404
Not Found
Examples and demos
Table of contents
This page contains usage examples and demos.
Demos
Usage demo
Demo use case
A demo on the benefits of using node feature discovery can be found in the source code repository under demo/.
Not Found
Not Found
You can reach us via the following channels:
This is a SIG-node subproject, hosted under the Kubernetes SIGs organization in Github. The project was established in 2016 and was migrated to Kubernetes SIGs in 2018.
This is open source software released under the Apache 2.0 License.
You can reach us via the following channels:
This is a SIG-node subproject, hosted under the Kubernetes SIGs organization in Github. The project was established in 2016 and was migrated to Kubernetes SIGs in 2018.
This is open source software released under the Apache 2.0 License.
Node Feature Discovery provides a Helm chart to manage its deployment.
NOTE: NFD is not ideal for other Helm charts to depend on as that may result in multiple parallel NFD deployments in the same cluster which is not fully supported by the NFD Helm chart.
Helm package manager should be installed.
To install the latest stable version:
export NFD_NS=node-feature-discovery
+ Helm · Node Feature Discovery
Deployment with Helm
Table of contents
Node Feature Discovery provides a Helm chart to manage its deployment.
NOTE: NFD is not ideal for other Helm charts to depend on as that may result in multiple parallel NFD deployments in the same cluster which is not fully supported by the NFD Helm chart.
Prerequisites
Helm package manager should be installed.
Deployment
To install the latest stable version:
export NFD_NS=node-feature-discovery
helm repo add nfd https://kubernetes-sigs.github.io/node-feature-discovery/charts
helm repo update
helm install nfd/node-feature-discovery --namespace $NFD_NS --create-namespace --generate-name
@@ -12,4 +12,4 @@ helm install nfd/node-feature-discovery install nfd/node-feature-discovery --set nameOverride=NFDinstance --set master.replicaCount=2 --namespace $NFD_NS --create-namespace
Uninstalling the chart
To uninstall the node-feature-discovery deployment:
export NFD_NS=node-feature-discovery
helm uninstall node-feature-discovery --namespace $NFD_NS
-
The command removes all the Kubernetes components associated with the chart and deletes the release. It also runs a post-delete hook that cleans up the nodes of all labels, annotations, taints and extended resources that were created by NFD.
Chart parameters
To tailor the deployment of the Node Feature Discovery to your needs following Chart parameters are available.
General parameters
Name Type Default Description image.repositorystring gcr.io/k8s-staging-nfd/node-feature-discoveryNFD image repository image.tagstring masterNFD image tag image.pullPolicystring AlwaysImage pull policy imagePullSecretslist [] ImagePullSecrets is an optional list of references to secrets in the same namespace to use for pulling any of the images used by this PodSpec. If specified, these secrets will be passed to individual puller implementations for them to use. For example, in the case of docker, only DockerConfig type secrets are honored. More info nameOverridestring Override the name of the chart fullnameOverridestring Override a default fully qualified app name tls.enablebool false Specifies whether to use TLS for communications between components. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release tls.certManagerbool false If enabled, requires cert-manager to be installed and will automatically create the required TLS certificates. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release tls.certManager.certManagerCertificate.issuerNamestring If specified, it will use a pre-existing issuer instead for the required TLS certificates. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release tls.certManager.certManagerCertificate.issuerKindstring Specifies on what kind of issuer is used, can be either ClusterIssuer or Issuer (default). Requires tls.certManager.certManagerCertificate.issuerName to be set. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release featureGates.NodeFeatureAPIbool true Enable the NodeFeature CRD API for communicating node features. This will automatically disable the gRPC communication. featureGates.NodeFeatureGroupAPIbool false Enable the NodeFeatureGroup CRD API. featureGates.DisableAutoPrefixbool false Enable DisableAutoPrefix feature gate. Disables automatic prefixing of unprefixed labels, annotations and extended resources. enableNodeFeatureApibool true Enable the NodeFeature CRD API for communicating node features. This will automatically disable the gRPC communication. DEPRECATED: will be removed in NFD v0.17, use featureGates.NodeFeatureAPI instead. prometheus.enablebool false Specifies whether to expose metrics using prometheus operator prometheus.labelsdict {} Specifies labels for use with the prometheus operator to control how it is selected prometheus.scrapeIntervalstring 10s Specifies the interval by which metrics are scraped priorityClassNamestring The name of the PriorityClass to be used for the NFD pods.
Metrics are configured to be exposed using prometheus operator API's by default. If you want to expose metrics using the prometheus operator API's you need to install the prometheus operator in your cluster.
Master pod parameters
Name Type Default description master.*dict NFD master deployment configuration master.enablebool true Specifies whether nfd-master should be deployed master.portinteger Specifies the TCP port that nfd-master listens for incoming requests. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release master.metricsPortinteger 8081 Port on which to expose metrics from components to prometheus operator master.instancestring Instance name. Used to separate annotation namespaces for multiple parallel deployments master.resyncPeriodstring NFD API controller resync period. master.extraLabelNsarray [] List of allowed extra label namespaces master.resourceLabelsarray [] List of labels to be registered as extended resources master.enableTaintsbool false Specifies whether to enable or disable node tainting master.crdControllerbool null Specifies whether the NFD CRD API controller is enabled. If not set, controller will be enabled if master.instance is empty. master.featureRulesControllerbool null DEPRECATED: use master.crdController instead master.replicaCountinteger 1 Number of desired pods. This is a pointer to distinguish between explicit zero and not specified master.podSecurityContextdict {} PodSecurityContext holds pod-level security attributes and common container settings master.securityContextdict {} Container security settings master.serviceAccount.createbool true Specifies whether a service account should be created master.serviceAccount.annotationsdict {} Annotations to add to the service account master.serviceAccount.namestring The name of the service account to use. If not set and create is true, a name is generated using the fullname template master.rbac.createbool true Specifies whether to create RBAC configuration for nfd-master master.service.typestring ClusterIP NFD master service type. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release master.service.portinteger 8080 NFD master service port. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release master.resources.limitsdict {memory: 4Gi} NFD master pod resources limits master.resources.requestsdict {cpu: 100m, memory: 128Mi} NFD master pod resources requests. You may want to use the same value for requests.memory and limits.memory. The "requests" value affects scheduling to accommodate pods on nodes. If there is a large difference between "requests" and "limits" and nodes experience memory pressure, the kernel may invoke the OOM Killer, even if the memory does not exceed the "limits" threshold. This can cause unexpected pod evictions. Memory cannot be compressed and once allocated to a pod, it can only be reclaimed by killing the pod. Natan Yellin 22/09/2022 that discusses this issue. master.tolerationsdict Scheduling to master node is disabled NFD master pod tolerations master.annotationsdict {} NFD master pod annotations master.affinitydict NFD master pod required node affinity master.deploymentAnnotationsdict {} NFD master deployment annotations master.nfdApiParallelisminteger 10 Specifies the maximum number of concurrent node updates. master.configdict NFD master configuration
Worker pod parameters
Name Type Default description worker.*dict NFD worker daemonset configuration worker.enablebool true Specifies whether nfd-worker should be deployed worker.metricsPort*int 8081 Port on which to expose metrics from components to prometheus operator worker.configdict NFD worker configuration worker.podSecurityContextdict {} PodSecurityContext holds pod-level security attributes and common container settings worker.securityContextdict {} Container security settings worker.serviceAccount.createbool true Specifies whether a service account for nfd-worker should be created worker.serviceAccount.annotationsdict {} Annotations to add to the service account for nfd-worker worker.serviceAccount.namestring The name of the service account to use for nfd-worker. If not set and create is true, a name is generated using the fullname template (suffixed with -worker) worker.rbac.createbool true Specifies whether to create RBAC configuration for nfd-worker worker.mountUsrSrcbool false Specifies whether to allow users to mount the hostpath /user/src. Does not work on systems without /usr/src AND a read-only /usr worker.resources.limitsdict {memory: 512Mi} NFD worker pod resources limits worker.resources.requestsdict {cpu: 5m, memory: 64Mi} NFD worker pod resources requests worker.nodeSelectordict {} NFD worker pod node selector worker.tolerationsdict {} NFD worker pod node tolerations worker.priorityClassNamestring NFD worker pod priority class worker.annotationsdict {} NFD worker pod annotations worker.daemonsetAnnotationsdict {} NFD worker daemonset annotations
Topology updater parameters
Name Type Default description topologyUpdater.*dict NFD Topology Updater configuration topologyUpdater.enablebool false Specifies whether the NFD Topology Updater should be created topologyUpdater.createCRDsbool false Specifies whether the NFD Topology Updater CRDs should be created topologyUpdater.serviceAccount.createbool true Specifies whether the service account for topology updater should be created topologyUpdater.serviceAccount.annotationsdict {} Annotations to add to the service account for topology updater topologyUpdater.serviceAccount.namestring The name of the service account for topology updater to use. If not set and create is true, a name is generated using the fullname template and -topology-updater suffix topologyUpdater.rbac.createbool true Specifies whether to create RBAC configuration for topology updater topologyUpdater.metricsPortinteger 8081 Port on which to expose prometheus metrics topologyUpdater.kubeletConfigPathstring "" Specifies the kubelet config host path topologyUpdater.kubeletPodResourcesSockPathstring "" Specifies the kubelet sock path to read pod resources topologyUpdater.updateIntervalstring 60s Time to sleep between CR updates. Non-positive value implies no CR update. topologyUpdater.watchNamespacestring *Namespace to watch pods, * for all namespaces topologyUpdater.podSecurityContextdict {} PodSecurityContext holds pod-level security attributes and common container settings topologyUpdater.securityContextdict {} Container security settings topologyUpdater.resources.limitsdict {memory: 60Mi} NFD Topology Updater pod resources limits topologyUpdater.resources.requestsdict {cpu: 50m, memory: 40Mi} NFD Topology Updater pod resources requests topologyUpdater.nodeSelectordict {} Topology updater pod node selector topologyUpdater.tolerationsdict {} Topology updater pod node tolerations topologyUpdater.annotationsdict {} Topology updater pod annotations topologyUpdater.daemonsetAnnotationsdict {} Topology updater daemonset annotations topologyUpdater.affinitydict {} Topology updater pod affinity topologyUpdater.configdict configuration topologyUpdater.podSetFingerprintbool true Enables compute and report of pod fingerprint in NRT objects. topologyUpdater.kubeletStateDirstring /var/lib/kubelet Specifies kubelet state directory path for watching state and checkpoint files. Empty value disables kubelet state tracking.
Garbage collector parameters
Name Type Default description gc.*dict NFD Garbage Collector configuration gc.enablebool true Specifies whether the NFD Garbage Collector should be created gc.serviceAccount.createbool true Specifies whether the service account for garbage collector should be created gc.serviceAccount.annotationsdict {} Annotations to add to the service account for garbage collector gc.serviceAccount.namestring The name of the service account for garbage collector to use. If not set and create is true, a name is generated using the fullname template and -gc suffix gc.rbac.createbool true Specifies whether to create RBAC configuration for garbage collector gc.intervalstring 1h Time between periodic garbage collector runs gc.podSecurityContextdict {} PodSecurityContext holds pod-level security attributes and common container settings gc.resources.limitsdict {memory: 1Gi} NFD Garbage Collector pod resources limits gc.resources.requestsdict {cpu: 10m, memory: 128Mi} NFD Garbage Collector pod resources requests gc.metricsPortinteger 8081 Port on which to serve Prometheus metrics gc.nodeSelectordict {} Garbage collector pod node selector gc.tolerationsdict {} Garbage collector pod node tolerations gc.annotationsdict {} Garbage collector pod annotations gc.deploymentAnnotationsdict {} Garbage collector deployment annotations gc.affinitydict {} Garbage collector pod affinity
Node Feature Discovery master
\ No newline at end of file
+The command removes all the Kubernetes components associated with the chart and deletes the release. It also runs a post-delete hook that cleans up the nodes of all labels, annotations, taints and extended resources that were created by NFD.
