nvml-mock

Mock NVIDIA driver infrastructure for Kubernetes testing

This page is generated from values.yaml. For installation guides and operational behaviour, see the Helm chart guide.

Requirements

Kubernetes: >= 1.28.0-0

Values

Key	Type	Default	Description
affinity	object	{}	Affinity rules for scheduling node-agent pods.
allocationWatcher.enabled	bool	false	Enable allocation-based synthetic memory metrics.
allocationWatcher.interval	string	"2s"	Poll period for the kubelet pod-resources socket.
allocationWatcher.podResourcesSocket	string	"/var/lib/kubelet/pod-resources/kubelet.sock"	Kubelet pod-resources socket path.
allocationWatcher.resources.limits.cpu	string	"100m"	CPU limit for the allocation watcher.
allocationWatcher.resources.limits.memory	string	"64Mi"	Memory limit for the allocation watcher.
allocationWatcher.resources.requests.cpu	string	"10m"	CPU request for the allocation watcher.
allocationWatcher.resources.requests.memory	string	"32Mi"	Memory request for the allocation watcher.
allocationWatcher.usedFractionPerClaim	float	0.5	Fraction of usable GPU memory attributed to each claim.
controlPlane.affinity	object	{}	Affinity rules for scheduling control-plane pods.
controlPlane.enabled	bool	false	Enable the Mokka control-plane service.
controlPlane.image.digest	string	""	Immutable digest; when set, it takes precedence over the tag.
controlPlane.image.pullPolicy	string	"IfNotPresent"	Kubernetes image pull policy for the control plane.
controlPlane.image.repository	string	"ghcr.io/nvidia/mokka-control-plane"	Control-plane image repository.
controlPlane.image.tag	string	""	Control-plane image tag; defaults to the chart appVersion.
controlPlane.kubeAPIBurst	int	100	Kubernetes API burst allowance for the control plane.
controlPlane.kubeAPIQPS	int	50	Kubernetes API request rate for the control plane.
controlPlane.leaderElection.name	string	"control-plane.mokka.nvidia.com"	Lease name used for control-plane leader election.
controlPlane.logging.format	string	"json"	Control-plane log encoding.
controlPlane.logging.level	string	"info"	Control-plane log level.
controlPlane.nodeSelector	object	{}	Node selector for control-plane pods.
controlPlane.pdb.enabled	bool	false	Enable a PodDisruptionBudget. Rendered only when replicas > 1.
controlPlane.podAnnotations	object	{}	Additional annotations applied to control-plane pods.
controlPlane.podLabels	object	{}	Additional labels applied to control-plane pods.
controlPlane.priorityClassName	string	""	Priority class for control-plane pods.
controlPlane.replicas	int	1	Number of control-plane replicas.
controlPlane.resources.limits.cpu	string	"500m"	CPU limit for the control plane.
controlPlane.resources.limits.memory	string	"1Gi"	Memory limit for the control plane.
controlPlane.resources.requests.cpu	string	"50m"	CPU requested by the control plane.
controlPlane.resources.requests.memory	string	"64Mi"	Memory requested by the control plane.
controlPlane.service.port	int	8080	Kubernetes Service port for the control plane.
controlPlane.service.type	string	"ClusterIP"	Kubernetes Service type for the control plane.
controlPlane.shutdownTimeout	string	"5s"	Graceful control-plane shutdown timeout.
controlPlane.terminationGracePeriodSeconds	int	30	Pod termination grace period for the control plane.
controlPlane.tolerations	list	[]	Tolerations for control-plane pods.
controlPlane.topologySpreadConstraints	list	[]	Topology spread constraints for control-plane pods.
controlPlane.workers	int	2	Number of control-plane worker goroutines.
driverBranch	string	""	Driver branch nvmlSystemGetDriverBranch reports; empty derives it from the driver version ("580.65.06" -> "r580_00").
driverVersion	string	""	Driver version reported by the simulated driver; empty derives it from the profile.
extraObjects	list	[]	Additional Kubernetes objects rendered with the release after tpl evaluation.
fabricmanager.enabled	string	""	Enable or disable the fake fabric manager; empty derives it from the profile.
fabricmanager.initDelay	string	""	Delay before publishing fabric readiness.
fabricmanager.stateDir	string	"/var/lib/nvml-mock/fabric-state"	Host directory for fabric-manager readiness state.
featureGates	object	{}
global.imagePullSecrets	list	[]	Pull secrets added to every pod created by this chart.
global.imageRegistry	string	""	Registry host override applied to every image, such as an internal mirror.
