is the number of host CPU cores.	Optional: {}

HostMemory¶

HostMemory is host memory capacity.

Appears in: - SGPUHost

Field	Description	Default	Validation
`capacity` Quantity	Capacity is the host memory capacity.
`coherentWithGPU` boolean	CoherentWithGPU indicates whether host and GPU memory are coherent.		Optional: {}

InfoROM¶

InfoROM is NVML inforom sub-object versions.

Appears in: - GPUFirmware

Field	Description	Validation
`imageVersion` string	ImageVersion is the information-ROM image version.	Optional: {}
`oemObjectVersion` string	OEMObjectVersion is the OEM information-ROM object version.	Optional: {}
`eccObjectVersion` string	ECCObjectVersion is the ECC information-ROM object version.	Optional: {}
`powerObjectVersion` string	PowerObjectVersion is the power information-ROM object version.	Optional: {}

InventoryCapacity¶

InventoryCapacity is realized rack/node/GPU counts.

Appears in: - RackGroupStatus - SGPUInventoryStatus

Field	Description	Default	Validation
`racks` integer	Racks is the number of materialized racks.
`nodes` integer	Nodes is the number of logical Nodes across those racks.
`gpus` integer	GPUs is the number of simulated GPUs across those Nodes.

InventoryUsage¶

InventoryUsage summarizes eligible placement Nodes, bound rack slots, and pending allocations.

Appears in: - RackGroupStatus - SGPUInventoryStatus

Field	Description	Default	Validation
`requestedNodes` integer	RequestedNodes counts eligible Nodes matching placement selectors. Inventory totals deduplicate matches across groups; per-group values count each group's matches.
`allocatedNodes` integer	AllocatedNodes counts rack Node slots with a live matching Kubernetes Node binding.
`availableNodes` integer	AvailableNodes is remaining capacity after bound rack slots, never below zero.
`pendingNodes` integer	PendingNodes counts eligible placement requests awaiting capacity in one matching group.

MIGCapability¶

MIGCapability is MIG partitioning support.

Appears in: - GPUCapabilities

Field	Description	Default	Validation
`supported` boolean	Supported indicates whether MIG is available on the GPU.		Optional: {}
`maxGPUInstances` integer	MaxGPUInstances is the maximum number of GPU instances supported.		Optional: {}

NetworkTopology¶

NetworkTopology is out-of-band adapters.

Appears in: - SGPUTopology

Field	Description	Validation
`type` string	Type names the out-of-band network adapter type.	Optional: {}
`adapterModel` string	AdapterModel is the adapter model name.	Optional: {}
`firmwareVersion` string	FirmwareVersion is the adapter firmware version.	Optional: {}
`linkSpeedGbps` integer	LinkSpeedGbps is the adapter link speed in gigabits per second.	Optional: {}
`adaptersPerGPU` integer	AdaptersPerGPU is the number of network adapters associated with each GPU.	Optional: {}

PCILink¶

PCILink is the PCIe max link generation and width.

Appears in: - GPUPCI

Field	Description	Default	Validation
`generation` integer	Generation is the maximum PCIe generation.		Optional: {}
`width` integer	Width is the maximum PCIe link width in lanes.		Optional: {}

PercentRange¶

PercentRange is a min/max percentage bound. Both bounds are required, so an override replaces an inherited range whole.

Appears in: - UtilizationPattern

Field	Description	Default	Validation
`minimum` integer	Minimum is the lower utilization percentage bound.		Maximum: 100 Minimum: 0
`maximum` integer	Maximum is the upper utilization percentage bound.		Maximum: 100 Minimum: 0

PolicyTargetRef¶

PolicyTargetRef selects the simulated GPUs a policy applies to. Each listed axis narrows the one above it, and an omitted axis selects every index. The deepest listed axis is the policy's scope: inventory, rack group, rack, Node, or GPU. A narrower scope overrides a broader one field by field. Every listed rack group must exist in the inventory, and every listed index must exist in at least one selected rack group.