To tailor the deployment of the Node Feature Discovery to your needs following Chart parameters are available.
| Name | Type | Default | Description |
|---|---|---|---|
image.repository | string | gcr.io/k8s-staging-nfd/node-feature-discovery | NFD image repository |
image.tag | string | master | NFD image tag |
image.pullPolicy | string | Always | Image pull policy |
imagePullSecrets | list | [] | ImagePullSecrets is an optional list of references to secrets in the same namespace to use for pulling any of the images used by this PodSpec. If specified, these secrets will be passed to individual puller implementations for them to use. For example, in the case of docker, only DockerConfig type secrets are honored. More info |
nameOverride | string | Override the name of the chart | |
fullnameOverride | string | Override a default fully qualified app name | |
tls.enable | bool | false | Specifies whether to use TLS for communications between components. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release |
tls.certManager | bool | false | If enabled, requires cert-manager to be installed and will automatically create the required TLS certificates. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release |
tls.certManager.certManagerCertificate.issuerName | string | If specified, it will use a pre-existing issuer instead for the required TLS certificates. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release | |
tls.certManager.certManagerCertificate.issuerKind | string | Specifies on what kind of issuer is used, can be either ClusterIssuer or Issuer (default). Requires tls.certManager.certManagerCertificate.issuerName to be set. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release | |
featureGates.NodeFeatureAPI | bool | true | Enable the NodeFeature CRD API for communicating node features. This will automatically disable the gRPC communication. |
featureGates.NodeFeatureGroupAPI | bool | false | Enable the NodeFeatureGroup CRD API. |
featureGates.DisableAutoPrefix | bool | false | Enable DisableAutoPrefix feature gate. Disables automatic prefixing of unprefixed labels, annotations and extended resources. |
enableNodeFeatureApi | bool | true | Enable the NodeFeature CRD API for communicating node features. This will automatically disable the gRPC communication. DEPRECATED: will be removed in NFD v0.17, use featureGates.NodeFeatureAPI instead. |
prometheus.enable | bool | false | Specifies whether to expose metrics using prometheus operator |
prometheus.labels | dict | {} | Specifies labels for use with the prometheus operator to control how it is selected |
prometheus.scrapeInterval | string | 10s | Specifies the interval by which metrics are scraped |
priorityClassName | string | The name of the PriorityClass to be used for the NFD pods. |
Metrics are configured to be exposed using prometheus operator API's by default. If you want to expose metrics using the prometheus operator API's you need to install the prometheus operator in your cluster.
| Name | Type | Default | description |
|---|---|---|---|
master.* | dict | NFD master deployment configuration | |
master.enable | bool | true | Specifies whether nfd-master should be deployed |
master.port | integer | Specifies the TCP port that nfd-master listens for incoming requests. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release | |
master.metricsPort | integer | 8081 | Port on which to expose metrics from components to prometheus operator |
master.instance | string | Instance name. Used to separate annotation namespaces for multiple parallel deployments | |
master.resyncPeriod | string | NFD API controller resync period. | |
master.extraLabelNs | array | [] | List of allowed extra label namespaces |
master.resourceLabels | array | [] | List of labels to be registered as extended resources |
master.enableTaints | bool | false | Specifies whether to enable or disable node tainting |
master.crdController | bool | null | Specifies whether the NFD CRD API controller is enabled. If not set, controller will be enabled if master.instance is empty. |
master.featureRulesController | bool | null | DEPRECATED: use master.crdController instead |
master.replicaCount | integer | 1 | Number of desired pods. This is a pointer to distinguish between explicit zero and not specified |
master.podSecurityContext | dict | {} | PodSecurityContext holds pod-level security attributes and common container settings |
master.securityContext | dict | {} | Container security settings |
master.serviceAccount.create | bool | true | Specifies whether a service account should be created |
master.serviceAccount.annotations | dict | {} | Annotations to add to the service account |
master.serviceAccount.name | string | The name of the service account to use. If not set and create is true, a name is generated using the fullname template | |
master.rbac.create | bool | true | Specifies whether to create RBAC configuration for nfd-master |
master.service.type | string | ClusterIP | NFD master service type. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release |
master.service.port | integer | 8080 | NFD master service port. NOTE: this parameter is related to the deprecated gRPC API and will be removed with it in a future release |
master.resources.limits | dict | {memory: 4Gi} | NFD master pod resources limits |
master.resources.requests | dict | {cpu: 100m, memory: 128Mi} | NFD master pod resources requests. You may want to use the same value for requests.memory and limits.memory. The "requests" value affects scheduling to accommodate pods on nodes. If there is a large difference between "requests" and "limits" and nodes experience memory pressure, the kernel may invoke the OOM Killer, even if the memory does not exceed the "limits" threshold. This can cause unexpected pod evictions. Memory cannot be compressed and once allocated to a pod, it can only be reclaimed by killing the pod. Natan Yellin 22/09/2022 that discusses this issue. |
master.tolerations | dict | Scheduling to master node is disabled | NFD master pod tolerations |
master.annotations | dict | {} | NFD master pod annotations |
master.affinity | dict | NFD master pod required node affinity | |
master.deploymentAnnotations | dict | {} | NFD master deployment annotations |
master.nfdApiParallelism | integer | 10 | Specifies the maximum number of concurrent node updates. |
master.config | dict | NFD master configuration |
| Name | Type | Default | description |
|---|---|---|---|
worker.* | dict | NFD worker daemonset configuration | |
worker.enable | bool | true | Specifies whether nfd-worker should be deployed |
worker.metricsPort* | int | 8081 | Port on which to expose metrics from components to prometheus operator |
worker.config | dict | NFD worker configuration | |
worker.podSecurityContext | dict | {} | PodSecurityContext holds pod-level security attributes and common container settings |
worker.securityContext | dict | {} | Container security settings |
worker.serviceAccount.create | bool | true | Specifies whether a service account for nfd-worker should be created |
worker.serviceAccount.annotations | dict | {} | Annotations to add to the service account for nfd-worker |
worker.serviceAccount.name | string | The name of the service account to use for nfd-worker. If not set and create is true, a name is generated using the fullname template (suffixed with -worker) | |
worker.rbac.create | bool | true | Specifies whether to create RBAC configuration for nfd-worker |
worker.mountUsrSrc | bool | false | Specifies whether to allow users to mount the hostpath /user/src. Does not work on systems without /usr/src AND a read-only /usr |
worker.resources.limits | dict | {memory: 512Mi} | NFD worker pod resources limits |
worker.resources.requests | dict | {cpu: 5m, memory: 64Mi} | NFD worker pod resources requests |
worker.nodeSelector | dict | {} | NFD worker pod node selector |
worker.tolerations | dict | {} | NFD worker pod node tolerations |
worker.priorityClassName | string | NFD worker pod priority class | |
worker.annotations | dict | {} | NFD worker pod annotations |
worker.daemonsetAnnotations | dict | {} | NFD worker daemonset annotations |
| Name | Type | Default | description |
|---|---|---|---|
topologyUpdater.* | dict | NFD Topology Updater configuration | |
topologyUpdater.enable | bool | false | Specifies whether the NFD Topology Updater should be created |
topologyUpdater.createCRDs | bool | false | Specifies whether the NFD Topology Updater CRDs should be created |
topologyUpdater.serviceAccount.create | bool | true | Specifies whether the service account for topology updater should be created |
topologyUpdater.serviceAccount.annotations | dict | {} | Annotations to add to the service account for topology updater |
topologyUpdater.serviceAccount.name | string | The name of the service account for topology updater to use. If not set and create is true, a name is generated using the fullname template and -topology-updater suffix | |
topologyUpdater.rbac.create | bool | true | Specifies whether to create RBAC configuration for topology updater |
topologyUpdater.metricsPort | integer | 8081 | Port on which to expose prometheus metrics |
topologyUpdater.kubeletConfigPath | string | "" | Specifies the kubelet config host path |
topologyUpdater.kubeletPodResourcesSockPath | string | "" | Specifies the kubelet sock path to read pod resources |
topologyUpdater.updateInterval | string | 60s | Time to sleep between CR updates. Non-positive value implies no CR update. |
topologyUpdater.watchNamespace | string | * | Namespace to watch pods, * for all namespaces |
topologyUpdater.podSecurityContext | dict | {} | PodSecurityContext holds pod-level security attributes and common container settings |
topologyUpdater.securityContext | dict | {} | Container security settings |
topologyUpdater.resources.limits | dict | {memory: 60Mi} | NFD Topology Updater pod resources limits |
topologyUpdater.resources.requests | dict | {cpu: 50m, memory: 40Mi} | NFD Topology Updater pod resources requests |
topologyUpdater.nodeSelector | dict | {} | Topology updater pod node selector |
topologyUpdater.tolerations | dict | {} | Topology updater pod node tolerations |
topologyUpdater.annotations | dict | {} | Topology updater pod annotations |
topologyUpdater.daemonsetAnnotations | dict | {} | Topology updater daemonset annotations |
topologyUpdater.affinity | dict | {} | Topology updater pod affinity |
topologyUpdater.config | dict | configuration | |
topologyUpdater.podSetFingerprint | bool | true | Enables compute and report of pod fingerprint in NRT objects. |
topologyUpdater.kubeletStateDir | string | /var/lib/kubelet | Specifies kubelet state directory path for watching state and checkpoint files. Empty value disables kubelet state tracking. |
| Name | Type | Default | description |
|---|---|---|---|
gc.* | dict | NFD Garbage Collector configuration | |
gc.enable | bool | true | Specifies whether the NFD Garbage Collector should be created |
gc.serviceAccount.create | bool | true | Specifies whether the service account for garbage collector should be created |
gc.serviceAccount.annotations | dict | {} | Annotations to add to the service account for garbage collector |
gc.serviceAccount.name | string | The name of the service account for garbage collector to use. If not set and create is true, a name is generated using the fullname template and -gc suffix | |
gc.rbac.create | bool | true | Specifies whether to create RBAC configuration for garbage collector |
gc.interval | string | 1h | Time between periodic garbage collector runs |
gc.podSecurityContext | dict | {} | PodSecurityContext holds pod-level security attributes and common container settings |
gc.resources.limits | dict | {memory: 1Gi} | NFD Garbage Collector pod resources limits |
gc.resources.requests | dict | {cpu: 10m, memory: 128Mi} | NFD Garbage Collector pod resources requests |
gc.metricsPort | integer | 8081 | Port on which to serve Prometheus metrics |
gc.nodeSelector | dict | {} | Garbage collector pod node selector |
gc.tolerations | dict | {} | Garbage collector pod node tolerations |
gc.annotations | dict | {} | Garbage collector pod annotations |
gc.deploymentAnnotations | dict | {} | Garbage collector deployment annotations |
gc.affinity | dict | {} | Garbage collector pod affinity |
NFD offers two variants of the container image. Released container images are available for x86_64 and Arm64 architectures.
The default is a minimal image based on scratch and only supports running statically linked binaries.
For backwards compatibility a container image tag with suffix -minimal (e.g. gcr.io/k8s-staging-nfd/node-feature-discovery:master-minimal) is provided.
This image is based on debian:bookworm-slim and contains a full Linux system for running shell-based nfd-worker hooks and doing live debugging and diagnosis of the NFD images.
The container image tag has suffix -full (e.g. gcr.io/k8s-staging-nfd/node-feature-discovery:master-full).
NFD offers two variants of the container image. Released container images are available for x86_64 and Arm64 architectures.
The default is a minimal image based on scratch and only supports running statically linked binaries.
For backwards compatibility a container image tag with suffix -minimal (e.g. gcr.io/k8s-staging-nfd/node-feature-discovery:master-minimal) is provided.
This image is based on debian:bookworm-slim and contains a full Linux system for running shell-based nfd-worker hooks and doing live debugging and diagnosis of the NFD images.
The container image tag has suffix -full (e.g. gcr.io/k8s-staging-nfd/node-feature-discovery:master-full).
Node Feature Discovery can be deployed on any recent version of Kubernetes (v1.24+).
See Image variants for description of the different NFD container images available.
