Define resource fields for api.Machine

[so-sahu]

Summary

This issue captures the efforts to extend the API machine object and the corresponding reconciler changes. The proposals and changes can be discussed in this issue.

[so-sahu]

type ResourceName string

const (
	ResourceCPU      ResourceName = "cpu"
	ResourceMemory   ResourceName = "memory"
	ResourceSGX      ResourceName = "sgx"
	ResourceHugepages ResourceName = "hugepages"
)

type MachineSpec struct {
        ...
	Resources ResourceList
	NumaNodes []int
        CPUPins map[int]int
        ...
}

type ResourceList map[ResourceName]resource.Quantity

func (rl *ResourceList) Get(name ResourceName) resource.Quantity {
	if val, ok := (*rl)[name]; ok {
		return val
	}
	return resource.Quantity{}
}

func (rl *ResourceList) Set(name ResourceName, value resource.Quantity) {
	(*rl)[name] = value
}

func (rl *ResourceList) CPU() resource.Quantity {
	return rl.Get(ResourceCPU)
}

func (rl *ResourceList) Memory() resource.Quantity {
	return rl.Get(ResourceMemory)
}

func (rl *ResourceList) SGX() resource.Quantity {
	return rl.Get(ResourceSGX)
}

func (rl *ResourceList) EnableHugepages() bool{
	if rl.Get(ResourceHugepages).Value() == 1 {
		return true
	}
	return false
}

Changes made

1. ResourceName and ResourceList Types

   • ResourceName defines string constants for different types of resources (cpu, memory, sgx, hugepages).
   • ResourceList is a map that associates ResourceName with resource.Quantity to manage resource quantities.

2. MachineSpec Struct

   • The MachineSpec struct now includes a Resources field of type ResourceList to store resource allocations.

3. ResourceList Methods

   • Get(name ResourceName) resource.Quantity: Retrieves the quantity of the specified resource.
   • Set(name ResourceName, value resource.Quantity): Sets the quantity of the specified resource.
   • CPU(), Memory(), SGX(): Convenience methods to retrieve specific resource quantities.
   • EnableHugepages(): Checks if hugepages are enabled.

Impact

• Improved readability and maintainability of resource configurations.
• Simplified handling of different resource types.
• Easier access and manipulation of resource quantities.

[lukas016]

Why do you need VCUAllocationRatio, BlockedHugepages and BlockedCPUs in machine spec? It is important for resource manager only for my point of view.

[lukas016]

Do you need NumaDomain? Isn't it len(NumaNodes)?

[lukas016]

network cards and storage can be resources too which ideally will manage by resource manager too. Could you think maybe how we can implement it too?

[lukas016]

what do you think about use resource.Quantity type? https://github.com/ironcore-dev/ironcore/blob/main/api/core/v1alpha1/resource.go

[afritzler]

How about the following. In the spec I have the possibility to assign a NUMA topoligy like this. If it is empty it is ignored (in case NUMA pinning is disabled) or a Scheduler/Resource Manager ensures that it is correctly filled later.

type MachineSpec struct {
	// Existing fields...

	NUMAPreferences NUMAPreferences `json:"numaPreferences,omitempty"`
}

type NUMAPreferences struct {
	CPUNodes    []int `json:"cpuNodes,omitempty"`
	MemoryNodes []int `json:"memoryNodes,omitempty"`
	IoNodes     []int `json:"ioNodes,omitempty"`
}

And in the Status you can update the information once the placement has been fulfilled.

type MachineStatus struct {
	// Existing fields...

	NUMAPlacement NUMAPlacement `json:"numaPlacement,omitempty"`
}

type NUMAPlacement struct {
	CPUNodes    []int `json:"cpuNodes"`
	MemoryNodes []int `json:"memoryNodes"`
	IoNodes     []int `json:"ioNodes"`
}