Appears in: - SGPURuntimePolicySpec

Field	Description	Validation
`group` string	Group is the API group containing the target resource.	Enum: [mokka.nvidia.com]
`kind` string	Kind is the target resource kind.	Enum: [SGPUInventory]
`name` string	Name is the name of the target SGPUInventory.	MinLength: 1
`rackGroups` string array	RackGroups optionally limits the policy to these rack-group IDs.	MaxItems: 64 MinItems: 1 Optional: {}
`rackIndexes` integer array	RackIndexes optionally limits the policy to these rack indexes.	MaxItems: 64 MinItems: 1 items:Minimum: 0 Optional: {}
`nodeIndexes` integer array	NodeIndexes optionally limits the policy to these logical Node indexes.	MaxItems: 64 MinItems: 1 items:Minimum: 0 Optional: {}
`gpuIndexes` integer array	GPUIndexes optionally limits the policy to these GPU indexes.	MaxItems: 64 MinItems: 1 items:Minimum: 0 Optional: {}

PowerLimits¶

PowerLimits is min/default/max power caps in milliwatts.

Appears in: - GPUPower

Field	Description	Validation
`minimum` integer	Minimum is the minimum power limit in milliwatts.	Optional: {}
`default` integer	Default is the default power limit in milliwatts.	Optional: {}
`maximum` integer	Maximum is the maximum power limit in milliwatts.	Optional: {}

PowerTelemetry¶

PowerTelemetry drives synthetic power draw.

Appears in: - RuntimeTelemetry

Field	Description	Default	Validation
`mode` string	Mode selects a constant or generated power draw.		Enum: [Fixed Pattern] Optional: {}
`drawMilliWatts` integer	DrawMilliWatts is the simulated GPU power draw in milliwatts.		Minimum: 0 Optional: {}

ProfileReference¶

ProfileReference targets an SGPURackProfile by name.

Appears in: - RackGroup

Field	Description	Default	Validation
`name` string	Name is the name of the SGPURackProfile to use.		MinLength: 1

RackGroup¶

RackGroup is a homogeneous group of racks sharing a profile.

Appears in: - SGPUInventorySpec

Field	Description	Validation
`id` string	ID uniquely identifies this group within its inventory.	MaxLength: 63 MinLength: 1
`count` integer	Count is the number of racks to materialize from this group.	Maximum: 100000 Minimum: 1
`profileRef` ProfileReference	ProfileRef names the rack profile used to materialize each rack.
`placement` RackPlacement	Placement optionally restricts the Nodes eligible for this rack group.	Optional: {}

RackGroupStatus¶

RackGroupStatus is the per-group Capacity + Usage projection.

Appears in: - SGPUInventoryStatus

Field	Description	Default	Validation
`id` string	ID identifies the rack group summarized by this status.
`profileName` string	ProfileName is the name of the profile used to materialize the group.
`capacity` InventoryCapacity	Capacity is the realized capacity for this group.
`usage` InventoryUsage	Usage summarizes Node allocation for this group.

RackPlacement¶

RackPlacement constrains which nodes a rack group may materialize on.

Appears in: - RackGroup

Field	Description	Default	Validation
`nodeSelector` LabelSelector	NodeSelector restricts placement to Nodes matching these labels.		Optional: {}

RuntimeModes¶

RuntimeModes is the persistent NVML/CUDA mode settings.

Appears in: - RuntimeState

Field	Description	Validation
`persistence` string	Persistence controls the simulated persistence mode.	Enum: [Enabled Disabled] Optional: {}
`compute` string	Compute selects the simulated compute mode.	Enum: [Default Exclusive Prohibited] Optional: {}
`mig` string	MIG controls whether Multi-Instance GPU mode is enabled.	Enum: [Enabled Disabled] Optional: {}
`ecc` string	ECC controls whether error-correcting code is enabled.	Enum: [Enabled Disabled] Optional: {}
`accounting` string	Accounting controls whether GPU accounting mode is enabled.	Enum: [Enabled Disabled] Optional: {}

RuntimeState¶

RuntimeState is the effective runtime settings of a simulated GPU. Sparse: omitted fields inherit; explicit zero is a set value.