Using Kustomize provides straightforward deployment with kubectl integration and declarative customization.
Using Helm provides easy management of NFD deployments with nice configuration management and easy upgrades.
Using Operator provides deployment and configuration management via CRDs.
Node Feature Discovery can be deployed on any recent version of Kubernetes (v1.24+).
See Image variants for description of the different NFD container images available.
Using Kustomize provides straightforward deployment with kubectl integration and declarative customization.
Using Helm provides easy management of NFD deployments with nice configuration management and easy upgrades.
Using Operator provides deployment and configuration management via CRDs.
Kustomize can be used to deploy NFD. Customization of the deployment is done by maintaining declarative overlays on top of the base overlays in NFD.
To follow the deployment instructions here, kubectl v1.24 or later is required.
The kustomize overlays provided in the repo can be used directly:
kubectl apply -k https://github.com/kubernetes-sigs/node-feature-discovery/deployment/overlays/default?ref=master
+ Kustomize · Node Feature Discovery
Deployment with Kustomize
Table of contents
Kustomize can be used to deploy NFD. Customization of the deployment is done by maintaining declarative overlays on top of the base overlays in NFD.
To follow the deployment instructions here, kubectl v1.24 or later is required.
The kustomize overlays provided in the repo can be used directly:
kubectl apply -k https://github.com/kubernetes-sigs/node-feature-discovery/deployment/overlays/default?ref=master
This will required RBAC rules and deploy nfd-master (as a deployment) and nfd-worker (as daemonset) in the node-feature-discovery namespace.
NOTE: nfd-topology-updater is not deployed as part of the default overlay. Refer to the Master Worker Topologyupdater and Topologyupdater below.
Alternatively you can clone the repository and customize the deployment by creating your own overlays. See kustomize for more information about managing deployment configurations.
Overlays
The NFD repository hosts a set of overlays for different usages and deployment scenarios under deployment/overlays
default: default deployment of nfd-worker as a daemonset, described abovedefault-job: see Worker one-shot belowmaster-worker-topologyupdater: see Master Worker Topologyupdater belowtopologyupdater: see Topology Updater belowprometheus: see Metrics belowprune: clean up the cluster after uninstallation, see Removing feature labelssamples/cert-manager: an example for supplementing the default deployment with cert-manager for TLS authentication, see Automated TLS certificate management using cert-manager for detailssamples/custom-rules: an example for spicing up the default deployment with a separately managed configmap of custom labeling rules, see Custom feature source for more information about custom node labels
Worker one-shot
Feature discovery can alternatively be configured as a one-shot job. The default-job overlay may be used to achieve this:
NUM_NODES=$(kubectl get no -o jsonpath='{.items[*].metadata.name}' | wc -w)
kubectl kustomize https://github.com/kubernetes-sigs/node-feature-discovery/deployment/overlays/default-job?ref=master | \
sed s"/NUM_NODES/$NUM_NODES/" | \
@@ -21,4 +21,4 @@ kubectl -n $NFD_NS delete svc nf
kubectl -n $NFD_NS delete sa nfd-master
kubectl delete clusterrole nfd-master
kubectl delete clusterrolebinding nfd-master
-
Node Feature Discovery master
\ No newline at end of file
+Metrics are configured to be exposed using prometheus operator API's by default. If you want to expose metrics using the prometheus operator API's you need to install the prometheus operator in your cluster. By default NFD Master and Worker expose metrics on port 8081.
The exposed metrics are
| Metric | Type | Description |
|---|---|---|
nfd_master_build_info | Gauge | Version from which nfd-master was built |
nfd_worker_build_info | Gauge | Version from which nfd-worker was built |
nfd_gc_build_info | Gauge | Version from which nfd-gc was built |
nfd_topology_updater_build_info | Gauge | Version from which nfd-topology-updater was built |
nfd_node_update_requests_total | Counter | Number of node update requests received by the master over gRPC |
nfd_node_updates_total | Counter | Number of nodes updated |
nfd_node_update_failures_total | Counter | Number of nodes update failures |
nfd_node_labels_rejected_total | Counter | Number of nodes labels rejected by nfd-master |
nfd_node_extendedresources_rejected_total | Counter | Number of nodes extended resources rejected by nfd-master |
nfd_node_taints_rejected_total | Counter | Number of nodes taints rejected by nfd-master |
nfd_nodefeaturerule_processing_duration_seconds | Histogram | Time taken to process NodeFeatureRule objects |
nfd_nodefeaturerule_processing_errors_total | Counter | Number or errors encountered while processing NodeFeatureRule objects |
nfd_feature_discovery_duration_seconds | Histogram | Time taken to discover features on a node |
nfd_topology_updater_scan_errors_total | Counter | Number of errors in scanning resource allocation of pods. |
nfd_gc_objects_deleted_total | Counter | Number of NodeFeature and NodeResourceTopology objects garbage collected. |
nfd_gc_object_delete_failures_total | Counter | Number of errors in deleting NodeFeature and NodeResourceTopology objects. |
To deploy NFD with metrics enabled using kustomize, you can use the prometheus overlay.
By default metrics are enabled when deploying NFD via Helm. To enable Prometheus to scrape metrics from NFD, you need to pass the following values to Helm:
--set prometheus.enable=true
-For more info on Helm deployment, see Helm.
It is recommended to specify --set prometheus.prometheusSpec.podMonitorSelectorNilUsesHelmValues=false when deploying prometheus-operator via Helm to enable the prometheus-operator to scrape metrics from any PodMonitor.
or setting labels on the PodMonitor via the helm parameter prometheus.labels to control which Prometheus instances will scrape this PodMonitor.
NFD contains an example Grafana dashboard. You can import examples/grafana-dashboard.json to your Grafana instance to visualize the NFD metrics.
Metrics are configured to be exposed using prometheus operator API's by default. If you want to expose metrics using the prometheus operator API's you need to install the prometheus operator in your cluster. By default NFD Master and Worker expose metrics on port 8081.
The exposed metrics are
| Metric | Type | Description |
|---|---|---|
nfd_master_build_info | Gauge | Version from which nfd-master was built |
nfd_worker_build_info | Gauge | Version from which nfd-worker was built |
nfd_gc_build_info | Gauge | Version from which nfd-gc was built |
nfd_topology_updater_build_info | Gauge | Version from which nfd-topology-updater was built |
nfd_node_update_requests_total | Counter | Number of node update requests received by the master over gRPC |
nfd_node_updates_total | Counter | Number of nodes updated |
nfd_node_update_failures_total | Counter | Number of nodes update failures |
nfd_node_labels_rejected_total | Counter | Number of nodes labels rejected by nfd-master |
nfd_node_extendedresources_rejected_total | Counter | Number of nodes extended resources rejected by nfd-master |
nfd_node_taints_rejected_total | Counter | Number of nodes taints rejected by nfd-master |
nfd_nodefeaturerule_processing_duration_seconds | Histogram | Time taken to process NodeFeatureRule objects |
nfd_nodefeaturerule_processing_errors_total | Counter | Number or errors encountered while processing NodeFeatureRule objects |
nfd_feature_discovery_duration_seconds | Histogram | Time taken to discover features on a node |
nfd_topology_updater_scan_errors_total | Counter | Number of errors in scanning resource allocation of pods. |
nfd_gc_objects_deleted_total | Counter | Number of NodeFeature and NodeResourceTopology objects garbage collected. |
nfd_gc_object_delete_failures_total | Counter | Number of errors in deleting NodeFeature and NodeResourceTopology objects. |
To deploy NFD with metrics enabled using kustomize, you can use the prometheus overlay.
By default metrics are enabled when deploying NFD via Helm. To enable Prometheus to scrape metrics from NFD, you need to pass the following values to Helm:
--set prometheus.enable=true
+For more info on Helm deployment, see Helm.
It is recommended to specify --set prometheus.prometheusSpec.podMonitorSelectorNilUsesHelmValues=false when deploying prometheus-operator via Helm to enable the prometheus-operator to scrape metrics from any PodMonitor.
or setting labels on the PodMonitor via the helm parameter prometheus.labels to control which Prometheus instances will scrape this PodMonitor.
NFD contains an example Grafana dashboard. You can import examples/grafana-dashboard.json to your Grafana instance to visualize the NFD metrics.
The Node Feature Discovery Operator automates installation, configuration and updates of NFD using a specific NodeFeatureDiscovery custom resource. This also provides good support for managing NFD as a dependency of other operators.
Deployment using the Node Feature Discovery Operator is recommended to be done via operatorhub.io.
Install the operator:
kubectl create -f https://operatorhub.io/install/nfd-operator.yaml
+ NFD Operator · Node Feature Discovery
Deployment with NFD Operator
Table of contents
The Node Feature Discovery Operator automates installation, configuration and updates of NFD using a specific NodeFeatureDiscovery custom resource. This also provides good support for managing NFD as a dependency of other operators.
Deployment
Deployment using the Node Feature Discovery Operator is recommended to be done via operatorhub.io.
- You need to have OLM installed. If you don't, take a look at the latest release for detailed instructions.
-
Install the operator:
kubectl create -f https://operatorhub.io/install/nfd-operator.yaml
-
Create NodeFeatureDiscovery object (in nfd namespace here):
cat << EOF | kubectl apply -f -
apiVersion: v1
kind: Namespace
@@ -17,4 +17,4 @@ spec:
EOF
Uninstallation
If you followed the deployment instructions above you can uninstall NFD with:
kubectl -n nfd delete NodeFeatureDiscovery my-nfd-deployment
Optionally, you can also remove the namespace:
kubectl delete ns nfd
-
See the node-feature-discovery-operator and OLM project documentation for instructions for uninstalling the operator and operator lifecycle manager, respectively.
Node Feature Discovery master
\ No newline at end of file
+See the node-feature-discovery-operator and OLM project documentation for instructions for uninstalling the operator and operator lifecycle manager, respectively.
DEPRECATED: this section only applies when the gRPC API is used, i.e. when the NodeFeature API is disabled (via the
-feature-gates NodeFeatureAPI=falseflag) on both nfd-master and nfd-worker. The gRPC API is deprecated and will be removed in a future release.
NFD supports mutual TLS authentication between the nfd-master and nfd-worker instances. That is, nfd-worker and nfd-master both verify that the other end presents a valid certificate.
TLS authentication is enabled by specifying -ca-file, -key-file and -cert-file args, on both the nfd-master and nfd-worker instances. The template specs provided with NFD contain (commented out) example configuration for enabling TLS authentication.
The Common Name (CN) of the nfd-master certificate must match the DNS name of the nfd-master Service of the cluster. By default, nfd-master only check that the nfd-worker has been signed by the specified root certificate (-ca-file).
Additional hardening can be enabled by specifying -verify-node-name in nfd-master args, in which case nfd-master verifies that the NodeName presented by nfd-worker matches the Common Name (CN) or a Subject Alternative Name (SAN) of its certificate.
cert-manager can be used to automate certificate management between nfd-master and the nfd-worker pods.
The NFD source code repository contains an example kustomize overlay and helm chart that can be used to deploy NFD with cert-manager supplied certificates enabled.
To install cert-manager itself, you can run:
kubectl apply -f https://github.com/cert-manager/cert-manager/releases/download/v1.13.2/cert-manager.yaml
+ TLS authentication · Node Feature Discovery
Communication security with TLS
Table of contents
DEPRECATED: this section only applies when the gRPC API is used, i.e. when the NodeFeature API is disabled (via the -feature-gates NodeFeatureAPI=false flag) on both nfd-master and nfd-worker. The gRPC API is deprecated and will be removed in a future release.
NFD supports mutual TLS authentication between the nfd-master and nfd-worker instances. That is, nfd-worker and nfd-master both verify that the other end presents a valid certificate.
TLS authentication is enabled by specifying -ca-file, -key-file and -cert-file args, on both the nfd-master and nfd-worker instances. The template specs provided with NFD contain (commented out) example configuration for enabling TLS authentication.
The Common Name (CN) of the nfd-master certificate must match the DNS name of the nfd-master Service of the cluster. By default, nfd-master only check that the nfd-worker has been signed by the specified root certificate (-ca-file).
