Skip to main content

HAMi Enterprise 2.10.0-r5

Chart versionDeclared appVersionRelease date
2.10.0-r52.10.0-r52026-11-06

This is a patch release. Changes below are compared with HAMi Enterprise v2.10.0-r4.

Software changes​

  • Extends license support across device vendors, adding licensing by device model and license-status annotations. Before upgrading, confirm through the delivery channel that the license covers the device models and quantities in use.
  • Fixes the compute-limit environment variable for NVIDIA workloads using the Downward API path, which is enabled by default (scheduler.useDownward: true). The scheduler now passes the compute quota from container annotations to the runtime through CUDA_DEVICE_SM_LIMIT.
  • Fixes PPU topology scheduling. Combined scheduling policies that include the topology policy now perform topology selection, and candidates with zero or negative topology scores are no longer omitted for that reason.
  • Updates Ascend mode selection. Pods without huawei.com/vnpu-mode use template slicing or hami-core soft slicing according to the candidate node's capabilities. A positive -core resource request automatically selects hami-core. Pods that explicitly select template retain template-slicing resource restrictions.
  • Rechecks namespace quotas against the actual Ascend allocation after template rounding. hami-core scheduling preserves the compute-capacity multiplier reported by the device, while each workload can still request at most 100% compute per device.

Chart changes​

  • Adds legacySelectorLabels, defaulting to false. Scheduler and NVIDIA device-plugin selectors now use the Chart name (or nameOverride) and Helm release name by default. Existing deployments must set it to true when upgrading to preserve historical selectors; see below.
  • Renames devices.ascend.vnpuDeviceSplitCount to devices.ascend.deviceSplitCount. It defaults to 10 and must be a positive integer. It controls virtual-device slots per physical NPU in soft-slicing mode. A positive node-level vDeviceCount takes precedence; template slicing still derives capacity from its templates.
  • Adds Ascend mode, compute-policy, and capacity-multiplier settings, listed below. The Ascend device plugin and runtime are delivered separately and are not installed by the main Chart.
  • Updates the main components' default image tag to v2.10.0-r5. The companion CRD Chart 0.4.2 and Resource Pool Manager Chart 0.0.6 retain their version numbers. The latter still defaults to image v0.0.2; confirm its capabilities against the separately delivered component version.
Main Chart settingDefaultMeaning and conditions
devices.ascend.hamiVnpuMode""Accepts template, hami-core (alias hamiCore), or enpu. An empty value retains hamiVnpuCore, which defaults to false. The plugin registers the actual node and device capabilities.
devices.ascend.hamiCorePolicyspatialhami-core runtime compute policy: spatial, dynamic, fixed, or overdraft. The Pod annotation huawei.com/npu-sched-policy takes precedence over node-level hamiCorePolicy, which takes precedence over the global setting.
devices.ascend.deviceCoreScaling1Multiplier for the compute percentage capacity advertised by hami-core; must be at least 1. For example, 2 advertises 200% total capacity without changing the per-workload limit of 100% per device.
devices.ascend.enpuPolicyelasticDefault policy for the optional ENPU backend: fixed-share, elastic, or best-effort. The Pod annotation huawei.com/enpu-policy can override it. Requires a separately deployed compatible plugin, runtime, and environment.

Upgrade notes​

Preserve selectors for existing deployments​

When upgrading an existing 2.10.0-r4 deployment, add the following setting to the existing configuration and retain it on subsequent upgrades:

legacySelectorLabels: true

The previous scheduler Deployment and NVIDIA device-plugin DaemonSet use fixed hami/hami selector labels. Applying the new default directly changes immutable selectors and causes the upgrade to fail. New installations use the default false. Custom monitoring or operational rules that select resources by these labels must match the labels of the actual deployment.