Appears in: - SGPURackProfileDefaults - SGPURuntimePolicySpec

Field	Description	Validation
`deviceState` DeviceState	DeviceState is the simulated health state of the GPU.	Enum: [Healthy Degraded Failed] Optional: {}
`modes` RuntimeModes	Modes contains persistent NVML/CUDA mode settings.	Optional: {}
`telemetry` RuntimeTelemetry	Telemetry contains synthetic utilization, power, temperature, and clock values.	Optional: {}

RuntimeTelemetry¶

RuntimeTelemetry is the synthetic NVML telemetry.

Appears in: - RuntimeState

Field	Description	Validation
`performanceState` string	NVML P-state, e.g. "P0".	Pattern: `^P[0-9]+$` Optional: {}
`utilization` UtilizationTelemetry	Utilization defines synthetic GPU and memory utilization values.	Optional: {}
`power` PowerTelemetry	Power defines synthetic power draw values.	Optional: {}
`temperature` TemperatureTelemetry	Temperature defines synthetic GPU and memory temperature values.	Optional: {}
`clocks` ClocksTelemetry	Clocks defines the current simulated clock rates.	Optional: {}

SGPUGPUs¶

SGPUGPUs is the shared template for every GPU on the node.

Appears in: - SGPUNode

Field	Description	Validation
`count` integer	Count is the number of GPUs exposed on each Node.	Maximum: 64 Minimum: 1
`model` GPUModel	Model describes the GPU vendor, product, and architecture.
`memory` GPUMemory	Memory describes device memory capacity and bus width.
`pci` GPUPCI	PCI describes device identifiers and maximum link properties.
`power` GPUPower	Power describes power-management support and device limits.	Optional: {}
`thermal` GPUThermal	Thermal describes device temperature thresholds.	Optional: {}
`clocks` GPUClocks	Clocks describes maximum and supported device clock rates.	Optional: {}
`capabilities` GPUCapabilities	Capabilities describes MIG support and additional named capabilities.	Optional: {}

SGPUHost¶

SGPUHost is host CPU + memory.

Appears in: - SGPUNode

Field	Description	Default	Validation
`cpu` HostCPU	CPU describes the host processor presented by the simulated Node.		Optional: {}
`memory` HostMemory	Memory describes host memory presented by the simulated Node.		Optional: {}

SGPUInventory¶

SGPUInventory is a set of simulated GPU racks to distribute across CPU nodes.

Field	Description	Validation
`apiVersion` string	`mokka.nvidia.com/v1alpha1`
`kind` string	`SGPUInventory`
`metadata` ObjectMeta	Refer to Kubernetes API documentation for fields of `metadata`.
`spec` SGPUInventorySpec	Spec defines the simulated rack groups managed by this inventory.
`status` SGPUInventoryStatus	Status reports the capacity and usage currently realized by the controller.	Optional: {}

SGPUInventorySpec¶

SGPUInventorySpec is the desired rack composition.

Appears in: - SGPUInventory

Field	Description	Default	Validation
`rackGroups` RackGroup array	RackGroups declares homogeneous sets of racks. One group can expand to many racks; the controller admits at most 64 groups across all Inventories.		MaxItems: 64 MinItems: 1

SGPUInventoryStatus¶

SGPUInventoryStatus is the Control Plane view of realized capacity and usage.

Appears in: - SGPUInventory

Field	Description	Validation
`rackGroupsSummary` string	RackGroupsSummary is a comma-joined rendering of rack-group IDs.	Optional: {}
`capacity` InventoryCapacity	Capacity is the total rack, Node, and GPU capacity realized by this inventory.	Optional: {}
`usage` InventoryUsage	Usage summarizes requested and allocated Nodes across the inventory.	Optional: {}
`rackGroups` RackGroupStatus array	RackGroups contains per-group capacity and usage summaries.	Optional: {}
`conditions` Condition array	Conditions report the latest inventory acceptance and programming results.	Optional: {}

SGPUNode¶

SGPUNode is a homogeneous logical node in the rack.