Additional hardening can be enabled by specifying -verify-node-name in nfd-master args, in which case nfd-master verifies that the NodeName presented by nfd-worker matches the Common Name (CN) or a Subject Alternative Name (SAN) of its certificate.
Automated TLS certificate management using cert-manager
cert-manager can be used to automate certificate management between nfd-master and the nfd-worker pods.
The NFD source code repository contains an example kustomize overlay and helm chart that can be used to deploy NFD with cert-manager supplied certificates enabled.
To install cert-manager itself, you can run:
kubectl apply -f https://github.com/cert-manager/cert-manager/releases/download/v1.13.2/cert-manager.yaml
Alternatively, you can refer to cert-manager documentation for other installation methods such as the Helm chart they provide.
When using the Helm chart to deploy NFD, override values.yaml to enable both the tls.enabled and tls.certManager options. Note that if you do not enable tls.certManager, helm will successfully install the application, but deployment will wait until certificates are manually created, as demonstrated below.
See the sample installation commands in the Helm Deployment and Configuration sections above for how to either override individual values, or provide a yaml file with which to override default values.
Manual TLS certificate management
If you do not with to make use of cert-manager, the certificates can be manually created and stored as secrets within the NFD namespace.
Create a CA certificate
openssl req -x509 -newkey rsa:4096 -keyout ca.key -nodes \
-subj "/CN=nfd-ca" -days 10000 -out ca.crt
Create a common openssl config file.
cat <<EOF > nfd-common.conf
@@ -77,4 +77,4 @@ data:
EOF
done
-
Node Feature Discovery master
\ No newline at end of file
+Follow the uninstallation instructions of the deployment method used (kustomize, helm or operator).
NOTE: This is unnecessary when using the Helm chart for deployment as it will clean up the nodes when NFD is uninstalled.
NFD-Master has a special -prune command line flag for removing all nfd-related node labels, annotations, extended resources and taints from the cluster.
kubectl apply -k https://github.com/kubernetes-sigs/node-feature-discovery/deployment/overlays/prune?ref=master
+ Uninstallation · Node Feature Discovery
Uninstallation
Follow the uninstallation instructions of the deployment method used (kustomize, helm or operator).
Removing feature labels
NOTE: This is unnecessary when using the Helm chart for deployment as it will clean up the nodes when NFD is uninstalled.
NFD-Master has a special -prune command line flag for removing all nfd-related node labels, annotations, extended resources and taints from the cluster.
kubectl apply -k https://github.com/kubernetes-sigs/node-feature-discovery/deployment/overlays/prune?ref=master
kubectl -n node-feature-discovery wait job.batch/nfd-master --for=condition=complete && \
kubectl delete -k https://github.com/kubernetes-sigs/node-feature-discovery/deployment/overlays/prune?ref=master
-
NOTE: You must run prune before removing the RBAC rules (serviceaccount, clusterrole and clusterrolebinding).
Node Feature Discovery master
\ No newline at end of file
+NOTE: You must run prune before removing the RBAC rules (serviceaccount, clusterrole and clusterrolebinding).
git clone https://github.com/kubernetes-sigs/node-feature-discovery
+ Developer guide · Node Feature Discovery
Developer guide
Table of contents
Building from source
Download the source code
git clone https://github.com/kubernetes-sigs/node-feature-discovery
cd node-feature-discovery
Docker build
Build the container image
See customizing the build below for altering the container image registry, for example.
make
Push the container image
Optional, this example with Docker.
docker push <IMAGE_TAG>
@@ -27,4 +27,4 @@ kubectl apply -k .
tilt up
This will override the default value(master) of IMAGE_TAG_NAME variable defined in the Tiltfile.
Documentation
All documentation resides under the docs directory in the source tree. It is designed to be served as a html site by GitHub Pages.
Building the documentation is containerized to fix the build environment. The recommended way for developing documentation is to run:
make site-serve
This will build the documentation in a container and serve it under localhost:4000/ making it easy to verify the results. Any changes made to the docs/ will automatically re-trigger a rebuild and are reflected in the served content and can be inspected with a browser refresh.
To just build the html documentation run:
make site-build
-
This will generate html documentation under docs/_site/.
Node Feature Discovery master
\ No newline at end of file
+This will generate html documentation under docs/_site/.
Welcome to Node Feature Discovery – a Kubernetes add-on for detecting hardware features and system configuration!
Continue to:
Introduction for more details on the project.
Quick start for quick step-by-step instructions on how to get NFD running on your cluster.
$ kubectl apply -k https://github.com/kubernetes-sigs/node-feature-discovery/deployment/overlays/default?ref=master
+ Get started · Node Feature Discovery
Node Feature Discovery
Welcome to Node Feature Discovery – a Kubernetes add-on for detecting hardware features and system configuration!
Continue to:
-
Introduction for more details on the project.
-
Quick start for quick step-by-step instructions on how to get NFD running on your cluster.
Quick-start – the short-short version
$ kubectl apply -k https://github.com/kubernetes-sigs/node-feature-discovery/deployment/overlays/default?ref=master
namespace/node-feature-discovery created
serviceaccount/nfd-master created
clusterrole.rbac.authorization.k8s.io/nfd-master created
@@ -22,4 +22,4 @@
"feature.node.kubernetes.io/cpu-cpuid.AESNI": "true",
...
-
Node Feature Discovery master
\ No newline at end of file
+This software enables node feature discovery for Kubernetes. It detects hardware features available on each node in a Kubernetes cluster, and advertises those features using node labels and optionally node extended resources, annotations and node taints. Node Feature Discovery is compatible with any recent version of Kubernetes (v1.24+).
NFD consists of four software components:
NFD-Master is the daemon responsible for communication towards the Kubernetes API. That is, it receives labeling requests from the worker and modifies node objects accordingly.
NFD-Worker is a daemon responsible for feature detection. It then communicates the information to nfd-master which does the actual node labeling. One instance of nfd-worker is supposed to be running on each node of the cluster,
NFD-Topology-Updater is a daemon responsible for examining allocated resources on a worker node to account for resources available to be allocated to new pod on a per-zone basis (where a zone can be a NUMA node). It then creates or updates a NodeResourceTopology custom resource object specific to this node. One instance of nfd-topology-updater is supposed to be running on each node of the cluster.
NFD-GC is a daemon responsible for cleaning obsolete NodeFeature and NodeResourceTopology objects.
One instance of nfd-gc is supposed to be running in the cluster.
Feature discovery is divided into domain-specific feature sources:
Each feature source is responsible for detecting a set of features which. in turn, are turned into node feature labels. Feature labels are prefixed with feature.node.kubernetes.io/ and also contain the name of the feature source. Non-standard user-specific feature labels can be created with the local and custom feature sources.
An overview of the default feature labels:
{
+ Introduction · Node Feature Discovery
Introduction
Table of contents
- NFD-Master
- NFD-Worker
- NFD-Topology-Updater
- NFD-GC
- Feature Discovery
- Node annotations
- Custom resources
This software enables node feature discovery for Kubernetes. It detects hardware features available on each node in a Kubernetes cluster, and advertises those features using node labels and optionally node extended resources, annotations and node taints. Node Feature Discovery is compatible with any recent version of Kubernetes (v1.24+).
NFD consists of four software components:
- nfd-master
- nfd-worker
- nfd-topology-updater
- nfd-gc
NFD-Master
NFD-Master is the daemon responsible for communication towards the Kubernetes API. That is, it receives labeling requests from the worker and modifies node objects accordingly.
NFD-Worker
NFD-Worker is a daemon responsible for feature detection. It then communicates the information to nfd-master which does the actual node labeling. One instance of nfd-worker is supposed to be running on each node of the cluster,
NFD-Topology-Updater
NFD-Topology-Updater is a daemon responsible for examining allocated resources on a worker node to account for resources available to be allocated to new pod on a per-zone basis (where a zone can be a NUMA node). It then creates or updates a NodeResourceTopology custom resource object specific to this node. One instance of nfd-topology-updater is supposed to be running on each node of the cluster.
NFD-GC
NFD-GC is a daemon responsible for cleaning obsolete NodeFeature and NodeResourceTopology objects.
One instance of nfd-gc is supposed to be running in the cluster.
Feature Discovery
Feature discovery is divided into domain-specific feature sources:
- CPU
- Kernel
- Memory
- Network
- PCI
- Storage
- System
- USB
- Custom (rule-based custom features)
- Local (hooks for user-specific features)
Each feature source is responsible for detecting a set of features which. in turn, are turned into node feature labels. Feature labels are prefixed with feature.node.kubernetes.io/ and also contain the name of the feature source. Non-standard user-specific feature labels can be created with the local and custom feature sources.
An overview of the default feature labels:
{
"feature.node.kubernetes.io/cpu-<feature-name>": "true",
"feature.node.kubernetes.io/custom-<feature-name>": "true",
"feature.node.kubernetes.io/kernel-<feature name>": "<feature value>",
@@ -10,4 +10,4 @@
"feature.node.kubernetes.io/usb-<device label>.present": "<feature value>",
"feature.node.kubernetes.io/<file name>-<feature name>": "<feature value>"
}
-
Node annotations
NFD also annotates nodes it is running on:
Annotation Description [<instance>.]nfd.node.kubernetes.io/feature-labels Comma-separated list of node labels managed by NFD. NFD uses this internally so must not be edited by users. [<instance>.]nfd.node.kubernetes.io/feature-annotations Comma-separated list of node annotations managed by NFD. NFD uses this internally so must not be edited by users. [<instance>.]nfd.node.kubernetes.io/extended-resources Comma-separated list of node extended resources managed by NFD. NFD uses this internally so must not be edited by users. [<instance>.]nfd.node.kubernetes.io/taints Comma-separated list of node taints managed by NFD. NFD uses this internally so must not be edited by users.
NOTE: the -instance command line flag affects the annotation names
Unapplicable annotations are not created, i.e. for example nfd.node.kubernetes.io/extended-resources is only placed if some extended resources were created by NFD.
Custom resources
NFD takes use of some Kubernetes Custom Resources.
NodeFeatures is be used for representing node features and requesting node labels to be generated.
NFD-Master uses NodeFeatureRules for custom labeling of nodes.
NFD-Topology-Updater creates NodeResourceTopology objects that describe the hardware topology of node resources.
Node Feature Discovery master
\ No newline at end of file
+NFD also annotates nodes it is running on:
| Annotation | Description |
|---|---|
| [<instance>.]nfd.node.kubernetes.io/feature-labels | Comma-separated list of node labels managed by NFD. NFD uses this internally so must not be edited by users. |
| [<instance>.]nfd.node.kubernetes.io/feature-annotations | Comma-separated list of node annotations managed by NFD. NFD uses this internally so must not be edited by users. |
| [<instance>.]nfd.node.kubernetes.io/extended-resources | Comma-separated list of node extended resources managed by NFD. NFD uses this internally so must not be edited by users. |
| [<instance>.]nfd.node.kubernetes.io/taints | Comma-separated list of node taints managed by NFD. NFD uses this internally so must not be edited by users. |
NOTE: the
-instancecommand line flag affects the annotation names
Unapplicable annotations are not created, i.e. for example nfd.node.kubernetes.io/extended-resources is only placed if some extended resources were created by NFD.
NFD takes use of some Kubernetes Custom Resources.
NodeFeatures is be used for representing node features and requesting node labels to be generated.
NFD-Master uses NodeFeatureRules for custom labeling of nodes.
NFD-Topology-Updater creates NodeResourceTopology objects that describe the hardware topology of node resources.
Minimal steps to deploy latest released version of NFD in your cluster.
Deploy with kustomize – creates a new namespace, service and required RBAC rules and deploys nfd-master and nfd-worker daemons.
kubectl apply -k https://github.com/kubernetes-sigs/node-feature-discovery/deployment/overlays/default?ref=master
+ Quick start · Node Feature Discovery
Quick start
Minimal steps to deploy latest released version of NFD in your cluster.
