feat: use HAMi with nvshare as GPU plugin (#1033)
This commit is contained in:
dkeven
219
frameworks/GPU/config/gpu/hami/values.yaml
Normal file
219
frameworks/GPU/config/gpu/hami/values.yaml
Normal file
@@ -0,0 +1,219 @@
|
||||
# Default values for hami-vgpu.
|
||||
|
||||
nameOverride: ""
|
||||
fullnameOverride: ""
|
||||
imagePullSecrets: [ ]
|
||||
version: "v2.5.0"
|
||||
|
||||
#Nvidia GPU Parameters
|
||||
resourceName: "nvidia.com/gpu"
|
||||
resourceMem: "nvidia.com/gpumem"
|
||||
resourceMemPercentage: "nvidia.com/gpumem-percentage"
|
||||
resourceCores: "nvidia.com/gpucores"
|
||||
resourcePriority: "nvidia.com/priority"
|
||||
|
||||
#MLU Parameters
|
||||
mluResourceName: "cambricon.com/vmlu"
|
||||
mluResourceMem: "cambricon.com/mlu.smlu.vmemory"
|
||||
mluResourceCores: "cambricon.com/mlu.smlu.vcore"
|
||||
|
||||
#Hygon DCU Parameters
|
||||
dcuResourceName: "hygon.com/dcunum"
|
||||
dcuResourceMem: "hygon.com/dcumem"
|
||||
dcuResourceCores: "hygon.com/dcucores"
|
||||
|
||||
#Iluvatar GPU Parameters
|
||||
iluvatarResourceName: "iluvatar.ai/vgpu"
|
||||
iluvatarResourceMem: "iluvatar.ai/vcuda-memory"
|
||||
iluvatarResourceCore: "iluvatar.ai/vcuda-core"
|
||||
|
||||
#Metax SGPU Parameters
|
||||
metaxResourceName: "metax-tech.com/sgpu"
|
||||
metaxResourceCore: "metax-tech.com/vcore"
|
||||
metaxResourceMem: "metax-tech.com/vmemory"
|
||||
|
||||
schedulerName: "hami-scheduler"
|
||||
|
||||
podSecurityPolicy:
|
||||
enabled: false
|
||||
|
||||
global:
|
||||
gpuHookPath: /usr/local
|
||||
labels: {}
|
||||
annotations: {}
|
||||
managedNodeSelectorEnable: false
|
||||
managedNodeSelector:
|
||||
usage: "gpu"
|
||||
|
||||
|
||||
scheduler:
|
||||
# @param nodeName defines the node name and the nvidia-vgpu-scheduler-scheduler will schedule to the node.
|
||||
# if we install the nvidia-vgpu-scheduler-scheduler as default scheduler, we need to remove the k8s default
|
||||
# scheduler pod from the cluster first, we must specify node name to skip the schedule workflow.
|
||||
nodeName: ""
|
||||
#nodeLabelSelector:
|
||||
# "gpu": "on"
|
||||
overwriteEnv: "false"
|
||||
defaultSchedulerPolicy:
|
||||
nodeSchedulerPolicy: binpack
|
||||
gpuSchedulerPolicy: spread
|
||||
metricsBindAddress: ":9395"
|
||||
livenessProbe: false
|
||||
leaderElect: true
|
||||
# when leaderElect is true, replicas is available, otherwise replicas is 1.
|
||||
replicas: 1
|
||||
kubeScheduler:
|
||||
# @param enabled indicate whether to run kube-scheduler container in the scheduler pod, it's true by default.
|
||||
enabled: true
|
||||
image: registry.k8s.io/kube-scheduler
|
||||
imageTag: ""
|
||||
imagePullPolicy: IfNotPresent
|
||||
resources: {}
|
||||
# If you do want to specify resources, uncomment the following lines, adjust them as necessary.
|
||||
# and remove the curly braces after 'resources:'.
|
||||
# limits:
|
||||
# cpu: 1000m
|
||||
# memory: 1000Mi
|
||||
# requests:
|
||||
# cpu: 100m
|
||||
# memory: 100Mi
|
||||
extraNewArgs:
|
||||
- --config=/config/config.yaml
|
||||
- -v=4
|
||||
extraArgs:
|
||||
- --policy-config-file=/config/config.json
|
||||
- -v=4
|
||||
extender:
|
||||
image: "beclab/hami"
|
||||
imagePullPolicy: IfNotPresent
|
||||
resources: {}
|
||||
# If you do want to specify resources, uncomment the following lines, adjust them as necessary,
|
||||
# and remove the curly braces after 'resources:'.
|
||||
# limits:
|
||||
# cpu: 1000m
|
||||
# memory: 1000Mi
|
||||
# requests:
|
||||
# cpu: 100m
|
||||
# memory: 100Mi
|
||||
extraArgs:
|
||||
- --debug
|
||||
- -v=4
|
||||
podAnnotations: {}
|
||||
tolerations: []
|
||||
#serviceAccountName: "hami-vgpu-scheduler-sa"
|
||||
admissionWebhook:
|
||||
customURL:
