Documentation
¶
Index ¶
- Constants
- Variables
- func ExtractTaints(nodeClaim *karpv1.NodeClaim) (generalTaints, startupTaints []corev1.Taint)
- func ExtractVersionFromVMSize(vmsize *skewer.VMSizeType) string
- func GetAKSGPUImageSHA(size string) string
- func GetAKSMachineNodeImageVersionFromImageID(imageID string) (string, error)
- func GetAKSMachineNodeImageVersionFromSIGImageID(imageID string) (string, error)
- func GetAlphanumericHash(input string, length int) (string, error)
- func GetGPUDriverType(size string) string
- func GetGPUDriverVersion(size string) string
- func GetGPUManufacturer(vmSize string) string
- func GetMaxPods(nodeClass *v1beta1.AKSNodeClass, networkPlugin, networkPluginMode string) int32
- func GetSubnetResourceID(subscriptionID, resourceGroupName, virtualNetworkName, subnetName string) string
- func HasChanged(existing, new any, options *hashstructure.HashOptions) bool
- func ImageReferenceToString(imageRef *armcompute.ImageReference) string
- func IsAKSManagedVNET(nodeResourceGroup string, subnetID string) (bool, error)
- func IsAMDEnabledSKU(vmSize string) bool
- func IsDriverInstallSupported(vmSize string) bool
- func IsGPUSKU(vmSize string) bool
- func IsGPUSKUSupportedOnOS(vmSize string, osName string) bool
- func IsMarinerEnabledGPUSKU(vmSize string) bool
- func IsNvidiaEnabledSKU(vmSize string) bool
- func IsVMDeleting(vm armcompute.VirtualMachine) bool
- func NewTerminatingResourceError(gr schema.GroupResource, name string) *errors.StatusError
- func NormalizeClusterResourceGroupNameForLabel(resourceGroupName string) string
- func PrettySlice[T any](s []T, maxItems int) string
- func ReadResponseBody(respErr *azcore.ResponseError) ([]byte, error)
- func StringMap(list v1.ResourceList) map[string]string
- func UseGridDrivers(size string) bool
- func VMResourceIDToProviderID(ctx context.Context, id string) string
- func WithDefaultFloat64(key string, def float64) float64
- type GPUSKUConfig
- type GPUSKUInfo
- type VnetSubnetResource
Constants ¶
const ( Nvidia470CudaDriverVersion = "470.82.01" // https://github.com/Azure/AgentBaker/blob/c0e684e5cecebcf61554cc7d2e2d2191972d35ed/parts/common/components.json#L797-L811 NvidiaCudaDriverVersion = "550.144.03" AKSGPUCudaVersionSuffix = "20250328201547" NvidiaGridDriverVersion = "550.144.06" AKSGPUGridVersionSuffix = "20250512225043" )
TODO: Get these from agentbaker
Variables ¶
var ConvergedGPUDriverSizes = map[string]bool{ "standard_nv6ads_a10_v5": true, "standard_nv12ads_a10_v5": true, "standard_nv18ads_a10_v5": true, "standard_nv36ads_a10_v5": true, "standard_nv72ads_a10_v5": true, "standard_nv36adms_a10_v5": true, "standard_nc8ads_a10_v4": true, "standard_nc16ads_a10_v4": true, "standard_nc32ads_a10_v4": true, }
ConvergedGPUDriverSizes : these sizes use a "converged" driver to support both cuda/grid workloads.
how do you figure this out? ask HPC or find out by trial and error. installing vanilla cuda drivers will fail to install with opaque errors. see https://github.com/Azure/azhpc-extensions/blob/daaefd78df6f27012caf30f3b54c3bd6dc437652/NvidiaGPU/resources.json
Functions ¶
func ExtractTaints ¶ added in v1.7.0
ExtractTaints returns the general taints and startup taints from a NodeClaim, ensuring that UnregisteredNoExecuteTaint is present in startup taints.
func ExtractVersionFromVMSize ¶ added in v1.6.1
func ExtractVersionFromVMSize(vmsize *skewer.VMSizeType) string
extractVersionFromVMSize extracts and normalizes the version from VMSizeType, dropping "v" prefix and backfilling "1"
func GetAKSGPUImageSHA ¶
func GetAKSMachineNodeImageVersionFromImageID ¶ added in v1.6.8
WARNING: not supporting CIG images yet.
func GetAKSMachineNodeImageVersionFromSIGImageID ¶ added in v1.6.8
Convert from "/subscriptions/10945678-1234-1234-1234-123456789012/resourceGroups/AKS-Ubuntu/providers/Microsoft.Compute/galleries/AKSUbuntu/images/2204gen2containerd/versions/2022.10.03" to "AKSUbuntu-2204gen2containerd-2022.10.03".
func GetAlphanumericHash ¶ added in v1.7.0
GetAlphanumericHash generates a base36 alphanumeric hash of the input string with the specified length. Be mindful of collision risks with short lengths. Also note that length > 13 provides no additional collision resistance because the underlying hashstructure library returns a 64-bit hash, which only fills ~13 base36 characters; extra characters are just leading zeros. At the time of writing, this is being used in AKS machine instance provider/GetAKSMachineNameFromNodeClaimName(). See that for context.
func GetGPUDriverType ¶ added in v0.5.5
GetGPUDriverType returns the type of GPU driver for given VM SKU ("grid" or "cuda")
func GetGPUDriverVersion ¶
NV series GPUs target graphics workloads vs NC which targets compute. they typically use GRID, not CUDA drivers, and will fail to install CUDA drivers. NVv1 seems to run with CUDA, NVv5 requires GRID. NVv3 is untested on AKS, NVv4 is AMD so n/a, and NVv2 no longer seems to exist (?).