Installation
Deploy with kustomize – creates a new namespace, service and required RBAC rules and deploys nfd-master and nfd-worker daemons.
kubectl apply -k https://github.com/kubernetes-sigs/node-feature-discovery/deployment/overlays/default?ref=master
Verify
Wait until NFD master and NFD worker are running.
$ kubectl -n node-feature-discovery get ds,deploy
NAME DESIRED CURRENT READY UP-TO-DATE AVAILABLE NODE SELECTOR AGE
daemonset.apps/nfd-worker 2 2 2 2 2 <none> 10s
@@ -40,4 +40,4 @@ daemonset.apps/nfd-topology-updater 2 2 2 2 2
NAME AGE
kind-control-plane 23s
kind-worker 23s
-
Node Feature Discovery master
\ No newline at end of file
+Feature gates are a set of key-value pairs that control the behavior of NFD. They are used to enable or disable certain features of NFD. The feature gates are set using the -feature-gates command line flag or featureGates value in the Helm chart. The following feature gates are available:
| Name | Default | Stage | Since | Until |
|---|---|---|---|---|
NodeFeatureAPI | true | Beta | V0.14 | |
DisableAutoPrefix | false | Alpha | V0.16 | |
NodeFeatureGroupAPI | false | Alpha | V0.16 |
The NodeFeatureAPI feature gate enables the Node Feature API. When enabled, NFD will register the Node Feature API with the Kubernetes API server. The Node Feature API is used to expose node-specific hardware and software features to the Kubernetes scheduler. The Node Feature API is a beta feature and is enabled by default.
The NodeFeatureGroupAPI feature gate enables the Node Feature Group API. When enabled, NFD will register the Node Feature Group API with the Kubernetes API server. The Node Feature Group API is used to create node groups based on hardware and software features. The Node Feature Group API is an alpha feature and is disabled by default.
The DisableAutoPrefix feature gate controls the automatic prefixing of names. When enabled nfd-master does not automatically add the default feature.node.kubernetes.io/ prefix to unprefixed labels, annotations and extended resources. Automatic prefixing is the default behavior in NFD v0.16 and earlier.
Note that enabling the feature gate effectively causes unprefixed names to be filtered out as NFD does not allow unprefixed names of labels, annotations or extended resources. For example, with the DisableAutoPrefix feature gate set to false, a NodeFeatureRule with
labels:
+ Feature Gates · Node Feature Discovery
Feature Gates
Feature gates are a set of key-value pairs that control the behavior of NFD. They are used to enable or disable certain features of NFD. The feature gates are set using the -feature-gates command line flag or featureGates value in the Helm chart. The following feature gates are available:
Name Default Stage Since Until NodeFeatureAPItrue Beta V0.14 DisableAutoPrefixfalse Alpha V0.16 NodeFeatureGroupAPIfalse Alpha V0.16
NodeFeatureAPI
The NodeFeatureAPI feature gate enables the Node Feature API. When enabled, NFD will register the Node Feature API with the Kubernetes API server. The Node Feature API is used to expose node-specific hardware and software features to the Kubernetes scheduler. The Node Feature API is a beta feature and is enabled by default.
NodeFeatureGroupAPI
The NodeFeatureGroupAPI feature gate enables the Node Feature Group API. When enabled, NFD will register the Node Feature Group API with the Kubernetes API server. The Node Feature Group API is used to create node groups based on hardware and software features. The Node Feature Group API is an alpha feature and is disabled by default.
DisableAutoPrefix
The DisableAutoPrefix feature gate controls the automatic prefixing of names. When enabled nfd-master does not automatically add the default feature.node.kubernetes.io/ prefix to unprefixed labels, annotations and extended resources. Automatic prefixing is the default behavior in NFD v0.16 and earlier.
Note that enabling the feature gate effectively causes unprefixed names to be filtered out as NFD does not allow unprefixed names of labels, annotations or extended resources. For example, with the DisableAutoPrefix feature gate set to false, a NodeFeatureRule with
labels:
foo: bar
-
will turn into feature.node.kubernetes.io/foo=bar node label. With DisableAutoPrefix set to true, no prefix is added and the label will be filtered out.
Note that taint keys are not affected by this feature gate.
Node Feature Discovery master
\ No newline at end of file
+will turn into feature.node.kubernetes.io/foo=bar node label. With DisableAutoPrefix set to true, no prefix is added and the label will be filtered out.
Note that taint keys are not affected by this feature gate.
To quickly view available command line flags execute nfd-gc -help. In a docker container:
docker run gcr.io/k8s-staging-nfd/node-feature-discovery:master \
+ Garbage Collector Cmdline Reference · Node Feature Discovery
NFD-GC Commandline Flags
Table of Contents
To quickly view available command line flags execute nfd-gc -help. In a docker container:
docker run gcr.io/k8s-staging-nfd/node-feature-discovery:master \
nfd-gc -help
-h, -help
Print usage and exit.
-version
Print version and exit.
-gc-interval
The -gc-interval specifies the interval between periodic garbage collector runs.
Default: 1h
Example:
nfd-gc -gc-interval=1h
-
Node Feature Discovery master
\ No newline at end of file
+Command line and configuration reference.
Command line and configuration reference.
To quickly view available command line flags execute nfd-master -help. In a docker container:
docker run gcr.io/k8s-staging-nfd/node-feature-discovery:master nfd-master -help
+ Master cmdline reference · Node Feature Discovery
Commandline flags of nfd-master
Table of contents
- -h, -help
- -version
- -feature-gates
- -prune
- -port
- -metrics
- -instance
- -ca-file
- -cert-file
- -key-file
- -verify-node-name
- -enable-nodefeature-api
- -enable-leader-election
- -enable-taints
- -no-publish
- -crd-controller
- -featurerules-controller
- -label-whitelist
- -extra-label-ns
- -deny-label-ns
- -resource-labels
- -config
- -options
- -nfd-api-parallelism
- Logging
- -resync-period
To quickly view available command line flags execute nfd-master -help. In a docker container:
docker run gcr.io/k8s-staging-nfd/node-feature-discovery:master nfd-master -help
-h, -help
Print usage and exit.
-version
Print version and exit.
-feature-gates
The -feature-gates flag is used to enable or disable non GA features. The list of available feature gates can be found in the feature gates documentation.
Example:
nfd-master -feature-gates NodeFeatureAPI=false
-prune
The -prune flag is a sub-command like option for cleaning up the cluster. It causes nfd-master to remove all NFD related labels, annotations and extended resources from all Node objects of the cluster and exit.
-port
The -port flag specifies the TCP port that nfd-master listens for incoming requests.
Default: 8080
Example:
nfd-master -port=443
-metrics
The -metrics flag specifies the port on which to expose Prometheus metrics. Setting this to 0 disables the metrics server on nfd-master.
Default: 8081
Example:
nfd-master -metrics=12345
@@ -21,4 +21,4 @@
-options
The -options flag may be used to specify and override configuration file options directly from the command line. The required format is the same as in the config file i.e. JSON or YAML. Configuration options specified via this flag will override those from the configuration file:
Default: empty
Example:
nfd-master -options='{"noPublish": true}'
-nfd-api-parallelism
The -nfd-api-parallelism flag can be used to specify the maximum number of concurrent node updates.
Does not have effect if the NodeFeatureAPI feature gate is disabled.
Default: 10
Example:
nfd-master -nfd-api-parallelism=1
Logging
The following logging-related flags are inherited from the klog package.
-add_dir_header
If true, adds the file directory to the header of the log messages.
Default: false
-alsologtostderr
Log to standard error as well as files.
Default: false
-log_backtrace_at
When logging hits line file:N, emit a stack trace.
Default: empty
-log_dir
If non-empty, write log files in this directory.
Default: empty
-log_file
If non-empty, use this log file.
Default: empty
-log_file_max_size
Defines the maximum size a log file can grow to. Unit is megabytes. If the value is 0, the maximum file size is unlimited.
Default: 1800
-logtostderr
Log to standard error instead of files
Default: true
-skip_headers
If true, avoid header prefixes in the log messages.
Default: false
-skip_log_headers
If true, avoid headers when opening log files.
Default: false
-stderrthreshold
Logs at or above this threshold go to stderr.
Default: 2
-v
Number for the log level verbosity.
Default: 0
-vmodule
Comma-separated list of pattern=N settings for file-filtered logging.
Default: empty
-resync-period
The -resync-period flag specifies the NFD API controller resync period. The resync means nfd-master replaying all NodeFeature and NodeFeatureRule objects, thus effectively re-syncing all nodes in the cluster (i.e. ensuring labels, annotations, extended resources and taints are in place).
Does not have effect if the NodeFeatureAPI feature gate is disabled.
Default: 1 hour.
Example:
nfd-master -resync-period=2h
-
Node Feature Discovery master
\ No newline at end of file
+See the sample configuration file for a full example configuration.
noPublish option disables updates to the Node objects in the Kubernetes API server, making a "dry-run" flag for nfd-master. No Labels, Annotations, Taints or ExtendedResources of nodes are updated.
Default: false
Example:
noPublish: true
+ Master config reference · Node Feature Discovery
Configuration file reference of nfd-master
Table of contents
- noPublish
- extraLabelNs
- denyLabelNs
- autoDefaultNs
- resourceLabels
- enableTaints
- labelWhiteList
- resyncPeriod
- leaderElection
- nfdApiParallelism
- klog
See the sample configuration file for a full example configuration.
noPublish
noPublish option disables updates to the Node objects in the Kubernetes API server, making a "dry-run" flag for nfd-master. No Labels, Annotations, Taints or ExtendedResources of nodes are updated.
Default: false
Example:
noPublish: true
extraLabelNs
extraLabelNs specifies a list of allowed feature label namespaces. This option can be used to allow other vendor or application specific namespaces for custom labels from the local and custom feature sources, even though these labels were denied using the denyLabelNs parameter.
The same namespace control and this option applies to Extended Resources (created with resourceLabels), too.
Default: empty
Example:
extraLabelNs: ["added.ns.io","added.kubernets.io"]
denyLabelNs
denyLabelNs specifies a list of excluded label namespaces. By default, nfd-master allows creating labels in all namespaces, excluding kubernetes.io namespace and its sub-namespaces (i.e. *.kubernetes.io). However, you should note that kubernetes.io and its sub-namespaces are always denied. This option can be used to exclude some vendors or application specific namespaces.
Default: empty
Example:
denyLabelNs: ["denied.ns.io","denied.kubernetes.io"]
autoDefaultNs
DEPRECATED: Will be removed in NFD v0.17. Use the DisableAutoPrefix feature gate instead.
The autoDefaultNs option controls the automatic prefixing of names. When set to true (the default in NFD version master) nfd-master automatically adds the default feature.node.kubernetes.io/ prefix to unprefixed labels, annotations and extended resources - this is also the default behavior in NFD v0.15 and earlier. When the option is set to false, no prefix will be prepended to unprefixed names, effectively causing them to be filtered out (as NFD does not allow unprefixed names of labels, annotations or extended resources). The default will be changed to false in a future release.
For example, with the autoDefaultNs set to true, a NodeFeatureRule with
labels:
@@ -15,4 +15,4 @@
leaderElection.retryPeriod
leaderElection.retryPeriod is the duration the LeaderElector clients should wait between tries of actions.
It has to be greater than 0.
Default: 2 seconds.
Example:
leaderElection:
retryPeriod: 2s
nfdApiParallelism
The nfdApiParallelism option can be used to specify the maximum number of concurrent node updates.
Does not have effect if the NodeFeatureAPI feature gate is disabled.
Default: 10
Example:
nfdApiParallelism: 1
-
klog
The following options specify the logger configuration. Most of which can be dynamically adjusted at run-time.
NOTE: The logger options can also be specified via command line flags which take precedence over any corresponding config file options.
klog.addDirHeader
If true, adds the file directory to the header of the log messages.
Default: false
Run-time configurable: yes
klog.alsologtostderr
Log to standard error as well as files.
Default: false
Run-time configurable: yes
klog.logBacktraceAt
When logging hits line file:N, emit a stack trace.