Migrate Ascend configuration​

Rename devices.ascend.vnpuDeviceSplitCount to devices.ascend.deviceSplitCount, retaining its value. Keeping only the old Helm key does not pass a custom value into the new default template. Deployments without an override continue to use 10.

devices:
ascend:
deviceSplitCount: 10

device-config.content replaces the entire default device configuration; it does not merge individual fields. Set deviceSplitCount under the vnpus mapping in that custom content and retain the configs model list and other required settings. Changing only devices.ascend.* does not update this custom content.

The scheduler and Ascend plugin must use the same device configuration and compatible plugin and runtime versions. The independently released Ascend plugin v1.5.0 (Chart 0.3.2) provides these configuration interfaces. To reuse the scheduler ConfigMap, set config.create: false in the independent plugin Chart and point config.deviceConfigMapName to the actual scheduler device ConfigMap in the same namespace. After upgrading, check that the registered device modes, virtual-device slot counts, and compute capacities match the configuration.

The companion plugin's default compute control now uses hamiCorePolicy, defaulting to spatial. spatial uses cooperative core budgets and does not guarantee exclusive hardware partitions or proportional throughput. dynamic schedules relative shares and allows borrowing idle compute; fixed controls the target percentage through run/sleep time; overdraft allows bounded borrowing while idle. To retain the previous fixed scheduling behavior, explicitly retain hamiVnpuCore.coreUtilizationPolicy: force in the independent plugin Chart. Existing workload-container settings for NPU_CORE_UTILIZATION_POLICY take precedence over the plugin default.

ENPU requires the explicit Pod annotation huawei.com/vnpu-mode: enpu. Each container is allocated one physical DIE, and ENPU cannot share that DIE with hami-core. ENPU instances on the same DIE must use the same policy: fixed-share allocates time slices by share, elastic can borrow idle compute, and best-effort does not enforce a compute-share limit. This backend requires compatible runtime assets, drivers, CANN, and a RuntimeClass. Setting main Chart parameters alone does not deploy it. Use the version-specific configuration supplied through the delivery channel for ENPU installation and upgrades; a main Chart upgrade does not enable ENPU or memory oversubscription by itself.

Update NVIDIA workloads​

CUDA_DEVICE_CORE_LIMIT was introduced with NVIDIA Downward API allocation. The first published Enterprise release containing it was v2.7.1-rc0 (January 16, 2026), and the name remained in use through v2.10.0-r4. The previous scheduler passed the Pod's requested compute percentage from container annotations into this environment variable, but the runtime reads CUDA_DEVICE_SM_LIMIT and could not obtain the quota from the old name.

On the default Downward API allocation path, if no other valid CUDA_DEVICE_SM_LIMIT or per-device setting is present, the runtime initializes the compute limit to 100%. For example, when a Pod requests 20% compute, the scheduler still allocates resources according to that request, but the runtime does not receive the 20% limit through the incorrectly named variable. This fix changes the injected name to CUDA_DEVICE_SM_LIMIT; the quota value and scheduler resource accounting remain unchanged. The non-Downward API allocation path already uses the correct variable and is unaffected by this naming error.

scheduler.useDownward defaults to true. For NVIDIA workloads using this default, recreate affected Pods through the normal workload rollout process after upgrading the scheduler so admission injects CUDA_DEVICE_SM_LIMIT again. Upgrading the scheduler does not replace environment variables in existing Pods. Update custom manifests that reference CUDA_DEVICE_CORE_LIMIT to use CUDA_DEVICE_SM_LIMIT.

Obtain packages and component manifests through the agreed delivery channel. See the Chart and appVersion mapping or the software center.

Values reference​

The following parameter descriptions are taken from the README bundled with the Chart 2.10.0-r5 release package and are preserved as published. Use this version's values.yaml as the source of default configuration.