Appears in: - SGPURackProfileSpec

Field	Description	Validation
`gpus` SGPUGPUs	GPUs defines the homogeneous simulated GPU set on each Node.
`host` SGPUHost	Host describes the CPU and memory presented by each Node.	Optional: {}
`topology` SGPUTopology	Topology describes PCIe slots and optional GPU/network fabrics.	Required: {}

SGPUNodeReference¶

SGPUNodeReference identifies an exact Kubernetes Node instance.

Appears in: - SGPURackNode

Field	Description	Default	Validation
`name` string	Name is the Kubernetes Node name.		MinLength: 1
`uid` UID	UID distinguishes this Node instance from a replacement with the same name.

SGPURack¶

SGPURack is the controller-owned materialization of one inventory rack. Its controller owner reference identifies the inventory pinned by spec.inventoryRef.

Field	Description	Validation
`apiVersion` string	`mokka.nvidia.com/v1alpha1`
`kind` string	`SGPURack`
`metadata` ObjectMeta	Refer to Kubernetes API documentation for fields of `metadata`.
`spec` SGPURackSpec	Spec is the rendered rack definition and its logical Node bindings.
`status` SGPURackStatus	Status summarizes the rack's durable Node assignments.	Optional: {}

SGPURackGPU¶

SGPURackGPU contains deterministic identity and structural placement.

Appears in: - SGPURackNode

Field	Description	Validation
`index` integer	Index is the zero-based GPU index within the logical Node.	Maximum: 63 Minimum: 0
`uuid` string	UUID is the stable GPU identifier.	Pattern: `^GPU-[0-9a-f]\{8\}-[0-9a-f]\{4\}-[0-9a-f]\{4\}-[0-9a-f]\{4\}-[0-9a-f]\{12\}$`
`serial` string	Serial is the GPU serial number.	MinLength: 1
`minorNumber` integer	MinorNumber is the device minor number exposed to workloads.	Minimum: 0
`pciAddress` string	PCIAddress is the GPU's PCI bus address.	Pattern: `^[0-9a-f]\{4\}:[0-9a-f]\{2\}:[0-9a-f]\{2\}\.[0-7]$`
`rootComplex` string	RootComplex identifies the PCI root complex connected to the GPU.	Pattern: `^pci[0-9a-f]\{4\}:[0-9a-f]\{2\}$`
`numaNode` integer	NUMANode is the host NUMA node associated with the GPU.	Minimum: 0
`hostProcessorIndex` integer	HostProcessorIndex is the host CPU index associated with the GPU.	Minimum: 0

SGPURackIdentity¶

SGPURackIdentity contains deterministic rack coordinates and fabric identity.

Appears in: - SGPURackSpec

Field	Description	Validation
`rackGroup` string	RackGroup is the rack-group identifier within the source inventory.	Format: dns1123Label MaxLength: 63 MinLength: 1
`rackIndex` integer	RackIndex is the zero-based rack index within its group.	Minimum: 0
`fabricUUID` string	FabricUUID identifies the fabric domain associated with the rack.	Format: uuid
`cliqueID` integer	CliqueID is reserved for future multi-clique layouts and remains zero.	Maximum: 0 Minimum: 0

SGPURackInventoryReference¶

SGPURackInventoryReference pins a rack to an exact inventory instance.

Appears in: - SGPURackSpec

Field	Description	Default	Validation
`name` string	Name is the inventory name.		MinLength: 1
`uid` UID	UID pins the reference to the exact inventory instance.

SGPURackNode¶

SGPURackNode is one logical Node and its optional Kubernetes Node binding.

Appears in: - SGPURackSpec

Field	Description	Validation
`index` integer	Index is the zero-based logical Node index within the rack.	Maximum: 1023 Minimum: 0
`nodeRef` SGPUNodeReference	NodeRef is absent while the logical Node is unbound. Its UID distinguishes a replacement Kubernetes Node that reuses the same name.	Optional: {}
`gpus` SGPURackGPU array	GPUs are rendered once for the logical Node and do not depend on its binding.	MaxItems: 64 MinItems: 1

SGPURackProfile¶

SGPURackProfile is the static shape of a simulated GPU rack.