Default: empty
Run-time configurable: yes
klog.logDir
If non-empty, write log files in this directory.
Default: empty
Run-time configurable: no
klog.logFile
If non-empty, use this log file.
Default: empty
Run-time configurable: no
klog.logFileMaxSize
Defines the maximum size a log file can grow to. Unit is megabytes. If the value is 0, the maximum file size is unlimited.
Default: 1800
Run-time configurable: no
klog.logtostderr
Log to standard error instead of files
Default: true
Run-time configurable: yes
klog.skipHeaders
If true, avoid header prefixes in the log messages.
Default: false
Run-time configurable: yes
klog.skipLogHeaders
If true, avoid headers when opening log files.
Default: false
Run-time configurable: no
klog.stderrthreshold
Logs at or above this threshold go to stderr (default 2)
Run-time configurable: yes
klog.v
Number for the log level verbosity.
Default: 0
Run-time configurable: yes
klog.vmodule
Comma-separated list of pattern=N settings for file-filtered logging.
Default: empty
Run-time configurable: yes
Node Feature Discovery master
\ No newline at end of file
+The following options specify the logger configuration. Most of which can be dynamically adjusted at run-time.
NOTE: The logger options can also be specified via command line flags which take precedence over any corresponding config file options.
If true, adds the file directory to the header of the log messages.
Default: false
Run-time configurable: yes
Log to standard error as well as files.
Default: false
Run-time configurable: yes
When logging hits line file:N, emit a stack trace.
Default: empty
Run-time configurable: yes
If non-empty, write log files in this directory.
Default: empty
Run-time configurable: no
If non-empty, use this log file.
Default: empty
Run-time configurable: no
Defines the maximum size a log file can grow to. Unit is megabytes. If the value is 0, the maximum file size is unlimited.
Default: 1800
Run-time configurable: no
Log to standard error instead of files
Default: true
Run-time configurable: yes
If true, avoid header prefixes in the log messages.
Default: false
Run-time configurable: yes
If true, avoid headers when opening log files.
Default: false
Run-time configurable: no
Logs at or above this threshold go to stderr (default 2)
Run-time configurable: yes
Number for the log level verbosity.
Default: 0
Run-time configurable: yes
Comma-separated list of pattern=N settings for file-filtered logging.
Default: empty
Run-time configurable: yes
To quickly view available command line flags execute kubectl nfd -help.
Print usage and exit.
Validate a NodeFeatureRule file.
The --nodefeature-file flag specifies the path to the NodeFeatureRule file to validate.
Test a NodeFeatureRule file against a node without applying it.
The --kubeconfig flag specifies the path to the kubeconfig file to use for CLI requests.
The --namespace flag specifies the namespace to use for CLI requests. Default: default.
The --nodename flag specifies the name of the node to test the NodeFeatureRule against.
The --nodefeaturerule-file flag specifies the path to the NodeFeatureRule file to test.
Process a NodeFeatureRule file against a NodeFeature file.
The --nodefeaturerule-file flag specifies the path to the NodeFeatureRule file to test.
The --nodefeature-file flag specifies the path to the NodeFeature file to test.
To quickly view available command line flags execute kubectl nfd -help.
Print usage and exit.
Validate a NodeFeatureRule file.
The --nodefeature-file flag specifies the path to the NodeFeatureRule file to validate.
Test a NodeFeatureRule file against a node without applying it.
The --kubeconfig flag specifies the path to the kubeconfig file to use for CLI requests.
The --namespace flag specifies the namespace to use for CLI requests. Default: default.
The --nodename flag specifies the name of the node to test the NodeFeatureRule against.
The --nodefeaturerule-file flag specifies the path to the NodeFeatureRule file to test.
Process a NodeFeatureRule file against a NodeFeature file.
The --nodefeaturerule-file flag specifies the path to the NodeFeatureRule file to test.
The --nodefeature-file flag specifies the path to the NodeFeature file to test.
To quickly view available command line flags execute nfd-topology-updater -help. In a docker container:
docker run gcr.io/k8s-staging-nfd/node-feature-discovery:master \
+ Topology Updater Cmdline Reference · Node Feature Discovery
NFD-Topology-Updater Commandline Flags
Table of Contents
- -h, -help
- -version
- -config
- -no-publish
- -oneshot
- -metrics
- -sleep-interval
- -watch-namespace
- -kubelet-config-uri
- -api-auth-token-file
- -podresources-socket
- -pods-fingerprint
- -kubelet-state-dir
To quickly view available command line flags execute nfd-topology-updater -help. In a docker container:
docker run gcr.io/k8s-staging-nfd/node-feature-discovery:master \
nfd-topology-updater -help
-h, -help
Print usage and exit.
-version
Print version and exit.
-config
The -config flag specifies the path of the nfd-topology-updater configuration file to use.
Default: /etc/kubernetes/node-feature-discovery/nfd-topology-updater.conf
Example:
nfd-topology-updater -config=/opt/nfd/nfd-topology-updater.conf
-no-publish
The -no-publish flag makes for a "dry-run" flag for nfd-topology-updater. NFD-Topology-Updater runs resource hardware topology detection normally, but NodeResourceTopology objects are not created or updated.
Default: false
Example:
nfd-topology-updater -no-publish
@@ -11,4 +11,4 @@ nfd-topology-updater -help
-podresources-socket
The -podresources-socket specifies the path to the Unix socket where kubelet exports a gRPC service to enable discovery of in-use CPUs and devices, and to provide metadata for them.
Default: /host-var/lib/kubelet/pod-resources/kubelet.sock
Example:
nfd-topology-updater -podresources-socket=/var/lib/kubelet/pod-resources/kubelet.sock
-pods-fingerprint
Enables compute and report the pod set fingerprint in the NRT. A pod fingerprint is a compact representation of the "node state" regarding resources.
Default: true
Example:
nfd-topology-updater -pods-fingerprint=false
-kubelet-state-dir
The -kubelet-state-dir specifies the path to the Kubelet state directory, where state and checkpoint files are stored. The files are mount as read-only and cannot be change by the updater. Enabled by default. Passing an empty string will disable the watching.
Default: /host-var/lib/kubelet
Example:
nfd-topology-updater -kubelet-state-dir=/var/lib/kubelet
-
Node Feature Discovery master
\ No newline at end of file
+See the sample configuration file for a full example configuration.
The excludeList specifies a key-value map of allocated resources that should not be examined by the topology-updater agent per node. Each key is a node name with a value as a list of resources that should not be examined by the agent for that specific node.
Default: empty
Example:
excludeList:
+ Topology-Updater config reference · Node Feature Discovery
Configuration file reference of nfd-topology-updater
Table of contents
See the sample configuration file for a full example configuration.
excludeList
The excludeList specifies a key-value map of allocated resources that should not be examined by the topology-updater agent per node. Each key is a node name with a value as a list of resources that should not be examined by the agent for that specific node.
Default: empty
Example:
excludeList:
nodeA: [hugepages-2Mi]
nodeB: [memory]
nodeC: [cpu, hugepages-2Mi]
excludeList.*
excludeList.* is a special value that use to specify all nodes. A resource that would be listed under this key, would be excluded from all nodes.
Default: empty
Example:
excludeList:
'*': [hugepages-2Mi]
-
Node Feature Discovery master
\ No newline at end of file
+Node Feature Discovery follows semantic versioning where the version number consists of three components, i.e. MAJOR.MINOR.PATCH.
The most recent two minor releases (or release branches) of Node Feature Discovery are supported. That is, with X being the latest release, X and X-1 are supported and X-1 reaches end-of-life when X+1 is released.
Built-in feature labels and features are supported for 2 releases after being deprecated, at minimum. That is, if a feature label is deprecated in version X, it will be supported in X+1 and X+2 and may be dropped in X+3.
Command-line flags and configuration file options are supported for 1 more release after being deprecated, at minimum. That is, if option/flag is deprecated in version X, it will be supported in X+1 and may be removed in X+2.
The same policy (support for 1 release after deprecation) also applies to Helm chart parameters.
Node Feature Discovery is compatible with Kubernetes v1.24 and later.
Node Feature Discovery follows semantic versioning where the version number consists of three components, i.e. MAJOR.MINOR.PATCH.
The most recent two minor releases (or release branches) of Node Feature Discovery are supported. That is, with X being the latest release, X and X-1 are supported and X-1 reaches end-of-life when X+1 is released.
Built-in feature labels and features are supported for 2 releases after being deprecated, at minimum. That is, if a feature label is deprecated in version X, it will be supported in X+1 and X+2 and may be dropped in X+3.
Command-line flags and configuration file options are supported for 1 more release after being deprecated, at minimum. That is, if option/flag is deprecated in version X, it will be supported in X+1 and may be removed in X+2.
The same policy (support for 1 release after deprecation) also applies to Helm chart parameters.
Node Feature Discovery is compatible with Kubernetes v1.24 and later.
To quickly view available command line flags execute nfd-worker -help. In a docker container:
docker run gcr.io/k8s-staging-nfd/node-feature-discovery:master nfd-worker -help
+ Worker cmdline reference · Node Feature Discovery
Commandline flags of nfd-worker
Table of contents
- -h, -help
- -version
- -feature-gates
- -config
- -options
- -server
- -ca-file
- -cert-file
- -key-file
- -kubeconfig
- -server-name-override
- -feature-sources
- -label-sources
- -metrics
- -no-publish
- -oneshot
- Logging
To quickly view available command line flags execute nfd-worker -help. In a docker container:
docker run gcr.io/k8s-staging-nfd/node-feature-discovery:master nfd-worker -help
-h, -help
Print usage and exit.
-version
Print version and exit.
-feature-gates
The -feature-gates flag is used to enable or disable non GA features. The list of available feature gates can be found in the feature gates documentation.
Example:
nfd-master -feature-gates NodeFeatureAPI=false
-config
The -config flag specifies the path of the nfd-worker configuration file to use.
Default: /etc/kubernetes/node-feature-discovery/nfd-worker.conf
Example:
nfd-worker -config=/opt/nfd/worker.conf
-options
The -options flag may be used to specify and override configuration file options directly from the command line. The required format is the same as in the config file i.e. JSON or YAML. Configuration options specified via this flag will override those from the configuration file:
Default: empty
Example:
nfd-worker -options='{"sources":{"cpu":{"cpuid":{"attributeWhitelist":["AVX","AVX2"]}}}}'
@@ -13,4 +13,4 @@
-metrics
The -metrics flag specifies the port on which to expose Prometheus metrics. Setting this to 0 disables the metrics server on nfd-worker.
Default: 8081
Example:
nfd-worker -metrics=12345
-no-publish
The -no-publish flag disables all communication with the nfd-master and the Kubernetes API server. It is effectively a "dry-run" flag for nfd-worker. NFD-Worker runs feature detection normally, but no labeling requests are sent to nfd-master and no NodeFeature objects are created or updated in the API server.
NOTE: This flag takes precedence over the core.noPublish configuration file option.
Default: false
Example:
nfd-worker -no-publish
-oneshot
The -oneshot flag causes nfd-worker to exit after one pass of feature detection.
Default: false
Example:
nfd-worker -oneshot -no-publish
-
Logging
The following logging-related flags are inherited from the klog package.
NOTE: The logger setup can also be specified via the core.klog configuration file options. However, the command line flags take precedence over any corresponding config file options specified.
-add_dir_header
If true, adds the file directory to the header of the log messages.
Default: false
-alsologtostderr
Log to standard error as well as files.
Default: false
-log_backtrace_at
When logging hits line file:N, emit a stack trace.
Default: empty
-log_dir
If non-empty, write log files in this directory.
Default: empty
-log_file
If non-empty, use this log file.
Default: empty
-log_file_max_size
Defines the maximum size a log file can grow to. Unit is megabytes. If the value is 0, the maximum file size is unlimited.
Default: 1800
-logtostderr
Log to standard error instead of files
Default: true
-skip_headers
If true, avoid header prefixes in the log messages.
Default: false
-skip_log_headers
If true, avoid headers when opening log files.