Expand values table
KeyTypeDefaultDescription
birenResourceNamestring"birentech.com/gpu"Biren GPU resource name
device-config.contentstring""
devicePlugin.cdiVendorstring""CDI vendor for generated specs and device annotations (default: k8s.device-plugin.nvidia.com)
devicePlugin.createRuntimeClassboolfalseWhether to create runtime class
devicePlugin.deviceCoreScalingint1Device core scaling factor
devicePlugin.deviceListStrategystring"envvar"Device list strategy. Supported values: envvar, volume-mounts, cdi-annotations
devicePlugin.deviceMemoryScalingint1Device memory scaling factor
devicePlugin.deviceSplitCountint10Number of slices to split each GPU into
devicePlugin.disablecorelimitstring"false"Disable core limit
devicePlugin.enableNumaTopologyboolfalseAdvertise the physical GPU's NUMA node on each vGPU replica
devicePlugin.enabledbooltrueEnable device plugin
devicePlugin.extraArgslist["-v=4"]Extra arguments for device plugin
devicePlugin.extraEnvslist[]Extra environment variables for device plugin as a Kubernetes env list.
devicePlugin.gdrcopyEnabledboolfalseEnable GDRCopy
devicePlugin.gdsEnabledboolfalseEnable GDS
devicePlugin.gpuOperatorToolkitReady.enabledboolfalseEnable GPU operator toolkit ready check
devicePlugin.gpuOperatorToolkitReady.hostPathstring"/run/nvidia/validations"Host path for GPU operator toolkit ready check
devicePlugin.gpuOperatorToolkitReady.securityContextobject{"privileged":true,"runAsUser":0}Security context for the GPU operator toolkit validation init container
devicePlugin.hostNetworkboolfalseShare the host network namespace with the device plugin pod
devicePlugin.hostPIDbooltrueShare the host PID namespace with the device plugin pod
devicePlugin.image.pullPolicystring"IfNotPresent"Device plugin image pull policy
devicePlugin.image.pullSecretslist[]Specify docker-registry secret names as an array
devicePlugin.image.registrystring"ghcr.io"Device plugin image registry
devicePlugin.image.repositorystring"dynamia-ai/hami-enterprise"Device plugin image repository
devicePlugin.image.tagstring""Device plugin image tag (immutable tags are recommended)
devicePlugin.libPathstring"/usr/local/vgpu"Library path for vGPU
devicePlugin.migStrategystring"none"MIG strategy. Supported values: none, single, mixed
devicePlugin.mofedEnabledboolfalseEnable Mellanox OFED
devicePlugin.monitor.ctrPathstring"/usr/local/vgpu/containers"Container path for monitor
devicePlugin.monitor.extraArgslist["-v=4"]Extra arguments for monitor
devicePlugin.monitor.extraEnvslist[]Extra environment variables for monitor as a Kubernetes env list.
devicePlugin.monitor.image.pullPolicystring"IfNotPresent"Monitor image pull policy
devicePlugin.monitor.image.pullSecretslist[]Specify docker-registry secret names as an array
devicePlugin.monitor.image.registrystring"ghcr.io"Monitor image registry
devicePlugin.monitor.image.repositorystring"dynamia-ai/hami-enterprise"Monitor image repository
devicePlugin.monitor.image.tagstring""Monitor image tag (immutable tags are recommended)
devicePlugin.monitor.resourcesobject{}Resources for monitor container
devicePlugin.monitor.resyncIntervalstring"5m"Resync interval for monitor
devicePlugin.monitor.securityContextobject{"allowPrivilegeEscalation":false,"capabilities":{"add":["SYS_ADMIN"],"drop":["ALL"]}}Security context for the monitor container
