HAMi Enterprise 2.9.0-rc2

Chart version	Declared appVersion	Release date
2.9.0-rc2	2.9.0-rc2	2026-06-08

This is a patch release. Changes below are compared with HAMi Enterprise v2.9.0-rc1.

Software changes

• Adjust GPU memory auto-expansion granularity. For containers with CUDA_MEMORY_AUTO_ADJUST=true, the target memory allocation is rounded up to the next 1024 MiB (1 GiB). For example, 1536 MiB becomes 2048 MiB, while an existing allocation of 2048 MiB remains unchanged. GPU memory auto-expansion is an existing feature; this change does not enable it automatically for other containers.
• Add Alibaba PPU to license management. The license state of a PPU device is checked when allocating it to a new workload. Devices without a valid license are no longer eligible for allocation. Verify that the required license is valid before upgrading an existing PPU deployment.

Chart changes

Configuration or identifier	2.9.0-rc1	2.9.0-rc2
Chart name	hami	hami-enterprise
Package	hami-2.9.0-rc1.tgz	hami-enterprise-2.9.0-rc2.tgz
Default schedulerName	hami-scheduler	hami-enterprise-scheduler
Default Enterprise component image repository	ghcr.io/dynamia-ai/hami	ghcr.io/dynamia-ai/hami-enterprise
Default global.imageTag	v2.9.0-rc1	v2.9.0-rc2
Scheduler app.kubernetes.io/component	hami-scheduler	hami-enterprise-scheduler
Device plugin app.kubernetes.io/component	hami-device-plugin	hami-enterprise-device-plugin
Default app.kubernetes.io/name	hami	hami-enterprise

The image repository changes correspond to scheduler.extender.image.repository, devicePlugin.image.repository, and devicePlugin.monitor.image.repository. If you use a private image registry, synchronize the target images and update these settings; changing only the tags is insufficient.

Renaming the Chart also changes default resource names. For example, with a Helm release named hami and no name overrides, the scheduler Deployment changes from hami-scheduler to hami-hami-enterprise-scheduler, and the device-plugin DaemonSet changes from hami-device-plugin to hami-hami-enterprise-device-plugin.

Upgrade notes

Before upgrading, save the current Helm values and resource manifests. Render the target Chart with the actual release name and compare resource names, selectors, and Service references. Switch from the old resources to the new ones during a maintenance window to avoid two schedulers or device plugins managing the same workloads and nodes at the same time.

• If a workload explicitly sets spec.schedulerName: hami-scheduler, make it match the target deployment's schedulerName: either update the workload template to use the new default or explicitly retain the old name in the Chart. The webhook does not automatically change an explicitly specified old HAMi scheduler name.
• Using nameOverride or fullnameOverride to retain old resource names does not retain old selectors. If the rendered .spec.selector changes for a Deployment or DaemonSet with the same name, recreate the resource; this field cannot be changed in place.
• Update monitoring, inspection, and automation scripts that select Pods using the old component labels. Check custom Service or webhook integrations for references to old resource names as well.
• For workloads with GPU memory auto-expansion enabled, verify available-memory headroom and capacity expectations against the new 1 GiB granularity.

Known limitation

The values validation introduced in this release restricts the following settings to null. Helm validation fails even when a valid, non-empty value is provided:

Setting	Affected custom value
devicePlugin.nvidiaDriverRoot, devicePlugin.nvidiaHookPath	Driver-root or hook-path string
devicePlugin.gdrcopyEnabled, devicePlugin.gdsEnabled, devicePlugin.mofedEnabled	Boolean
scheduler.admissionWebhook.whitelistNamespaces	Namespace list
scheduler.admissionWebhook.namespaceSelector.matchLabels	Label map

Deployments that depend on these settings cannot install this release with their existing custom values unchanged. The validation issue is fixed in 2.9.0-r3. Do not remove settings that are still required by your workloads just to pass validation.