gpu.count	string	""	Number of simulated GPUs; empty derives the count from the selected profile.
gpu.customConfig	string	""	Complete YAML configuration that replaces the selected GPU profile.
gpu.dynamicMetrics.enabled	bool	false	Enable time-varying synthetic temperature, power, and utilization values.
gpu.failureInjection.after_calls	int	0	Trigger the selected failure after this many guarded calls; zero disables it.
gpu.failureInjection.enabled	bool	false	Enable deterministic or probabilistic simulated GPU failures.
gpu.failureInjection.mode	string	"healthy"	Failure mode to expose when failure injection is enabled.
gpu.failureInjection.probability	float	0	Probability that each guarded NVML call triggers the selected failure.
gpu.failureInjection.seed	int	0	Random seed; zero uses a time-based seed.
gpu.failureInjection.xid.code	int	0	Xid code reported when the failure mode emits an Xid event.
gpu.mig.enabled	bool	false	Enable MIG partitioning for profiles that support it.
gpu.mig.gpuInstances	list	[]	GPU instance layout to create on each simulated device.
gpu.profile	string	"gb300"	GPU profile and per-device simulation settings.
hic	list	[]	Host interface cards nvidia-smi -q -u reports, as {id, firmwareVersion} entries.
image.digest	string	""	Immutable image digest; when set, it takes precedence over the tag.
image.pullPolicy	string	"IfNotPresent"	Kubernetes image pull policy.
image.repository	string	"ghcr.io/nvidia/nvml-mock"	Container image repository for the node agent.
image.tag	string	""	Image tag; defaults to the chart appVersion.
imagePullSecrets	list	[]	Pull secrets for nvml-mock, NRI, and control-plane pods.
imex.binaries.download.timeout	string	"1m"	Time limit of one IMEX archive download attempt while the IMEXDownload feature gate is on.
imex.binaries.image.digest	string	""	Immutable IMEX binaries image digest; when set, it takes precedence over the tag.
imex.binaries.image.pullPolicy	string	""	IMEX binaries image pull policy; empty picks IfNotPresent, or Always for a "latest" tag.
imex.binaries.image.repository	string	""	IMEX binaries image repository; empty (default) renders no init container.
imex.binaries.image.tag	string	""	IMEX binaries image tag, the IMEX version it carries; required unless digest is set.
imex.binaries.resources.requests.cpu	string	"10m"	CPU requested by the IMEX binaries init container.
imex.binaries.resources.requests.memory	string	"32Mi"	Memory requested by the IMEX binaries init container.
imex.deviceNumbers.capsMajor	int	236	Major number used for IMEX capability devices.
imex.deviceNumbers.channelMajor	int	235	Major number used for IMEX channel devices.
infiniband.mockTier	string	""	InfiniBand mock tier; empty derives the tier from the selected profile. mockTier sets the container's MOCK_IB value: off
infiniband.ping.networkPolicy.enabled	bool	true	Restrict mock InfiniBand traffic to peer nvml-mock pods. Rendered only at the full tier, the only one that listens on the port.
infiniband.ping.networkPolicy.peerNamespaces	list	[]	Namespaces besides the release's whose nvml-mock pods may reach the port, for a fleet split across namespaces. Empty admits the release's namespace only, which keeps OVN-Kubernetes from re-listing every Pod in the cluster on each Pod event.
infiniband.ping.port	int	18515	TCP port used by the mock InfiniBand ping service.
integrations.fakeGpuOperator.enabled	bool	false	Enable fake-gpu-operator profile integration.
integrations.fakeGpuOperator.profileLabels."run.ai/gpu-profile"	string	"true"	Additional label applied to generated fake-gpu-operator profiles.
integrations.fakeGpuOperator.targetNamespace	string	""	Namespace where fake-gpu-operator reads profile ConfigMaps.
nodeAgent.kernelLog.enabled	bool	false	Announce injected Xids through the host kernel log.
nodeAgent.livenessProbe	object	{"httpGet":{"path":"/healthz","port":"health"}}	Liveness probe; /healthz fails when the last Stage wave failed. Set to null to disable.
nodeAgent.livenessProbe.httpGet.path	string	"/healthz"	Liveness probe endpoint path.
nodeAgent.livenessProbe.httpGet.port	string	"health"	Liveness probe endpoint port name or number.
nodeAgent.logging	object	{"format":"json","level":"info"}	Node-agent logging settings and lifecycle behavior.
nodeAgent.logging.format	string	"json"	Log encoding for the node agent.
nodeAgent.logging.level	string	"info"	Log level for the node agent.
nodeAgent.nriStartupTimeoutSeconds	int	120