devicePlugin.nodeConfiguration.configstring"{\n \"nodeconfig\": [\n {\n \"name\": \"your-node-name\",\n \"operatingmode\": \"hami-core\",\n \"devicememoryscaling\": 1,\n \"devicesplitcount\": 10,\n \"preconfigureddevicememory\": 0,\n \"enablenumatopology\": false,\n \"migstrategy\": \"none\",\n \"filterdevices\": {\n \"uuid\": [],\n \"index\": []\n },\n \"enablegetpreferredallocation\": false\n }\n ]\n}\n"Node configuration for device plugin. If set, overrides the default config.json
devicePlugin.nodeConfiguration.externalConfigNamestring""Name of an existing ConfigMap to use for node configuration
devicePlugin.numaRefitobject{"caFile":"","caSecret":"","enabled":false,"schedulerEndpoint":"","tlsInsecure":true}NUMA alignment refit configuration
devicePlugin.numaRefit.caFilestring""Path to a CA bundle already mounted in the device-plugin Pod
devicePlugin.numaRefit.caSecretstring""Secret containing ca.crt, mounted at /etc/hami/numa-refit-ca
devicePlugin.numaRefit.enabledboolfalseEnable NUMA alignment refit
devicePlugin.numaRefit.schedulerEndpointstring""Scheduler base URL override. Defaults to the in-cluster Scheduler service
devicePlugin.numaRefit.tlsInsecurebooltrueSkip Scheduler TLS verification. Set false with caFile or caSecret to verify TLS
devicePlugin.nvidiaDriverRootstring""NVIDIA driver root on the host, mounted into the device plugin and monitor containers
devicePlugin.nvidiaHookPathstring""Path for NVIDIA hook
devicePlugin.nvidiaNodeSelectorobject{"gpu":"on"}Node selector for NVIDIA device plugin
devicePlugin.passDeviceSpecsEnabledboolfalseEnable passing device specs
devicePlugin.pluginPathstring"/var/lib/kubelet/device-plugins"Path for device plugin socket
devicePlugin.podAnnotationsobject{}Annotations for device plugin pod
devicePlugin.preConfiguredDeviceMemoryint0Pre-configured device memory in MB for GPUs that don't support memory query. Set to 0 for auto-detection
devicePlugin.resourcesobject{}Resources for device plugin container
devicePlugin.runtimeClassNamestring""Runtime class name to be used by the device plugin
devicePlugin.securityContextobject{"allowPrivilegeEscalation":true,"capabilities":{"add":["SYS_ADMIN"],"drop":["ALL"]},"privileged":true}Security context for the device plugin container
devicePlugin.service.annotationsobject{}Annotations for device plugin service
devicePlugin.service.httpPortint31992HTTP port for device plugin service
devicePlugin.service.labelsobject{}Labels for device plugin service
devicePlugin.service.typestring"NodePort"Service type for device plugin
devicePlugin.tolerationslist[{"effect":"NoSchedule","key":"nvidia.com/gpu","operator":"Exists"}]Tolerations for device plugin pod
devicePlugin.updateStrategy.rollingUpdate.maxUnavailableint1Maximum unavailable pods during update
devicePlugin.updateStrategy.typestring"RollingUpdate"Update strategy for device plugin DaemonSet
devices.alibaba.customresourceslist["alibaba.com/ppu","alibaba.com/ppu-memory","alibaba.com/gpu-core"]Custom resource definitions for Alibaba devices
devices.alibaba.enabledbooltrueEnable Alibaba PPU device
devices.alibaba.ppuCorePolicystring"default"PPU core policy for Alibaba devices
devices.amd.customresourceslist["amd.com/gpu","amd.com/gpumem","amd.com/gpucores"]Custom resource definitions for AMD devices
devices.ascend.customresourceslist["huawei.com/Ascend910A","huawei.com/Ascend910A-memory","huawei.com/Ascend910A-core","huawei.com/Ascend910B2","huawei.com/Ascend910B2-memory","huawei.com/Ascend910B2-core","huawei.com/Ascend910B3","huawei.com/Ascend910B3-memory","huawei.com/Ascend910B3-core","huawei.com/Ascend910B4","huawei.com/Ascend910B4-memory","huawei.com/Ascend910B4-core","huawei.com/Ascend910B4-1","huawei.com/Ascend910B4-1-memory","huawei.com/Ascend910B4-1-core","huawei.com/Ascend310P","huawei.com/Ascend310P-memory","huawei.com/Ascend310P-core","huawei.com/Ascend910C","huawei.com/Ascend910C-memory","huawei.com/Ascend910C-core"]Custom resource definitions for Ascend devices