Obtain the package 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 2.9.0-rc2 release package and are preserved as published. Use this version's values.yaml as the source of default configuration.

Expand values table

Key	Type	Default	Description
dcuResourceCores	string	"hygon.com/dcucores"
dcuResourceMem	string	"hygon.com/dcumem"
dcuResourceName	string	"hygon.com/dcunum"
devicePlugin.createRuntimeClass	bool	false
devicePlugin.deviceCoreScaling	int	1
devicePlugin.deviceListStrategy	string	"envvar"
devicePlugin.deviceMemoryScaling	int	1
devicePlugin.deviceSplitCount	int	10
devicePlugin.disablecorelimit	string	"false"
devicePlugin.enabled	bool	true
devicePlugin.extraArgs[0]	string	"-v=4"
devicePlugin.extraEnvs	object	{}
devicePlugin.gdrcopyEnabled	string	nil
devicePlugin.gdsEnabled	string	nil
devicePlugin.gpuOperatorToolkitReady.enabled	bool	false
devicePlugin.gpuOperatorToolkitReady.hostPath	string	"/run/nvidia/validations"
devicePlugin.image.pullPolicy	string	"IfNotPresent"
devicePlugin.image.pullSecrets	list	[]
devicePlugin.image.registry	string	"ghcr.io"
devicePlugin.image.repository	string	"dynamia-ai/hami-enterprise"
devicePlugin.image.tag	string	""
devicePlugin.libPath	string	"/usr/local/vgpu"
devicePlugin.migStrategy	string	"none"
devicePlugin.mofedEnabled	string	nil
devicePlugin.monitor.ctrPath	string	"/usr/local/vgpu/containers"
devicePlugin.monitor.extraArgs[0]	string	"-v=4"
devicePlugin.monitor.extraEnvs	object	{}
devicePlugin.monitor.image.pullPolicy	string	"IfNotPresent"
devicePlugin.monitor.image.pullSecrets	list	[]
devicePlugin.monitor.image.registry	string	"ghcr.io"
devicePlugin.monitor.image.repository	string	"dynamia-ai/hami-enterprise"
devicePlugin.monitor.image.tag	string	""
devicePlugin.monitor.resources	object	{}
devicePlugin.monitor.resyncInterval	string	"5m"
devicePlugin.nodeConfiguration.config	string	"{\n \"nodeconfig\": [\n {\n \"name\": \"your-node-name\",\n \"operatingmode\": \"hami-core\",\n \"devicememoryscaling\": 1,\n \"devicesplitcount\": 10,\n \"preconfigureddevicememory\": 0,\n \"migstrategy\": \"none\",\n \"filterdevices\": {\n \"uuid\": [],\n \"index\": []\n },\n \"enablegetpreferredallocation\": false\n }\n ]\n}\n"
devicePlugin.nodeConfiguration.externalConfigName	string	""
devicePlugin.nvidiaDriverRoot	string	nil
devicePlugin.nvidiaHookPath	string	nil
devicePlugin.nvidiaNodeSelector.gpu	string	"on"
devicePlugin.passDeviceSpecsEnabled	bool	false
devicePlugin.pluginPath	string	"/var/lib/kubelet/device-plugins"
devicePlugin.podAnnotations	object	{}
devicePlugin.preConfiguredDeviceMemory	int	0
devicePlugin.resources	object	{}
devicePlugin.runtimeClassName	string	""
devicePlugin.service.annotations	object	{}
devicePlugin.service.httpPort	int	31992
devicePlugin.service.labels	object	{}
devicePlugin.service.type	string	"NodePort"
devicePlugin.tolerations	list	[]
devicePlugin.updateStrategy.rollingUpdate.maxUnavailable	int	1
devicePlugin.updateStrategy.type	string	"RollingUpdate"
devices.alibaba.customresources[0]	string	"alibaba.com/ppu"