|
||||
enabled: false
|
||||
# must be an endpoint using https.
|
||||
# should generate host certs here
|
||||
host: 127.0.0.1 # hostname or ip, can be your node'IP if you want to use https://<nodeIP>:<schedulerPort>/<path>
|
||||
port: 31998
|
||||
path: /webhook
|
||||
whitelistNamespaces:
|
||||
# Specify the namespaces that the webhook will not be applied to.
|
||||
# - default
|
||||
# - kube-system
|
||||
# - istio-system
|
||||
reinvocationPolicy: Never
|
||||
failurePolicy: Ignore
|
||||
patch:
|
||||
image: jettech/kube-webhook-certgen:v1.5.2
|
||||
imageNew: liangjw/kube-webhook-certgen:v1.1.1
|
||||
imagePullPolicy: IfNotPresent
|
||||
priorityClassName: ""
|
||||
podAnnotations: {}
|
||||
nodeSelector: {}
|
||||
tolerations: []
|
||||
runAsUser: 2000
|
||||
service:
|
||||
type: NodePort # Default type is NodePort, can be changed to ClusterIP
|
||||
httpPort: 443 # HTTP port
|
||||
schedulerPort: 31998 # NodePort for HTTP
|
||||
monitorPort: 31993 # Monitoring port
|
||||
labels: {}
|
||||
annotations: {}
|
||||
|
||||
devicePlugin:
|
||||
image: "beclab/hami"
|
||||
monitorimage: "beclab/hami"
|
||||
monitorctrPath: /usr/local/vgpu/containers
|
||||
imagePullPolicy: IfNotPresent
|
||||
deviceSplitCount: 100
|
||||
deviceMemoryScaling: 100
|
||||
deviceCoreScaling: 100
|
||||
runtimeClassName: ""
|
||||
migStrategy: "none"
|
||||
disablecorelimit: "false"
|
||||
passDeviceSpecsEnabled: false
|
||||
extraArgs:
|
||||
- -v=4
|
||||
|
||||
service:
|
||||
type: NodePort # Default type is NodePort, can be changed to ClusterIP
|
||||
httpPort: 31992
|
||||
labels: {}
|
||||
annotations: {}
|
||||
|
||||
pluginPath: /var/lib/kubelet/device-plugins
|
||||
libPath: /usr/local/vgpu
|
||||
|
||||
podAnnotations: {}
|
||||
nvidianodeSelector:
|
||||
gpu.bytetrade.io/cuda-supported: 'true'
|
||||
tolerations: []
|
||||
# The updateStrategy for DevicePlugin DaemonSet.
|
||||
# If you want to update the DaemonSet by manual, set type as "OnDelete".
|
||||
# We recommend use OnDelete update strategy because DevicePlugin pod restart will cause business pod restart, this behavior is destructive.
|
||||
# Otherwise, you can use RollingUpdate update strategy to rolling update DevicePlugin pod.
|
||||
updateStrategy:
|
||||
type: RollingUpdate
|
||||
rollingUpdate:
|
||||
maxUnavailable: 1
|
||||
|
||||
resources: {}
|
||||
# If you do want to specify resources, uncomment the following lines, adjust them as necessary.
|
||||
# and remove the curly braces after 'resources:'.
|
||||
# limits:
|
||||
# cpu: 1000m
|
||||
# memory: 1000Mi
|
||||
# requests:
|
||||
# cpu: 100m
|
||||
# memory: 100Mi
|
||||
|
||||
vgpuMonitor:
|
||||
resources: {}
|
||||
# If you do want to specify resources, uncomment the following lines, adjust them as necessary.
|
||||
# and remove the curly braces after 'resources:'.
|
||||
# limits:
|
||||
# cpu: 1000m
|
||||
# memory: 1000Mi
|
||||
# requests:
|
||||
# cpu: 100m
|
||||
# memory: 100Mi
|
||||
|
||||
devices:
|
||||
mthreads:
|
||||
enabled: false
|
||||
customresources:
|
||||
- mthreads.com/vgpu
|
||||
nvidia:
|
||||
gpuCorePolicy: default
|
||||
ascend:
|
||||
enabled: false
|
||||
image: ""
|
||||
imagePullPolicy: IfNotPresent
|
||||
extraArgs: []
|
||||
nodeSelector:
|
||||
ascend: "on"
|
||||
tolerations: []
|
||||
customresources:
|
||||
- huawei.com/Ascend910A
|
||||
- huawei.com/Ascend910A-memory
|
||||
- huawei.com/Ascend910B2
|
||||
- huawei.com/Ascend910B2-memory
|
||||
- huawei.com/Ascend910B
|
||||
- huawei.com/Ascend910B-memory
|
||||
- huawei.com/Ascend910B4
|
||||
- huawei.com/Ascend910B4-memory
|
||||
- huawei.com/Ascend310P
|
||||
- huawei.com/Ascend310P-memory
|
||||
Reference in New Issue
Block a user