devices.ascend.deviceCoreScalingint1hami-core advertised percentage capacity multiplier; each workload still requests at most 100 percent.
devices.ascend.deviceSplitCountint10Number of virtual devices advertised per physical Ascend NPU in soft-slicing mode. A positive node-level vDeviceCount overrides it.
devices.ascend.enabledboolfalseEnable Ascend device plugin
devices.ascend.enpuPolicystring"elastic"Default ENPU policy; enabling ENPU also requires compatible runtime assets on the node.
devices.ascend.extraArgslist[]Extra arguments for Ascend device plugin
devices.ascend.hamiCorePolicystring"spatial"Default hami-core runtime policy; Pod and node settings take precedence. Explicit enterprise force/disable controls remain effective.
devices.ascend.hamiVnpuCoreboolfalseEnable soft-partitioning based on hami-vnpu-core (node-level annotation takes higher priority)
devices.ascend.hamiVnpuModestring""Backend: template, hami-core (alias hamiCore), or enpu. Empty preserves hamiVnpuCore compatibility.
devices.ascend.imagestring""Ascend device plugin image
devices.ascend.imagePullPolicystring"IfNotPresent"Image pull policy for Ascend device plugin
devices.ascend.nodeSelectorobject{"ascend":"on"}Node selector for Ascend device plugin
devices.ascend.runtimeClassNamestring""Runtime class name for Ascend NPU pods
devices.ascend.tolerationslist[]Tolerations for Ascend device plugin
devices.awsneuron.customresourceslist["aws.amazon.com/neuron","aws.amazon.com/neuroncore"]Custom resource definitions for AWS Neuron devices
devices.biren.customresourceslist["birentech.com/gpu"]Custom resource definitions for Biren devices
devices.biren.enabledbooltrueEnable Biren device plugin
devices.enflame.customresourceslist["enflame.com/drs-gcu","enflame.com/gcu-memory","enflame.com/gcu-core","enflame.com/gcu","enflame.com/vgcu","enflame.com/vgcu-percentage"]Custom resource definitions for Enflame devices
devices.enflame.enabledbooltrueEnable Enflame device plugin
devices.iluvatar.customresourceslist["iluvatar.ai/BI-V100-vgpu","iluvatar.ai/BI-V100.vCore","iluvatar.ai/BI-V100.vMem","iluvatar.ai/BI-V150-vgpu","iluvatar.ai/BI-V150.vCore","iluvatar.ai/BI-V150.vMem","iluvatar.ai/MR-V100-vgpu","iluvatar.ai/MR-V100.vCore","iluvatar.ai/MR-V100.vMem","iluvatar.ai/MR-V50-vgpu","iluvatar.ai/MR-V50.vCore","iluvatar.ai/MR-V50.vMem"]Custom resource definitions for Iluvatar devices
devices.iluvatar.enabledboolfalseEnable Iluvatar device plugin
devices.kunlun.customresourceslist["kunlunxin.com/xpu","kunlunxin.com/vxpu","kunlunxin.com/vxpu-memory"]Custom resource definitions for Kunlun devices
devices.kunlun.enabledbooltrueEnable Kunlun device plugin
devices.mthreads.customresourceslist["mthreads.com/vgpu"]Custom resource definitions for MThreads devices
devices.mthreads.enabledbooltrueEnable MThreads device plugin
devices.nvidia.gpuCorePolicystring"default"GPU core policy for NVIDIA devices
devices.nvidia.libCudaLogLevelint2libCUDA log level
devices.vastai.customresourceslist["vastaitech.com/va"]Custom resource definitions for Vastai devices
devices.vastai.enabledbooltrueEnable Vastai device plugin
dra.enabledboolfalseEnable DRA (Dynamic Resource Allocation)
enflameResourceNameDRSGCUstring"enflame.com/drs-gcu"Enflame DRS GCU resource name
enflameResourceNameGCUCorestring"enflame.com/gcu-core"Enflame GCU core resource name
enflameResourceNameGCUMemorystring"enflame.com/gcu-memory"Enflame GCU memory resource name