Field	Description	Default	Validation
`apiVersion` string	`mokka.nvidia.com/v1alpha1`
`kind` string	`SGPURackProfile`
`metadata` ObjectMeta	Refer to Kubernetes API documentation for fields of `metadata`.
`spec` SGPURackProfileSpec	Spec defines the hardware shape and default software/runtime metadata.

SGPURackProfileDefaults¶

SGPURackProfileDefaults is initial runtime state for fresh sGPUs.

Appears in: - SGPURackProfileSpec

Field	Description	Default	Validation
`runtime` RuntimeState	Runtime provides initial sparse runtime settings for newly materialized GPUs.		Optional: {}

SGPURackProfileReference¶

SGPURackProfileReference pins rendered data to an exact profile revision.

Appears in: - SGPURackSpec

Field	Description	Validation
`name` string	Name is the profile name.	MinLength: 1
`uid` UID	UID pins the reference to the exact profile instance.
`generation` integer	Generation records the profile generation used for rendering.	Minimum: 1
`revision` string	Revision identifies the rendered profile content, independent of its name.	Pattern: `^[0-9a-f]\{64\}$`

SGPURackProfileSpec¶

SGPURackProfileSpec is one logical rack profile.

Appears in: - SGPURackProfile

Field	Description	Validation
`rack` SGPURackShape	Rack is the logical rack shape described by this profile.
`node` SGPUNode	Node describes the shared simulated hardware for each rack Node.
`software` SGPUSoftware	Software records optional driver, NVML, and CUDA versions.	Optional: {}
`defaults` SGPURackProfileDefaults	Defaults provides initial runtime settings for newly materialized GPUs.	Optional: {}

SGPURackShape¶

SGPURackShape defines the dimensions of a logical rack template. It is nested profile data rather than a materialized SGPURack resource.

Appears in: - SGPURackProfileSpec

Field	Description	Default	Validation
`nodesPerRack` integer	NodesPerRack is the number of logical Nodes in each rack.		Maximum: 1024 Minimum: 1

SGPURackSpec¶

SGPURackSpec is a rendered rack with durable logical Node bindings.

Appears in: - SGPURack

Field	Description	Validation
`inventoryRef` SGPURackInventoryReference	InventoryRef pins ownership to an exact inventory instance so a same-name replacement cannot inherit this rack.
`profileRef` SGPURackProfileReference	ProfileRef records the exact profile revision used to render the rack.
`identity` SGPURackIdentity	Identity is the stable coordinate used by agents and topology consumers.
`nodes` SGPURackNode array	Nodes retain stable device identities while Kubernetes Node assignments change. Fabric and network capabilities remain in the referenced profile.	MaxItems: 1024 MinItems: 1

SGPURackStatus¶

SGPURackStatus summarizes durable logical Node assignments.

Appears in: - SGPURack

Field	Description	Validation
`observedGeneration` integer	ObservedGeneration is the latest spec generation reflected by this status.	Minimum: 0 Optional: {}
`assignedNodes` integer	AssignedNodes is the number of logical Nodes with an exact Kubernetes Node binding.	Minimum: 0 Optional: {}
`conditions` Condition array	Conditions report whether the rendered assignment is ready for agents.	Optional: {}

SGPURuntimePolicy¶

SGPURuntimePolicy applies a sparse RuntimeState override to a subset of an SGPUInventory.

Field	Description	Validation
`apiVersion` string	`mokka.nvidia.com/v1alpha1`
`kind` string	`SGPURuntimePolicy`
`metadata` ObjectMeta	Refer to Kubernetes API documentation for fields of `metadata`.
`spec` SGPURuntimePolicySpec	Spec defines the target selection and runtime settings to apply.
`status` SGPURuntimePolicyStatus	Status summarizes the target scope selected for this policy.	Optional: {}

SGPURuntimePolicySpec¶

SGPURuntimePolicySpec pairs a target selector with the RuntimeState to apply.