func GetGPUManufacturer ¶ added in v1.12.0
GetGPUManufacturer returns the GPU manufacturer for a VM SKU ("nvidia", "amd", or "")
func GetMaxPods ¶ added in v0.7.5
func GetMaxPods(nodeClass *v1beta1.AKSNodeClass, networkPlugin, networkPluginMode string) int32
GetMaxPods resolves what we should set max pods to for a given nodeclass. If not specified, defaults based on network-plugin. 30 for "azure", 110 for "kubenet", or 250 for "none" and network plugin mode overlay.
func GetSubnetResourceID ¶ added in v0.4.0
func GetSubnetResourceID(subscriptionID, resourceGroupName, virtualNetworkName, subnetName string) string
GetSubnetResourceID constructs the subnet resource id
func HasChanged ¶ added in v1.5.0
func HasChanged(existing, new any, options *hashstructure.HashOptions) bool
HasChanged returns if the given value has changed, given the existing and new instance
This option is accessible in place of using a ChangeMonitor, when there's access to both the existing and new data.
func ImageReferenceToString ¶ added in v0.7.0
func ImageReferenceToString(imageRef *armcompute.ImageReference) string
func IsAKSManagedVNET ¶ added in v1.5.0
IsAKSManagedVNET determines if the vnet managed or not. Note: You can "trick" this function if you really try by (for example) createding a VNET that looks like an AKS managed VNET, with the same resource group as the MC RG, in a different subscription, or by creating your own VNET in the MC RG whose name matches the AKS pattern but the VNET is actually yours rather than ours.
func IsAMDEnabledSKU ¶ added in v1.12.0
IsAMDEnabledSKU determines if a VM SKU is an AMD GPU SKU
func IsDriverInstallSupported ¶ added in v1.12.0
IsDriverInstallSupported returns true if the system knows how to install GPU drivers for this VM SKU. Currently all NVIDIA SKUs have driver installation support, while AMD SKUs do not. This is the single abstraction point for this decision — when AMD driver support is added, only this function needs to change.
func IsGPUSKU ¶ added in v1.12.0
IsGPUSKU determines if a VM SKU is a known GPU SKU (any vendor: nvidia, amd, etc.)
func IsGPUSKUSupportedOnOS ¶ added in v1.12.0
IsGPUSKUSupportedOnOS checks if a GPU SKU supports a given OS identifier (e.g., "ubuntu", "azurelinux", "azurelinux3")
func IsMarinerEnabledGPUSKU ¶
IsNvidiaEnabledSKU determines if an VM SKU has nvidia driver support
func IsNvidiaEnabledSKU ¶
IsNvidiaEnabledSKU determines if an VM SKU has nvidia driver support
func IsVMDeleting ¶ added in v0.7.0
func IsVMDeleting(vm armcompute.VirtualMachine) bool
func NewTerminatingResourceError ¶ added in v1.6.2
func NewTerminatingResourceError(gr schema.GroupResource, name string) *errors.StatusError
NewTerminatingResourceError returns a NotFound error indicating that the resource is terminating. This is useful for resources where termination should be treated as not found.
func NormalizeClusterResourceGroupNameForLabel ¶ added in v1.7.0
func PrettySlice ¶ added in v0.7.0
PrettySlice truncates a slice after a certain number of max items to ensure that the Slice isn't too long
func ReadResponseBody ¶ added in v1.12.0
func ReadResponseBody(respErr *azcore.ResponseError) ([]byte, error)
ReadResponseBody reads the body from a ResponseError's RawResponse.
func StringMap ¶ added in v0.7.0
func StringMap(list v1.ResourceList) map[string]string
StringMap returns the string map representation of the resource list
func UseGridDrivers ¶ added in v0.5.5
func VMResourceIDToProviderID ¶ added in v1.6.6
func WithDefaultFloat64 ¶ added in v0.7.0
WithDefaultFloat64 returns the float64 value of the supplied environment variable or, if not present, the supplied default value. If the float64 conversion fails, returns the default
Types ¶
type GPUSKUConfig ¶ added in v1.10.0
type GPUSKUConfig map[string]GPUSKUInfo
type GPUSKUInfo ¶ added in v1.10.0
type VnetSubnetResource ¶ added in v1.6.2
type VnetSubnetResource struct {
SubscriptionID string
ResourceGroupName string
VNetName string
SubnetName string
}
this parsing function replaces three different functions in different packages that all had bugs. Please don't use a regex to parse these
func GetVnetSubnetIDComponents ¶ added in v0.4.0
func GetVnetSubnetIDComponents(vnetSubnetID string) (VnetSubnetResource, error)
GetVnetSubnetIDComponents parses an Azure subnet resource ID into its component parts. Input: A fully qualified Azure subnet resource ID in the format:
/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.Network/virtualNetworks/{virtualNetworkName}/subnets/{subnetName}
The input is case-insensitive and must contain exactly 11 slash-separated segments. Output: A vnetSubnetResource struct containing:
- SubscriptionID: The Azure subscription ID
- ResourceGroupName: The resource group name
- VNetName: The virtual network name
- SubnetName: The subnet name
Returns an error if the input format is invalid or doesn't match the expected structure.
func (VnetSubnetResource) IsSameVNET ¶ added in v1.6.2
func (v VnetSubnetResource) IsSameVNET(cmp VnetSubnetResource) bool