devices.alibaba.customresources[1]	string	"alibaba.com/ppu-memory"
devices.alibaba.customresources[2]	string	"alibaba.com/gpu-core"
devices.alibaba.enabled	bool	true
devices.alibaba.ppuCorePolicy	string	"default"
devices.amd.customresources[0]	string	"amd.com/gpu"
devices.amd.customresources[1]	string	"amd.com/gpumem"
devices.ascend.customresources[0]	string	"huawei.com/Ascend910A"
devices.ascend.customresources[10]	string	"huawei.com/Ascend910B4-memory"
devices.ascend.customresources[11]	string	"huawei.com/Ascend910B4-core"
devices.ascend.customresources[12]	string	"huawei.com/Ascend910B4-1"
devices.ascend.customresources[13]	string	"huawei.com/Ascend910B4-1-memory"
devices.ascend.customresources[14]	string	"huawei.com/Ascend910B4-1-core"
devices.ascend.customresources[15]	string	"huawei.com/Ascend310P"
devices.ascend.customresources[16]	string	"huawei.com/Ascend310P-memory"
devices.ascend.customresources[17]	string	"huawei.com/Ascend310P-core"
devices.ascend.customresources[18]	string	"huawei.com/Ascend910C"
devices.ascend.customresources[19]	string	"huawei.com/Ascend910C-memory"
devices.ascend.customresources[1]	string	"huawei.com/Ascend910A-memory"
devices.ascend.customresources[20]	string	"huawei.com/Ascend910C-core"
devices.ascend.customresources[2]	string	"huawei.com/Ascend910A-core"
devices.ascend.customresources[3]	string	"huawei.com/Ascend910B2"
devices.ascend.customresources[4]	string	"huawei.com/Ascend910B2-memory"
devices.ascend.customresources[5]	string	"huawei.com/Ascend910B2-core"
devices.ascend.customresources[6]	string	"huawei.com/Ascend910B3"
devices.ascend.customresources[7]	string	"huawei.com/Ascend910B3-memory"
devices.ascend.customresources[8]	string	"huawei.com/Ascend910B3-core"
devices.ascend.customresources[9]	string	"huawei.com/Ascend910B4"
devices.ascend.enabled	bool	false
devices.ascend.extraArgs	list	[]
devices.ascend.hamiVnpuCore	bool	false
devices.ascend.image	string	""
devices.ascend.imagePullPolicy	string	"IfNotPresent"
devices.ascend.nodeSelector.ascend	string	"on"
devices.ascend.runtimeClassName	string	""
devices.ascend.tolerations	list	[]
devices.awsneuron.customresources[0]	string	"aws.amazon.com/neuron"
devices.awsneuron.customresources[1]	string	"aws.amazon.com/neuroncore"
devices.enflame.customresources[0]	string	"enflame.com/vgcu"
devices.enflame.customresources[1]	string	"enflame.com/vgcu-percentage"
devices.enflame.customresources[2]	string	"enflame.com/gcu"
devices.enflame.enabled	bool	true
devices.iluvatar.customresources[0]	string	"iluvatar.ai/BI-V100-vgpu"
devices.iluvatar.customresources[10]	string	"iluvatar.ai/MR-V50.vCore"
devices.iluvatar.customresources[11]	string	"iluvatar.ai/MR-V50.vMem"
devices.iluvatar.customresources[1]	string	"iluvatar.ai/BI-V100.vCore"
devices.iluvatar.customresources[2]	string	"iluvatar.ai/BI-V100.vMem"
devices.iluvatar.customresources[3]	string	"iluvatar.ai/BI-V150-vgpu"
devices.iluvatar.customresources[4]	string	"iluvatar.ai/BI-V150.vCore"
devices.iluvatar.customresources[5]	string	"iluvatar.ai/BI-V150.vMem"