Appears in: - SGPURuntimePolicy

Field	Description	Default	Validation
`targetRef` PolicyTargetRef	TargetRef selects the inventory and optional rack, Node, and GPU subset.
`runtime` RuntimeState	Runtime contains sparse settings to apply to the selected GPUs.		Optional: {}

SGPURuntimePolicyStatus¶

SGPURuntimePolicyStatus reports whether the policy is accepted and holds comma-joined denormalizations of spec.targetRef axes for print columns.

Appears in: - SGPURuntimePolicy

Field	Description	Validation
`rackGroupsSummary` string	RackGroupsSummary is the comma-separated selected rack-group IDs.	Optional: {}
`rackIndexesSummary` string	RackIndexesSummary is the comma-separated selected rack indexes.	Optional: {}
`nodeIndexesSummary` string	NodeIndexesSummary is the comma-separated selected logical Node indexes.	Optional: {}
`gpuIndexesSummary` string	GPUIndexesSummary is the comma-separated selected GPU indexes.	Optional: {}
`conditions` Condition array	Conditions report whether the policy is accepted.	Optional: {}

SGPUSoftware¶

SGPUSoftware is driver/NVML/CUDA versions.

Appears in: - SGPURackProfileSpec

Field	Description	Validation
`driverVersion` string	DriverVersion is the NVIDIA driver version.	Optional: {}
`nvmlVersion` string	NVMLVersion is the NVIDIA Management Library version.	Optional: {}
`cudaVersion` string	CUDAVersion is the CUDA version.	Optional: {}

SGPUTopology¶

SGPUTopology is PCIe slots, GPU fabric, and network visible to the node.

Appears in: - SGPUNode

Field	Description	Validation
`gpuSlots` GPUSlot array	GPUSlots maps each GPU index to its PCI and host topology coordinates.	MaxItems: 64 MinItems: 1
`gpuFabric` GPUFabric	GPUFabric describes optional NVLink/NVSwitch connectivity.	Optional: {}
`network` NetworkTopology	Network describes optional out-of-band network adapters.	Optional: {}

SupportedClocks¶

SupportedClocks pairs a memory clock with supported graphics clocks.

Appears in: - GPUClocks

Field	Description	Default	Validation
`memoryMHz` integer	MemoryMHz is a memory clock rate paired with GraphicsMHz values.
`graphicsMHz` integer array	GraphicsMHz lists graphics clock rates supported at MemoryMHz.

TemperatureTelemetry¶

TemperatureTelemetry drives synthetic GPU/memory temperature.

Appears in: - RuntimeTelemetry

Field	Description	Validation
`mode` string	Mode selects constant or generated temperature values.	Enum: [Fixed Pattern] Optional: {}
`gpuCelsius` integer	GPUCelsius is the simulated GPU temperature in degrees Celsius.	Optional: {}
`memoryCelsius` integer	MemoryCelsius is the simulated memory temperature in degrees Celsius.	Optional: {}

UtilizationPattern¶

UtilizationPattern shapes the generated utilization curve.

Appears in: - UtilizationTelemetry

Field	Description	Validation
`type` string	Type selects the shape of the generated utilization curve.	Enum: [Steady Bursty Wave] Optional: {}
`gpuPercent` PercentRange	GPUPercent sets the minimum and maximum GPU utilization percentages.	Optional: {}
`memoryPercent` PercentRange	MemoryPercent sets the minimum and maximum memory utilization percentages.	Optional: {}

UtilizationTelemetry¶

UtilizationTelemetry drives synthetic GPU/memory utilization.

Appears in: - RuntimeTelemetry

Field	Description	Default	Validation
`mode` string	Mode selects pattern-generated or fixed utilization values.		Enum: [Pattern Fixed] Optional: {}
`pattern` UtilizationPattern	Pattern defines the utilization values generated over time.		Optional: {}