Default: false
-stderrthreshold
Logs at or above this threshold go to stderr.
Default: 2
-v
Number for the log level verbosity.
Default: 0
-vmodule
Comma-separated list of pattern=N settings for file-filtered logging.
Default: empty
Node Feature Discovery master
\ No newline at end of file
+The following logging-related flags are inherited from the klog package.
NOTE: The logger setup can also be specified via the
core.klogconfiguration file options. However, the command line flags take precedence over any corresponding config file options specified.
If true, adds the file directory to the header of the log messages.
Default: false
Log to standard error as well as files.
Default: false
When logging hits line file:N, emit a stack trace.
Default: empty
If non-empty, write log files in this directory.
Default: empty
If non-empty, use this log file.
Default: empty
Defines the maximum size a log file can grow to. Unit is megabytes. If the value is 0, the maximum file size is unlimited.
Default: 1800
Log to standard error instead of files
Default: true
If true, avoid header prefixes in the log messages.
Default: false
If true, avoid headers when opening log files.
Default: false
Logs at or above this threshold go to stderr.
Default: 2
Number for the log level verbosity.
Default: 0
Comma-separated list of pattern=N settings for file-filtered logging.
Default: empty
See the sample configuration file for a full example configuration.
The core section contains common configuration settings that are not specific to any particular feature source.
core.sleepInterval specifies the interval between consecutive passes of feature (re-)detection, and thus also the interval between node re-labeling. A non-positive value implies infinite sleep interval, i.e. no re-detection or re-labeling is done.
Default: 60s
Example:
core:
+ Worker config reference · Node Feature Discovery
Configuration file reference of nfd-worker
Table of contents
See the sample configuration file for a full example configuration.
core
The core section contains common configuration settings that are not specific to any particular feature source.
core.sleepInterval
core.sleepInterval specifies the interval between consecutive passes of feature (re-)detection, and thus also the interval between node re-labeling. A non-positive value implies infinite sleep interval, i.e. no re-detection or re-labeling is done.
Default: 60s
Example:
core:
sleepInterval: 60s
core.featureSources
core.featureSources specifies the list of enabled feature sources. A special value all enables all sources. Prefixing a source name with - indicates that the source will be disabled instead - this is only meaningful when used in conjunction with all. This option allows completely disabling the feature detection so that neither standard feature labels are generated nor the raw feature data is available for custom rule processing.
Default: [all]
Example:
core:
# Enable all but cpu and local sources
@@ -68,4 +68,4 @@
matchExpressions:
class: {op: In, value: ["0200"]}
vendor: {op: In, value: ["8086"]}
-
Node Feature Discovery master
\ No newline at end of file
+NFD uses some Kubernetes custom resources.
NodeFeature is an NFD-specific custom resource for communicating node features and node labeling requests. The nfd-master pod watches for NodeFeature objects, labels nodes as specified and uses the listed features as input when evaluating NodeFeatureRules. NodeFeature objects can be used for implementing 3rd party extensions (see customization guide for more details).
apiVersion: nfd.k8s-sigs.io/v1alpha1
+ CRDs · Node Feature Discovery
Custom Resources
Table of contents
NFD uses some Kubernetes custom resources.
NodeFeature
NodeFeature is an NFD-specific custom resource for communicating node features and node labeling requests. The nfd-master pod watches for NodeFeature objects, labels nodes as specified and uses the listed features as input when evaluating NodeFeatureRules. NodeFeature objects can be used for implementing 3rd party extensions (see customization guide for more details).
apiVersion: nfd.k8s-sigs.io/v1alpha1
kind: NodeFeature
metadata:
labels:
@@ -88,4 +88,4 @@
capacity: 3
allocatable: 3
available: 3
-
The NodeResourceTopology objects created by NFD can be used to gain insight into the allocatable resources along with the granularity of those resources at a per-zone level (represented by node-0 and node-1 in the above example) or can be used by an external entity (e.g. topology-aware scheduler plugin) to take an action based on the gathered information.
Node Feature Discovery master
\ No newline at end of file
+The NodeResourceTopology objects created by NFD can be used to gain insight into the allocatable resources along with the granularity of those resources at a per-zone level (represented by node-0 and node-1 in the above example) or can be used by an external entity (e.g. topology-aware scheduler plugin) to take an action based on the gathered information.
NFD provides multiple extension points for vendor and application specific labeling:
NodeFeature objects can be used to communicate "raw" node features and node labeling requests to nfd-master.NodeFeatureRule objects provide a way to deploy custom labeling rules via the Kubernetes API.local feature source of nfd-worker creates labels by reading text files and executing hooks.custom feature source of nfd-worker creates labels based on user-specified rules.NodeFeature objects provide a way for 3rd party extensions to advertise custom features, both as "raw" features that serve as input to NodeFeatureRule objects and as feature labels directly.
Note that RBAC rules must be created for each extension for them to be able to create and manipulate NodeFeature objects in their namespace.
The NodeFeature CRD API can be disabled with the NodeFeatureAPI feature gate (-feature-gates NodeFeatureAPI=false on command line). The -feature-gates command line flag must be specified for both nfd-master and nfd-worker as it will enable the gRPC communication between them. Note that the gRPC API is DEPRECATED and will be removed in a future release, at which point the NodeFeature API cannot be disabled.
Consider the following referential example:
apiVersion: nfd.k8s-sigs.io/v1alpha1
+ Customization guide · Node Feature Discovery
Customization guide
Table of contents
- Overview
- NodeFeature custom resource
- NodeFeatureRule custom resource
- NodeFeatureGroup custom resource
- Local feature source
- Custom feature source
- Node labels
- Feature rule format
Overview
NFD provides multiple extension points for vendor and application specific labeling:
NodeFeature objects can be used to communicate "raw" node features and node labeling requests to nfd-master.NodeFeatureRule objects provide a way to deploy custom labeling rules via the Kubernetes API.local feature source of nfd-worker creates labels by reading text files and executing hooks.custom feature source of nfd-worker creates labels based on user-specified rules.
NodeFeature custom resource
NodeFeature objects provide a way for 3rd party extensions to advertise custom features, both as "raw" features that serve as input to NodeFeatureRule objects and as feature labels directly.
Note that RBAC rules must be created for each extension for them to be able to create and manipulate NodeFeature objects in their namespace.
The NodeFeature CRD API can be disabled with the NodeFeatureAPI feature gate (-feature-gates NodeFeatureAPI=false on command line). The -feature-gates command line flag must be specified for both nfd-master and nfd-worker as it will enable the gRPC communication between them. Note that the gRPC API is DEPRECATED and will be removed in a future release, at which point the NodeFeature API cannot be disabled.
A NodeFeature example
Consider the following referential example:
apiVersion: nfd.k8s-sigs.io/v1alpha1
kind: NodeFeature
metadata:
labels:
@@ -366,4 +366,4 @@ vendor.io/my-feature=value
- pci.device:
vendor: "0fff"
device: "abcd"
-
Node Feature Discovery master
\ No newline at end of file
+This page contains usage examples and demos.
A demo on the benefits of using node feature discovery can be found in the source code repository under demo/.
This page contains usage examples and demos.
A demo on the benefits of using node feature discovery can be found in the source code repository under demo/.
Features are advertised as labels in the Kubernetes Node object.
Label creation in nfd-worker is performed by a set of separate modules called label sources. The core.labelSources configuration option (or -label-sources flag) of nfd-worker controls which sources to enable for label generation.
All built-in labels use the feature.node.kubernetes.io label namespace and have the following format.
feature.node.kubernetes.io/<feature> = <value>
-NOTE: Consecutive runs of nfd-worker will update the labels on a given node. If features are not discovered on a consecutive run, the corresponding label will be removed. This includes any restrictions placed on the consecutive run, such as restricting discovered features with the
-label-whitelistflag of nfd-master orcore.labelWhiteListoption of nfd-worker.
| Feature name | Value | Description |
|---|---|---|
cpu-cpuid.<cpuid-flag> | true | CPU capability is supported. NOTE: the capability might be supported but not enabled. |
cpu-cpuid.<cpuid-attribute> | string | CPU attribute value |
cpu-hardware_multithreading | true | Hardware multithreading, such as Intel HTT, enabled (number of logical CPUs is greater than physical CPUs) |
cpu-coprocessor.nx_gzip | true | Nest Accelerator for GZIP is supported(Power). |
cpu-power.sst_bf.enabled | true | Intel SST-BF (Intel Speed Select Technology - Base frequency) enabled |
cpu-pstate.status | string | The status of the Intel pstate driver when in use and enabled, either ‘active' or ‘passive'. |
cpu-pstate.turbo | bool | Set to ‘true' if turbo frequencies are enabled in Intel pstate driver, set to ‘false' if they have been disabled. |
cpu-pstate.scaling_governor | string | The value of the Intel pstate scaling_governor when in use, either ‘powersave' or ‘performance'. |
cpu-cstate.enabled | bool | Set to ‘true' if cstates are set in the intel_idle driver, otherwise set to ‘false'. Unset if intel_idle cpuidle driver is not active. |
cpu-security.sgx.enabled | true | Set to ‘true' if Intel SGX is enabled in BIOS (based on a non-zero sum value of SGX EPC section sizes). |
cpu-security.se.enabled | true | Set to ‘true' if IBM Secure Execution for Linux (IBM Z & LinuxONE) is available and enabled (requires /sys/firmware/uv/prot_virt_host facility) |
cpu-security.tdx.enabled | true | Set to ‘true' if Intel TDX is available on the host and has been enabled (requires /sys/module/kvm_intel/parameters/tdx). |
cpu-security.tdx.protected | true | Set to ‘true' if Intel TDX was used to start the guest node, based on the existence of the "TDX_GUEST" information as part of cpuid features. |
cpu-security.sev.enabled | true | Set to ‘true' if ADM SEV is available on the host and has been enabled (requires /sys/module/kvm_amd/parameters/sev). |
cpu-security.sev.es.enabled | true | Set to ‘true' if ADM SEV-ES is available on the host and has been enabled (requires /sys/module/kvm_amd/parameters/sev_es). |
cpu-security.sev.snp.enabled | true | Set to ‘true' if ADM SEV-SNP is available on the host and has been enabled (requires /sys/module/kvm_amd/parameters/sev_snp). |
cpu-model.vendor_id | string | Comparable CPU vendor ID. |
cpu-model.family | int | CPU family. |
cpu-model.id | int | CPU model number. |
The CPU label source is configurable, see worker configuration and sources.cpu configuration options for details.
| Flag | Description |
|---|---|
| ADX | Multi-Precision Add-Carry Instruction Extensions (ADX) |
| AESNI | Advanced Encryption Standard (AES) New Instructions (AES-NI) |
| APX_F | Intel Advanced Performance Extensions (APX) |
| AVX10 | Intel Advanced Vector Extensions 10 (AVX10) |
| AVX10_256, AVX10_512 | Intel AVX10 256-bit and 512-bit vector support |
| AVX | Advanced Vector Extensions (AVX) |
| AVX2 | Advanced Vector Extensions 2 (AVX2) |
| AVXIFMA | AVX-IFMA instructions |
| AVXVNNI | AVX (VEX encoded) VNNI neural network instructions |
| AMXBF16 | Advanced Matrix Extension, tile multiplication operations on BFLOAT16 numbers |
| AMXINT8 | Advanced Matrix Extension, tile multiplication operations on 8-bit integers |
| AMXFP16 | Advanced Matrix Extension, tile multiplication operations on FP16 numbers |
| AMXTILE | Advanced Matrix Extension, base tile architecture support |
| AVX512BF16 | AVX-512 BFLOAT16 instructions |
| AVX512BITALG | AVX-512 bit Algorithms |
| AVX512BW | AVX-512 byte and word Instructions |
| AVX512CD | AVX-512 conflict detection instructions |
| AVX512DQ | AVX-512 doubleword and quadword instructions |
| AVX512ER | AVX-512 exponential and reciprocal instructions |
| AVX512F | AVX-512 foundation |
| AVX512FP16 | AVX-512 FP16 instructions |
| AVX512IFMA | AVX-512 integer fused multiply-add instructions |
| AVX512PF | AVX-512 prefetch instructions |
| AVX512VBMI | AVX-512 vector bit manipulation instructions |
| AVX512VBMI2 | AVX-512 vector bit manipulation instructions, version 2 |
| AVX512VL | AVX-512 vector length extensions |
| AVX512VNNI | AVX-512 vector neural network instructions |
| AVX512VP2INTERSECT | AVX-512 intersect for D/Q |
| AVX512VPOPCNTDQ | AVX-512 vector population count doubleword and quadword |
| AVXNECONVERT | AVX-NE-CONVERT instructions |
| AVXVNNIINT8 | AVX-VNNI-INT8 instructions |
| AVXVNNIINT16 | AVX-VNNI-INT16 instructions |
| CMPCCXADD | CMPCCXADD instructions |
| ENQCMD | Enqueue Command |
| GFNI | Galois Field New Instructions |
| HYPERVISOR | Running under hypervisor |
| MSRLIST | Read/Write List of Model Specific Registers |
| PREFETCHI | PREFETCHIT0/1 instructions |
| VAES | AVX-512 vector AES instructions |
| VPCLMULQDQ | Carry-less multiplication quadword |
| WRMSRNS | Non-Serializing Write to Model Specific Register |
By default, the following CPUID flags have been blacklisted: AVX10 (use AVX10_VERSION instead), BMI1, BMI2, CLMUL, CMOV, CX16, ERMS, F16C, HTT, LZCNT, MMX, MMXEXT, NX, POPCNT, RDRAND, RDSEED, RDTSCP, SGX, SSE, SSE2, SSE3, SSE4, SSE42, SSSE3 and TDX_GUEST. See sources.cpu configuration options to change the behavior.
