Skip to main content

Resource Pool Manager Chart 0.0.7

Chart: 0.0.7 · appVersion: 0.0.3 · RPM release: rpm-0.0.3 · Released: 2026-11-10 (UTC)

Software changes​

Ascend inventory and admission​

RPM 0.0.3 uses Ascend runtime modes when reporting device capacity, selecting a resource pool, and validating allocations. It reads vnpus.hamiVnpuMode from the existing scheduler device ConfigMap and matches Pod requests against the modes advertised by nodes and individual devices. Only devices that support the requested mode count toward candidate pool capacity.

For hami-core devices, RPM normalizes legacy physical core counts to a percentage capacity of 100, while retaining advertised percentage capacities above 100. Template devices do not report physical AI core counts as percentage capacity. ENPU devices report a core capacity of 100 and require an explicit huawei.com/vnpu-mode: enpu Pod annotation. ENPU admission requires exactly one Ascend device per container.

NVIDIA Downward API SM limits​

For NVIDIA workloads using useDownward, RPM now requires CUDA_DEVICE_SM_LIMIT to reference the current container's HAMi core-limit annotation through the Downward API. RPM adds this explicit reference when it is missing, so a value imported through envFrom cannot override the runtime SM limit. This applies to regular and init containers and works with either webhook order.

Older schedulers may also inject CUDA_DEVICE_CORE_LIMIT. RPM accepts that additional variable when it references the same annotation, while still requiring CUDA_DEVICE_SM_LIMIT. RPM preserves existing explicit entries and rejects incorrect references or literal values during final admission.

Chart changes​

  • appVersion changes from 0.0.2 to 0.0.3; the default image changes from ghcr.io/dynamia-ai/resource-pool-manager:v0.0.2 to ghcr.io/dynamia-ai/resource-pool-manager:v0.0.3.
  • Chart templates and all other Values defaults are unchanged from Chart 0.0.6. No new Values keys are required.
  • The separately published resource-pool CRD Chart remains hami-enterprise-crds 0.4.2, with appVersion: 0.4.0. This release does not introduce a CRD schema migration.

Upgrade notes​

Upgrade the Manager Chart and RPM image together. If installation values override image.tag, set it explicitly to v0.0.3; an existing override retains the older image.

For Ascend installations, keep devices.ascend.enabled: true and check that the existing scheduler device ConfigMap, node/device registration, and workload mode agree. vnpus.hamiVnpuMode accepts template, hami-core, or enpu. When it is unset, the legacy vnpus.hamiVnpuCore: true selects hami-core; otherwise the default is template. For ENPU workloads, set the explicit Pod mode and request one Ascend device per container. The RPM device switch does not install or upgrade the Ascend device plugin or runtime; prepare components that support the selected mode. See the Ascend device guide for mode requirements.

For NVIDIA installations using Downward API limits, check workload templates for explicit CUDA_DEVICE_SM_LIMIT or CUDA_DEVICE_CORE_LIMIT entries. Remove application-provided literal overrides or correct their references to the current container's HAMi core-limit annotation. Incorrect explicit entries can reject new or recreated Pods after upgrade.

The Chart continues to require Kubernetes 1.30 or newer, separately installed resource-pool CRDs, HAMi extender scheduling, and hami.existingConfigMap pointing to the existing scheduler device ConfigMap. After changing scheduler device configuration, upgrade RPM with the complete manager values file to regenerate webhook rules and manager configuration. See RPM installation in the HAMi Enterprise guide.

Values reference​

The parameter descriptions below come from the README shipped with Chart 0.0.7. For defaults, use this version's values.yaml.

Expand values table
KeyTypeDefaultDescription
config.createbooltrueInitialize the runtime ConfigMap if absent. It is retained outside Helm ownership for external policy updates. For render-and-apply GitOps, precreate the ConfigMap and set this to false.
config.namestring""Runtime ConfigMap name; defaults to <workload name>-config. With config.create=false it must already exist in the manager namespace.
config.poolSchedulerPolicystring"binpack"Initial pool selection policy. Later changes belong in the runtime ConfigMap, not Helm values.
devices.ascend.enabledboolfalseInclude Ascend profiles from the scheduler device ConfigMap in resource-pool admission.
devices.hygon.enabledboolfalseInclude Hygon HCU resources from the scheduler device ConfigMap in resource-pool admission.
devices.nvidia.enabledbooltrueInclude NVIDIA resources from the scheduler device ConfigMap in resource-pool admission.
fullnameOverridestring""Override the workload name. Names are limited to 43 characters to leave room for resource suffixes.
hami.existingConfigMapstring""Required name of the scheduler device ConfigMap in this release namespace. Helm reads data.device-config.yaml from the cluster and the manager mounts the same ConfigMap. Inline configuration and overrides are not supported. Upgrade the manager after scheduler device configuration changes.
hami.schedulerNamestring"hami-scheduler"Scheduler name used by the existing HAMi installation. The scheduler must use HAMi extender admission and scheduling, not DRA.
healthProbeBindAddressstring":8081"Health server listen address.
image.pullPolicystring"IfNotPresent"Resource-pool-manager image pull policy.
image.repositorystring"ghcr.io/dynamia-ai/resource-pool-manager"Resource-pool-manager image repository, including registry.
image.tagstring"v0.0.3"Required published image tag. Independent of this chart's appVersion; update explicitly when releasing a new manager image.
imagePullSecretslist[]Registry Secret names used by the manager and certificate Jobs.
leaderElectbooltrueElect a leader for inventory and status reconcilers.
metricsBindAddressstring":8080"Metrics server listen address.
nameOverridestring""Override the chart name used in names and labels.
nodeSelectorobject{}Node selector for the manager.
podAnnotationsobject{}Additional annotations for manager Pods.
replicasint1Number of manager replicas. Use leader election when running multiple replicas.
resourcesobject{}Resource requests and limits for the manager.
timeoutSecondsint10Admission webhook timeout in seconds.
tolerationslist[]Tolerations for the manager.
webhook.tls.certManagerboolfalseUse an installed cert-manager controller for certificates and CA injection. When false, certgen hook Jobs manage webhook certificates.
webhook.tls.certgen.image.pullPolicystring"IfNotPresent"Certificate generator image pull policy.
webhook.tls.certgen.image.repositorystring"registry.k8s.io/ingress-nginx/kube-webhook-certgen"Certificate generator image repository, including registry.
webhook.tls.certgen.image.tagstring"v1.6.9"Certificate generator image tag.
webhook.tls.certgen.nodeSelectorobject{}Node selector for certificate Jobs.
webhook.tls.certgen.podAnnotationsobject{}Additional annotations for certificate Job Pods.
webhook.tls.certgen.priorityClassNamestring""Priority class for certificate Jobs.
webhook.tls.certgen.runAsUserint65532Non-root UID for certificate Jobs.
webhook.tls.certgen.tolerationslist[]Tolerations for certificate Jobs.
webhookPortint9443HTTPS port in the manager container.