enflameResourceNameVGCUstring"enflame.com/vgcu"Enflame VGCU resource name
enflameResourceNameVGCUPercentagestring"enflame.com/vgcu-percentage"Enflame VGCU percentage resource name
fullnameOverridestring""Override the fullname of the chart
global.annotationsobject{}Global annotations to add to all resources
global.gpuHookPathstring"/usr/local"Path for GPU hook installation on nodes
global.imagePullSecretslist[]Global Docker image pull secrets
global.imageRegistrystring""Global Docker image registry
global.imageTagstring"v2.10.0-r5"Global image tag for HAMi Enterprise components
global.labelsobject{}Global labels to add to all resources
global.managedNodeSelectorobject{"usage":"gpu"}Node selector configuration for managed GPU nodes
global.managedNodeSelectorEnableboolfalseEnable managed node selector for GPU nodes
hami-dra.drivers.nvidia.containerDriverbooltrueEnable container driver for NVIDIA DRA
hami-dra.drivers.nvidia.enabledbooltrueEnable NVIDIA DRA driver
hami-dra.drivers.nvidia.image.repositorystring"ghcr.io/project-hami/k8s-dra-driver"NVIDIA DRA driver image repository
hami-dra.drivers.nvidia.image.tagstring"main"NVIDIA DRA driver image tag
hami-dra.monitor.enabledbooltrueEnable monitor for hami-dra
hcuResourceCoresstring"hygon.com/hcucores"Hygon HCU cores resource name
hcuResourceMemstring"hygon.com/hcumem"Hygon HCU memory resource name
hcuResourceNamestring"hygon.com/hcunum"Hygon HCU resource name
kunlunResourceNamestring"kunlunxin.com/xpu"Kunlun XPU resource name
kunlunResourceVCountNamestring"kunlunxin.com/vxpu"Kunlun virtual XPU count resource name
kunlunResourceVMemoryNamestring"kunlunxin.com/vxpu-memory"Kunlun virtual XPU memory resource name
legacyMetricsboolfalseEnable legacy metrics
legacySelectorLabelsboolfalseKeep historical hami/hami scheduler and NVIDIA device-plugin selector labels. Enable when upgrading an existing Enterprise release and keep enabled on subsequent upgrades.
metaxResourceCorestring"metax-tech.com/vcore"Metax sGPU core resource name
metaxResourceMemstring"metax-tech.com/vmemory"Metax sGPU memory resource name
metaxResourceNamestring"metax-tech.com/sgpu"Metax sGPU resource name
metaxsGPUTopologyAwarestring"false"Enable Metax sGPU topology awareness
mluResourceCoresstring"cambricon.com/mlu.smlu.vcore"MLU cores resource name
mluResourceMemstring"cambricon.com/mlu.smlu.vmemory"MLU memory resource name
mluResourceNamestring"cambricon.com/vmlu"MLU resource name
mockDevicePlugin.enabledboolfalseEnable mock device plugin
mockDevicePlugin.image.pullPolicystring"IfNotPresent"Mock device plugin image pull policy
mockDevicePlugin.image.pullSecretslist[]Specify docker-registry secret names as an array
mockDevicePlugin.image.registrystring"docker.io"Mock device plugin image registry
mockDevicePlugin.image.repositorystring"projecthami/mock-device-plugin"Mock device plugin image repository
mockDevicePlugin.image.tagstring"1.0.1"Mock device plugin image tag
nameOverridestring""Override the chart name
namespaceOverridestring""Override the namespace
openshift.securityContextConstraints.createbooltrueCreate the named device-plugin SCC and its use ClusterRole when OpenShift support is enabled. Set this to false only when both the SCC and system:openshift:scc:<name> ClusterRole already exist, such as for the built-in privileged SCC.
openshift.securityContextConstraints.namestring"hami-device-plugin"SCC granted to enabled device-plugin service accounts. When create=false, the matching system:openshift:scc:<name> ClusterRole must already exist. The built-in privileged SCC requires create=false.