nodeAgent.readinessProbe	object	{"httpGet":{"path":"/readyz","port":"health"}}	Readiness probe; /readyz reports whether every simulator is serving. Set to null to disable.
nodeAgent.readinessProbe.httpGet.path	string	"/readyz"	Readiness probe endpoint path.
nodeAgent.readinessProbe.httpGet.port	string	"health"	Readiness probe endpoint port name or number.
nodeAgent.resources	object	{"requests":{"cpu":"10m","memory":"32Mi"}}	Resource requests for the node agent.
nodeAgent.resources.requests.cpu	string	"10m"	CPU requested by the node agent.
nodeAgent.resources.requests.memory	string	"32Mi"	Memory requested by the node agent.
nodeAgent.resyncInterval	string	"1m"	Node-agent profile and topology resync period.
nodeAgent.shutdownTimeout	string	"5s"	Graceful shutdown timeout for the node agent.
nodeLabels.featuresDir	string	"/etc/kubernetes/node-feature-discovery/features.d"	Host directory from which NFD reads generated feature files.
nodeSelector	object	{}	Existing node labels required to schedule the mock DaemonSet.
nri.cdiSpecDir	string	"/var/run/cdi"	Host directory containing staged CDI specifications.
nri.deviceAnnotation	string	"nvml-mock.nvidia.com/devices"	Pod annotation enabling mock GPU device injection.
nri.deviceInjectionMode	string	"raw"	Device injection mode: raw nodes or CDI references.
nri.enabled	bool	true	Enable node-wide container injection through the NRI plugin. Inject the mock into containers that hold a GPU allocation or opt in by annotation. Needs containerd NRI enabled on the node: containerd 2.x enables it by default, containerd 1.7 does not. Set false where it is off; otherwise the node pod stays NotReady.
nri.excludedNamespaces	list	[]	Namespaces excluded from NRI injection.
nri.healthPort	int	8080	NRI plugin health and readiness port.
nri.image	object	{}	Optional container image overrides for the NRI plugin.
nri.imexChannelAnnotation	string	"nvml-mock.nvidia.com/imex-channels"	Pod annotation enabling IMEX channel injection.
nri.infinibandAnnotation	string	"nvml-mock.nvidia.com/infiniband"
nri.livenessProbe.failureThreshold	int	3	Consecutive failures before the plugin is restarted.
nri.livenessProbe.httpGet.path	string	"/healthz"	Liveness endpoint path.
nri.livenessProbe.httpGet.port	string	"nri-health"	Liveness endpoint port name.
nri.livenessProbe.periodSeconds	int	10	Liveness probe interval.
nri.livenessProbe.timeoutSeconds	int	2	Liveness probe timeout.
nri.logging.format	string	"json"	Log encoding for the NRI plugin.
nri.logging.level	string	"info"	Log level for the NRI plugin.
nri.nativeSidecar	bool	true
nri.optOutAnnotation	string	"nvml-mock.nvidia.com/inject"	Pod annotation that disables ambient mock injection when set to "false".
nri.overlay.hostPath	string	"/var/lib/nvml-mock"	Host path containing the staged mock driver overlay.
nri.overlay.mountPath	string	"/opt/nvml-mock"	Container mount path for the mock driver overlay.
nri.pluginIndex	string	"10"	NRI plugin ordering index.
nri.pluginName	string	"nvml-mock"	NRI plugin name.
nri.readinessProbe.failureThreshold	int	2	Consecutive failures before the plugin is marked unready.
nri.readinessProbe.httpGet.path	string	"/readyz"	Readiness endpoint path.
nri.readinessProbe.httpGet.port	string	"nri-health"	Readiness endpoint port name.
nri.readinessProbe.periodSeconds	int	10	Readiness probe interval.
nri.readinessProbe.timeoutSeconds	int	2	Readiness probe timeout.
nri.resources	object	{"requests":{"cpu":"10m","memory":"32Mi"}}	NRI plugin resource requests; limits are unset by default. Requests only: the plugin runs on every node by default, and no limit is set until its usage under load has been measured.
nri.socketPath	string	"/var/run/nri/nri.sock"	Runtime NRI socket path.
podAnnotations	object	{}	Additional annotations applied to node-agent pods.
podLabels	object	{}	Additional labels for node-agent pods; do not set app.kubernetes.io selector labels.
priorityClassName	string	""	Priority class for node-agent DaemonSet pods.
terminationGracePeriodSeconds	int	10	Pod termination grace period for the node-agent DaemonSet.
tolerations	list	[{"operator":"Exists"}]	Pod tolerations for the mock DaemonSet.
topology.domains	list	[]	Synthetic topology domains and clique membership.
topology.enabled	bool	false	Enable simulation of multi-node GPU fabric topology.
updateStrategy.rollingUpdate.maxUnavailable	string	"25%"	Maximum number of DaemonSet pods unavailable during an update.
updateStrategy.type	string	"RollingUpdate"	DaemonSet update strategy.