devices.iluvatar.customresources[6]	string	"iluvatar.ai/MR-V100-vgpu"
devices.iluvatar.customresources[7]	string	"iluvatar.ai/MR-V100.vCore"
devices.iluvatar.customresources[8]	string	"iluvatar.ai/MR-V100.vMem"
devices.iluvatar.customresources[9]	string	"iluvatar.ai/MR-V50-vgpu"
devices.iluvatar.enabled	bool	false
devices.kunlun.customresources[0]	string	"kunlunxin.com/xpu"
devices.kunlun.customresources[1]	string	"kunlunxin.com/vxpu"
devices.kunlun.customresources[2]	string	"kunlunxin.com/vxpu-memory"
devices.kunlun.enabled	bool	true
devices.mthreads.customresources[0]	string	"mthreads.com/vgpu"
devices.mthreads.enabled	bool	true
devices.nvidia.gpuCorePolicy	string	"default"
devices.nvidia.libCudaLogLevel	int	2
devices.vastai.customresources[0]	string	"vastaitech.com/va"
devices.vastai.enabled	bool	true
dra.enabled	bool	false
enflameResourceNameVGCU	string	"enflame.com/vgcu"
enflameResourceNameVGCUPercentage	string	"enflame.com/vgcu-percentage"
fullnameOverride	string	""
global.annotations	object	{}
global.gpuHookPath	string	"/usr/local"
global.imagePullSecrets	list	[]
global.imageRegistry	string	""
global.imageTag	string	"v2.9.0-rc2"
global.labels	object	{}
global.managedNodeSelector.usage	string	"gpu"
global.managedNodeSelectorEnable	bool	false
hami-dra.drivers.nvidia.containerDriver	bool	true
hami-dra.drivers.nvidia.enabled	bool	true
hami-dra.drivers.nvidia.image.repository	string	"ghcr.io/project-hami/k8s-dra-driver"
hami-dra.drivers.nvidia.image.tag	string	"main"
hami-dra.monitor.enabled	bool	true
kunlunResourceName	string	"kunlunxin.com/xpu"
kunlunResourceVCountName	string	"kunlunxin.com/vxpu"
kunlunResourceVMemoryName	string	"kunlunxin.com/vxpu-memory"
legacyMetrics	bool	false
metaxResourceCore	string	"metax-tech.com/vcore"
metaxResourceMem	string	"metax-tech.com/vmemory"
metaxResourceName	string	"metax-tech.com/sgpu"
metaxsGPUTopologyAware	string	"false"
mluResourceCores	string	"cambricon.com/mlu.smlu.vcore"
mluResourceMem	string	"cambricon.com/mlu.smlu.vmemory"
mluResourceName	string	"cambricon.com/vmlu"
mockDevicePlugin.enabled	bool	false
mockDevicePlugin.image.pullPolicy	string	"IfNotPresent"
mockDevicePlugin.image.pullSecrets	list	[]
mockDevicePlugin.image.registry	string	"docker.io"
mockDevicePlugin.image.repository	string	"projecthami/mock-device-plugin"
mockDevicePlugin.image.tag	string	"1.0.1"
nameOverride	string	""
namespaceOverride	string	""
podSecurityPolicy.enabled	bool	false
prometheus.enabled	bool	false
resourceCores	string	"nvidia.com/gpucores"
resourceMem	string	"nvidia.com/gpumem"
resourceMemPercentage	string	"nvidia.com/gpumem-percentage"
resourceName	string	"nvidia.com/gpu"
resourcePriority	string	"nvidia.com/priority"
scheduler.admissionWebhook.customURL.enabled	bool	false
scheduler.admissionWebhook.customURL.host	string	"127.0.0.1"
scheduler.admissionWebhook.customURL.path	string	"/webhook"
scheduler.admissionWebhook.customURL.port	int	31998
scheduler.admissionWebhook.enabled	bool	true
scheduler.admissionWebhook.failurePolicy	string	"Ignore"
scheduler.admissionWebhook.namespaceSelector.matchExpressions	list	[]