platform.openshiftboolfalseEnable OpenShift-specific resources and handling
platform.openshiftVersionstring""OpenShift version used to resolve the Red Hat OpenShift hyperkube image tag when platform.openshift=true and scheduler.kubeScheduler.image.tag is empty
podSecurityPolicy.enabledboolfalseEnable PodSecurityPolicy
prometheus.enabledboolfalseEnable Prometheus metrics
resourceCoresstring"nvidia.com/gpucores"NVIDIA GPU cores resource name
resourceMemstring"nvidia.com/gpumem"NVIDIA GPU memory resource name
resourceMemPercentagestring"nvidia.com/gpumem-percentage"NVIDIA GPU memory percentage resource name
resourceNamestring"nvidia.com/gpu"NVIDIA GPU resource name
resourcePrioritystring"nvidia.com/priority"NVIDIA GPU priority resource name
scheduler.admissionWebhook.customURL.enabledboolfalseEnable custom webhook URL
scheduler.admissionWebhook.customURL.hoststring"127.0.0.1"Webhook host (must be an endpoint using https)
scheduler.admissionWebhook.customURL.pathstring"/webhook"Webhook path
scheduler.admissionWebhook.customURL.portint31998Webhook port
scheduler.admissionWebhook.enabledbooltrueEnable admission webhook. If disabled, pods must be configured with correct schedulerName manually
scheduler.admissionWebhook.failurePolicystring"Ignore"Failure policy for webhook
scheduler.admissionWebhook.manageNamespaceSelectorbooltrueWhether the chart renders and manages the webhook's namespaceSelector field
scheduler.admissionWebhook.namespaceSelectorobject{"matchExpressions":[],"matchLabels":{}}Namespace selector for webhook
scheduler.admissionWebhook.objectSelectorobject{"matchExpressions":[]}Object selector for webhook
scheduler.admissionWebhook.reinvocationPolicystring"Never"Reinvocation policy for webhook
scheduler.admissionWebhook.whitelistNamespaceslist[]Namespaces that the webhook will not be applied to
scheduler.certManager.enabledboolfalseEnable cert-manager for TLS certificate generation
scheduler.defaultSchedulerPolicy.gpuSchedulerPolicystring"spread"GPU-level scheduling policy. Supported values: binpack, spread
scheduler.defaultSchedulerPolicy.nodeSchedulerPolicystring"binpack"Node-level scheduling policy. Supported values: binpack, spread
scheduler.extender.extraArgslist["--debug","-v=4"]Extra arguments for scheduler extender
scheduler.extender.image.pullPolicystring"IfNotPresent"Scheduler extender image pull policy
scheduler.extender.image.pullSecretslist[]Specify docker-registry secret names as an array
scheduler.extender.image.registrystring"ghcr.io"Scheduler extender image registry
scheduler.extender.image.repositorystring"dynamia-ai/hami-enterprise"Scheduler extender image repository
scheduler.extender.image.tagstring""Scheduler extender image tag (immutable tags are recommended)
scheduler.extender.resourcesobject{}Resources for scheduler extender container
scheduler.forceOverwriteDefaultSchedulerbooltrueForce overwrite the default scheduler name in pods when it equals the K8s default scheduler name
scheduler.kubeScheduler.enabledbooltrueWhether to run kube-scheduler container in the scheduler pod
scheduler.kubeScheduler.extraArgslist["--policy-config-file=/config/config.json","-v=4"]Extra arguments for kube-scheduler (legacy config format)
scheduler.kubeScheduler.extraNewArgslist["--config=/config/config.yaml","-v=4"]Extra arguments for kube-scheduler (new config format)
scheduler.kubeScheduler.image.pullPolicystring"IfNotPresent"kube-scheduler image pull policy
scheduler.kubeScheduler.image.pullSecretslist[]Specify docker-registry secret names as an array
