Adding GPU nodes to self-managed OpenShift on Azure🔗
When deploying self-managed OpenShift on Azure, the compute nodes are represented as one or more OpenShift MachineSets. If your cluster is deployed in a single zone, there will be 1 MachineSet which defines the number and type of virtual machiness created in the Azure resource group. For multi-zone clusters, there will be 3 MachineSets.
Find the compute node MachineSet🔗
Below is an example of a compute node (worker) MachineSet created by the OpenShift installer. The instance type defines the node that is spun up, with 32 vCPUs and 128 GB of memory.
apiVersion: machine.openshift.io/v1beta1
kind: MachineSet
metadata:
annotations:
capacity.cluster-autoscaler.kubernetes.io/labels: kubernetes.io/arch=amd64
machine.openshift.io/GPU: '0'
machine.openshift.io/memoryMb: '131072'
machine.openshift.io/vCPU: '32'
resourceVersion: '67642'
name: fk-openshift-shtmp-worker-germanywestcentral1
uid: 0104469b-71d9-413e-ad52-40a1a2168a84
creationTimestamp: '2024-12-06T14:24:56Z'
generation: 3
managedFields:
- apiVersion: machine.openshift.io/v1beta1
fieldsType: FieldsV1
fieldsV1:
'f:spec':
'f:replicas': {}
manager: Mozilla
operation: Update
subresource: scale
- apiVersion: machine.openshift.io/v1beta1
fieldsType: FieldsV1
fieldsV1:
'f:metadata':
'f:labels':
.: {}
'f:machine.openshift.io/cluster-api-cluster': {}
'f:machine.openshift.io/cluster-api-machine-role': {}
'f:machine.openshift.io/cluster-api-machine-type': {}
'f:spec':
.: {}
'f:selector': {}
'f:template':
.: {}
'f:metadata':
.: {}
'f:labels':
.: {}
'f:machine.openshift.io/cluster-api-cluster': {}
'f:machine.openshift.io/cluster-api-machine-role': {}
'f:machine.openshift.io/cluster-api-machine-type': {}
'f:machine.openshift.io/cluster-api-machineset': {}
'f:spec':
.: {}
'f:lifecycleHooks': {}
'f:metadata': {}
'f:providerSpec':
.: {}
'f:value':
'f:zone': {}
'f:metadata':
.: {}
'f:creationTimestamp': {}
'f:publicIP': {}
'f:image':
.: {}
'f:offer': {}
'f:publisher': {}
'f:resourceID': {}
'f:sku': {}
'f:version': {}
'f:acceleratedNetworking': {}
'f:kind': {}
'f:location': {}
'f:managedIdentity': {}
'f:vmSize': {}
.: {}
'f:subnet': {}
'f:apiVersion': {}
'f:securityProfile':
.: {}
'f:settings': {}
'f:osDisk':
.: {}
'f:diskSettings': {}
'f:diskSizeGB': {}
'f:managedDisk':
.: {}
'f:securityProfile':
.: {}
'f:diskEncryptionSet': {}
'f:storageAccountType': {}
'f:osType': {}
'f:networkResourceGroup': {}
'f:diagnostics': {}
'f:credentialsSecret':
.: {}
'f:name': {}
'f:namespace': {}
'f:publicLoadBalancer': {}
'f:userDataSecret':
.: {}
'f:name': {}
'f:vnet': {}
'f:resourceGroup': {}
manager: cluster-bootstrap
operation: Update
time: '2024-12-06T14:24:56Z'
- apiVersion: machine.openshift.io/v1beta1
fieldsType: FieldsV1
fieldsV1:
'f:status': {}
manager: cluster-bootstrap
operation: Update
subresource: status
time: '2024-12-06T14:24:57Z'
- apiVersion: machine.openshift.io/v1beta1
fieldsType: FieldsV1
fieldsV1:
'f:metadata':
'f:annotations':
.: {}
'f:capacity.cluster-autoscaler.kubernetes.io/labels': {}
'f:machine.openshift.io/GPU': {}
'f:machine.openshift.io/memoryMb': {}
'f:machine.openshift.io/vCPU': {}
manager: machine-controller-manager
operation: Update
time: '2024-12-06T14:37:14Z'
- apiVersion: machine.openshift.io/v1beta1
fieldsType: FieldsV1
fieldsV1:
'f:status':
'f:availableReplicas': {}
'f:fullyLabeledReplicas': {}
'f:observedGeneration': {}
'f:readyReplicas': {}
'f:replicas': {}
manager: machineset-controller
operation: Update
subresource: status
time: '2024-12-06T15:54:53Z'
namespace: openshift-machine-api
labels:
machine.openshift.io/cluster-api-cluster: fk-openshift-shtmp
machine.openshift.io/cluster-api-machine-role: worker
machine.openshift.io/cluster-api-machine-type: worker
spec:
replicas: 3
selector:
matchLabels:
machine.openshift.io/cluster-api-cluster: fk-openshift-shtmp
machine.openshift.io/cluster-api-machineset: fk-openshift-shtmp-worker-germanywestcentral1
template:
metadata:
labels:
machine.openshift.io/cluster-api-cluster: fk-openshift-shtmp
machine.openshift.io/cluster-api-machine-role: worker
machine.openshift.io/cluster-api-machine-type: worker
machine.openshift.io/cluster-api-machineset: fk-openshift-shtmp-worker-germanywestcentral1
spec:
lifecycleHooks: {}
metadata: {}
providerSpec:
value:
osDisk:
diskSettings: {}
diskSizeGB: 300
managedDisk:
securityProfile:
diskEncryptionSet: {}
storageAccountType: Premium_LRS
osType: Linux
networkResourceGroup: fk-rg
publicLoadBalancer: fk-openshift-shtmp
userDataSecret:
name: worker-user-data
vnet: fk-openshift
securityProfile:
settings: {}
credentialsSecret:
name: azure-cloud-credentials
namespace: openshift-machine-api
diagnostics: {}
zone: '1'
metadata:
creationTimestamp: null
publicIP: false
resourceGroup: fk-openshift-rg
kind: AzureMachineProviderSpec
location: germanywestcentral
vmSize: Standard_D32s_v3
image:
offer: ''
publisher: ''
resourceID: /resourceGroups/fk-openshift-rg/providers/Microsoft.Compute/galleries/gallery_fk_openshift_shtmp/images/fk-openshift-shtmp-gen2/versions/latest
sku: ''
version: ''
acceleratedNetworking: true
managedIdentity: fk-openshift-shtmp-identity
subnet: control-plane
apiVersion: machine.openshift.io/v1beta1
status:
availableReplicas: 3
fullyLabeledReplicas: 3
observedGeneration: 3
readyReplicas: 3
replicas: 3
Transform to GPU MachineSet🔗
To create a GPU MachineSet in the same region, copy the yaml into your favourite text editor and remove all the unnecessary properties. Below is an example of the resulting yaml, highlighting the items that have changed to define the GPU node(s). In this example there is only 1 GPU node of type Standard_NC96ads_A100_v4. Only the highlighted properties should be changed.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 | |
Create the GPU MachineSet🔗
Once ready, do the following to create the MachineSet:
- Go to the OpenShift console
- Click the "+" sign at the top of the page
- Paste the updated yaml into the the window
The MachineSet will create Machine CRs in the openshift-machines project. If the instance type is available in the selected Azure region and there is enough capacity, the Azure VM(s) will be created and after a 5-10 minutes, it/they will appear as nodes in the OpenShift cluster.