scheduler.admissionWebhook.namespaceSelector.matchLabels	string	nil
scheduler.admissionWebhook.objectSelector.matchExpressions	list	[]
scheduler.admissionWebhook.reinvocationPolicy	string	"Never"
scheduler.admissionWebhook.whitelistNamespaces	string	nil
scheduler.certManager.enabled	bool	false
scheduler.defaultSchedulerPolicy.gpuSchedulerPolicy	string	"spread"
scheduler.defaultSchedulerPolicy.nodeSchedulerPolicy	string	"binpack"
scheduler.extender.extraArgs[0]	string	"--debug"
scheduler.extender.extraArgs[1]	string	"-v=4"
scheduler.extender.image.pullPolicy	string	"IfNotPresent"
scheduler.extender.image.pullSecrets	list	[]
scheduler.extender.image.registry	string	"ghcr.io"
scheduler.extender.image.repository	string	"dynamia-ai/hami-enterprise"
scheduler.extender.image.tag	string	""
scheduler.extender.resources	object	{}
scheduler.forceOverwriteDefaultScheduler	bool	true
scheduler.kubeScheduler.enabled	bool	true
scheduler.kubeScheduler.extraArgs[0]	string	"--policy-config-file=/config/config.json"
scheduler.kubeScheduler.extraArgs[1]	string	"-v=4"
scheduler.kubeScheduler.extraNewArgs[0]	string	"--config=/config/config.yaml"
scheduler.kubeScheduler.extraNewArgs[1]	string	"-v=4"
scheduler.kubeScheduler.image.pullPolicy	string	"IfNotPresent"
scheduler.kubeScheduler.image.pullSecrets	list	[]
scheduler.kubeScheduler.image.registry	string	"registry.cn-hangzhou.aliyuncs.com"
scheduler.kubeScheduler.image.repository	string	"google_containers/kube-scheduler"
scheduler.kubeScheduler.image.tag	string	""
scheduler.kubeScheduler.resources	object	{}
scheduler.leaderElect	bool	true
scheduler.livenessProbe	bool	false
scheduler.metricsBindAddress	string	":9395"
scheduler.nodeLockExpire	string	"5m"
scheduler.nodeName	string	""
scheduler.overwriteEnv	string	"false"
scheduler.patch.enabled	bool	true
scheduler.patch.image.pullPolicy	string	"IfNotPresent"
scheduler.patch.image.pullSecrets	list	[]
scheduler.patch.image.registry	string	"docker.io"
scheduler.patch.image.repository	string	"jettech/kube-webhook-certgen"
scheduler.patch.image.tag	string	"v1.5.2"
scheduler.patch.imageNew.pullPolicy	string	"IfNotPresent"
scheduler.patch.imageNew.pullSecrets	list	[]
scheduler.patch.imageNew.registry	string	"docker.io"
scheduler.patch.imageNew.repository	string	"liangjw/kube-webhook-certgen"
scheduler.patch.imageNew.tag	string	"v1.1.1"
scheduler.patch.nodeSelector	object	{}
scheduler.patch.podAnnotations	object	{}
scheduler.patch.priorityClassName	string	""
scheduler.patch.runAsUser	int	2000
scheduler.patch.tolerations	list	[]
scheduler.podAnnotations	object	{}
scheduler.replicas	int	1
scheduler.service.annotations	object	{}
scheduler.service.httpPort	int	443
scheduler.service.httpTargetPort	int	443
scheduler.service.labels	object	{}
scheduler.service.monitorPort	int	31993
scheduler.service.monitorTargetPort	int	9395
scheduler.service.schedulerPort	int	31998
scheduler.service.type	string	"NodePort"
scheduler.tolerations	list	[]
schedulerName	string	"hami-enterprise-scheduler"
vastaiResourceName	string	"vastaitech.com/va"