scheduler.kubeScheduler.image.registrystring"registry.cn-hangzhou.aliyuncs.com"kube-scheduler image registry
scheduler.kubeScheduler.image.repositorystring"google_containers/kube-scheduler"kube-scheduler image repository
scheduler.kubeScheduler.image.tagstring""kube-scheduler image tag (immutable tags are recommended)
scheduler.kubeScheduler.resourcesobject{}Resources for kube-scheduler container
scheduler.leaderElectbooltrueEnable leader election for scheduler
scheduler.livenessProbeboolfalseEnable liveness probe for scheduler
scheduler.metricsBindAddressstring":9395"Address to bind metrics server
scheduler.nodeLockExpirestring"5m"Node lock expiration time
scheduler.nodeNamestring""Node name for scheduler pod. If installing as default scheduler, set this to skip the schedule workflow
scheduler.overwriteEnvstring"false"Overwrite environment variables
scheduler.patch.enabledbooltrueEnable kube-webhook-certgen patch job
scheduler.patch.image.pullPolicystring"IfNotPresent"kube-webhook-certgen image pull policy
scheduler.patch.image.pullSecretslist[]Specify docker-registry secret names as an array
scheduler.patch.image.registrystring"docker.io"kube-webhook-certgen image registry
scheduler.patch.image.repositorystring"jettech/kube-webhook-certgen"kube-webhook-certgen image repository
scheduler.patch.image.tagstring"v1.5.2"kube-webhook-certgen image tag
scheduler.patch.imageNew.pullPolicystring"IfNotPresent"Kubernetes 1.22+ kube-webhook-certgen image pull policy
scheduler.patch.imageNew.pullSecretslist[]Specify docker-registry secret names as an array
scheduler.patch.imageNew.registrystring"registry.k8s.io"Kubernetes 1.22+ kube-webhook-certgen image registry
scheduler.patch.imageNew.repositorystring"ingress-nginx/kube-webhook-certgen"Kubernetes 1.22+ kube-webhook-certgen image repository
scheduler.patch.imageNew.tagstring"v1.6.9"Kubernetes 1.22+ kube-webhook-certgen image tag
scheduler.patch.nodeSelectorobject{}Node selector for patch job
scheduler.patch.podAnnotationsobject{}Annotations for patch job pod
scheduler.patch.priorityClassNamestring""Priority class name for patch job
scheduler.patch.runAsUserint2000User ID to run patch job as. OpenShift omits this value at render time so its SCC can allocate an allowed UID
scheduler.patch.tolerationslist[]Tolerations for patch job
scheduler.podAnnotationsobject{}Annotations for scheduler pod
scheduler.replicasint1Number of scheduler replicas (only effective when leaderElect is true)
scheduler.service.annotationsobject{}Annotations for scheduler service
scheduler.service.httpPortint443HTTP port for scheduler service
scheduler.service.httpTargetPortint9443Target port on which the scheduler extender listens
scheduler.service.labelsobject{}Labels for scheduler service
scheduler.service.monitorPortint31993Monitoring port for scheduler service
scheduler.service.monitorTargetPortint9395Target port for monitoring
scheduler.service.schedulerPortint31998NodePort for scheduler HTTP
scheduler.service.typestring"NodePort"Service type for scheduler
scheduler.tolerationslist[]Tolerations for scheduler pod
scheduler.useDownwardbooltrueWhether to enable useDownward in the generated scheduler device config for NVIDIA devices
schedulerNamestring"hami-scheduler"Name of the scheduler
selinux.enabledboolfalseRelabel shared vGPU host directories on SELinux-enabled Kubernetes nodes
selinux.levelstring"s0"SELinux level applied to shared vGPU host directories
selinux.typestring"container_file_t"SELinux type applied to shared vGPU host directories
vastaiResourceNamestring"vastaitech.com/va"Vastai resource name