See the full list in github.com/klauspost/cpuid.
| Attribute | Description |
|---|---|
| AVX10_VERSION | AVX10 vector ISA version (if supported) |
| Flag | Description |
|---|---|
| IDIVA | Integer divide instructions available in ARM mode |
| IDIVT | Integer divide instructions available in Thumb mode |
| THUMB | Thumb instructions |
| FASTMUL | Fast multiplication |
| VFP | Vector floating point instruction extension (VFP) |
| VFPv3 | Vector floating point extension v3 |
| VFPv4 | Vector floating point extension v4 |
| VFPD32 | VFP with 32 D-registers |
| HALF | Half-word loads and stores |
| EDSP | DSP extensions |
| NEON | NEON SIMD instructions |
| LPAE | Large Physical Address Extensions |
| Flag | Description |
|---|---|
| AES | Announcing the Advanced Encryption Standard |
| EVSTRM | Event Stream Frequency Features |
| FPHP | Half Precision(16bit) Floating Point Data Processing Instructions |
| ASIMDHP | Half Precision(16bit) Asimd Data Processing Instructions |
| ATOMICS | Atomic Instructions to the A64 |
| ASIMRDM | Support for Rounding Double Multiply Add/Subtract |
| PMULL | Optional Cryptographic and CRC32 Instructions |
| JSCVT | Perform Conversion to Match Javascript |
| DCPOP | Persistent Memory Support |
| Feature | Value | Description |
|---|---|---|
kernel-config.<option> | true | Kernel config option is enabled (set ‘y' or ‘m'). Default options are NO_HZ, NO_HZ_IDLE, NO_HZ_FULL and PREEMPT |
kernel-selinux.enabled | true | Selinux is enabled on the node |
kernel-version.full | string | Full kernel version as reported by /proc/sys/kernel/osrelease (e.g. ‘4.5.6-7-g123abcde') |
kernel-version.major | string | First component of the kernel version (e.g. ‘4') |
kernel-version.minor | string | Second component of the kernel version (e.g. ‘5') |
kernel-version.revision | string | Third component of the kernel version (e.g. ‘6') |
The kernel label source is configurable, see worker configuration and sources.kernel configuration options for details.
| Feature | Value | Description |
|---|---|---|
memory-numa | true | Multiple memory nodes i.e. NUMA architecture detected |
memory-nv.present | true | NVDIMM device(s) are present |
memory-nv.dax | true | NVDIMM region(s) configured in DAX mode are present |
memory-swap.enabled | true | Swap is enabled on the node |
| Feature | Value | Description |
|---|---|---|
network-sriov.capable | true | Single Root Input/Output Virtualization (SR-IOV) enabled Network Interface Card(s) present |
network-sriov.configured | true | SR-IOV virtual functions have been configured |
| Feature | Value | Description |
|---|---|---|
pci-<device label>.present | true | PCI device is detected |
pci-<device label>.sriov.capable | true | Single Root Input/Output Virtualization (SR-IOV) enabled PCI device present |
<device label> is format is configurable and set to <class>_<vendor> by default. For more more details about configuration of the pci labels, see sources.pci options and worker configuration instructions.
| Feature | Value | Description |
|---|---|---|
usb-<device label>.present | true | USB device is detected |
<device label> is format is configurable and set to <class>_<vendor>_<device> by default. For more more details about configuration of the usb labels, see sources.usb options and worker configuration instructions.
| Feature | Value | Description |
|---|---|---|
storage-nonrotationaldisk | true | Non-rotational disk, like SSD, is present in the node |
| Feature | Value | Description |
|---|---|---|
system-os_release.ID | string | Operating system identifier |
system-os_release.VERSION_ID | string | Operating system version identifier (e.g. ‘6.7') |
system-os_release.VERSION_ID.major | string | First component of the OS version id (e.g. ‘6') |
system-os_release.VERSION_ID.minor | string | Second component of the OS version id (e.g. ‘7') |
The custom label source is designed for creating user defined labels. However, it has a few statically defined built-in labels:
| Feature | Value | Description |
|---|---|---|
custom-rdma.capable | true | The node has an RDMA capable Network adapter |
custom-rdma.enabled | true | The node has the needed RDMA modules loaded to run RDMA traffic |
NFD has many extension points for creating vendor and application specific labels. See the customization guide for detailed documentation.
NFD is able to create extended resources, see the NodeFeatureRule CRD and its extendedResources field for more details.
Note that NFD is not a replacement for the usage of device plugins.
An example use-case for extended resources could be based on custom feature (created e.g. with feature files that exposes the node SGX EPC memory section size. This value will then be turned into an extended resource of the node, allowing PODs to request that resource and the Kubernetes scheduler to schedule such PODs to only those nodes which have a sufficient capacity of said resource left.
Features are advertised as labels in the Kubernetes Node object.
Label creation in nfd-worker is performed by a set of separate modules called label sources. The core.labelSources configuration option (or -label-sources flag) of nfd-worker controls which sources to enable for label generation.
All built-in labels use the feature.node.kubernetes.io label namespace and have the following format.
feature.node.kubernetes.io/<feature> = <value>
+NOTE: Consecutive runs of nfd-worker will update the labels on a given node. If features are not discovered on a consecutive run, the corresponding label will be removed. This includes any restrictions placed on the consecutive run, such as restricting discovered features with the
-label-whitelistflag of nfd-master orcore.labelWhiteListoption of nfd-worker.
| Feature name | Value | Description |
|---|---|---|
cpu-cpuid.<cpuid-flag> | true | CPU capability is supported. NOTE: the capability might be supported but not enabled. |
cpu-cpuid.<cpuid-attribute> | string | CPU attribute value |
cpu-hardware_multithreading | true | Hardware multithreading, such as Intel HTT, enabled (number of logical CPUs is greater than physical CPUs) |
cpu-coprocessor.nx_gzip | true | Nest Accelerator for GZIP is supported(Power). |
cpu-power.sst_bf.enabled | true | Intel SST-BF (Intel Speed Select Technology - Base frequency) enabled |
cpu-pstate.status | string | The status of the Intel pstate driver when in use and enabled, either ‘active' or ‘passive'. |
cpu-pstate.turbo | bool | Set to ‘true' if turbo frequencies are enabled in Intel pstate driver, set to ‘false' if they have been disabled. |
cpu-pstate.scaling_governor | string | The value of the Intel pstate scaling_governor when in use, either ‘powersave' or ‘performance'. |
cpu-cstate.enabled | bool | Set to ‘true' if cstates are set in the intel_idle driver, otherwise set to ‘false'. Unset if intel_idle cpuidle driver is not active. |
cpu-security.sgx.enabled | true | Set to ‘true' if Intel SGX is enabled in BIOS (based on a non-zero sum value of SGX EPC section sizes). |
cpu-security.se.enabled | true | Set to ‘true' if IBM Secure Execution for Linux (IBM Z & LinuxONE) is available and enabled (requires /sys/firmware/uv/prot_virt_host facility) |
cpu-security.tdx.enabled | true | Set to ‘true' if Intel TDX is available on the host and has been enabled (requires /sys/module/kvm_intel/parameters/tdx). |
cpu-security.tdx.protected | true | Set to ‘true' if Intel TDX was used to start the guest node, based on the existence of the "TDX_GUEST" information as part of cpuid features. |
cpu-security.sev.enabled | true | Set to ‘true' if ADM SEV is available on the host and has been enabled (requires /sys/module/kvm_amd/parameters/sev). |
cpu-security.sev.es.enabled | true | Set to ‘true' if ADM SEV-ES is available on the host and has been enabled (requires /sys/module/kvm_amd/parameters/sev_es). |
cpu-security.sev.snp.enabled | true | Set to ‘true' if ADM SEV-SNP is available on the host and has been enabled (requires /sys/module/kvm_amd/parameters/sev_snp). |
cpu-model.vendor_id | string | Comparable CPU vendor ID. |
cpu-model.family | int | CPU family. |
cpu-model.id | int | CPU model number. |
The CPU label source is configurable, see worker configuration and sources.cpu configuration options for details.
| Flag | Description |
|---|---|
| ADX | Multi-Precision Add-Carry Instruction Extensions (ADX) |
| AESNI | Advanced Encryption Standard (AES) New Instructions (AES-NI) |
| APX_F | Intel Advanced Performance Extensions (APX) |
| AVX10 | Intel Advanced Vector Extensions 10 (AVX10) |
| AVX10_256, AVX10_512 | Intel AVX10 256-bit and 512-bit vector support |
| AVX | Advanced Vector Extensions (AVX) |
| AVX2 | Advanced Vector Extensions 2 (AVX2) |
| AVXIFMA | AVX-IFMA instructions |
| AVXVNNI | AVX (VEX encoded) VNNI neural network instructions |
| AMXBF16 | Advanced Matrix Extension, tile multiplication operations on BFLOAT16 numbers |
| AMXINT8 | Advanced Matrix Extension, tile multiplication operations on 8-bit integers |
| AMXFP16 | Advanced Matrix Extension, tile multiplication operations on FP16 numbers |
| AMXTILE | Advanced Matrix Extension, base tile architecture support |
| AVX512BF16 | AVX-512 BFLOAT16 instructions |
| AVX512BITALG | AVX-512 bit Algorithms |
| AVX512BW | AVX-512 byte and word Instructions |
| AVX512CD | AVX-512 conflict detection instructions |
| AVX512DQ | AVX-512 doubleword and quadword instructions |
| AVX512ER | AVX-512 exponential and reciprocal instructions |
| AVX512F | AVX-512 foundation |
| AVX512FP16 | AVX-512 FP16 instructions |
| AVX512IFMA | AVX-512 integer fused multiply-add instructions |
| AVX512PF | AVX-512 prefetch instructions |
| AVX512VBMI | AVX-512 vector bit manipulation instructions |
| AVX512VBMI2 | AVX-512 vector bit manipulation instructions, version 2 |
| AVX512VL | AVX-512 vector length extensions |
| AVX512VNNI | AVX-512 vector neural network instructions |
| AVX512VP2INTERSECT | AVX-512 intersect for D/Q |
| AVX512VPOPCNTDQ | AVX-512 vector population count doubleword and quadword |
| AVXNECONVERT | AVX-NE-CONVERT instructions |
| AVXVNNIINT8 | AVX-VNNI-INT8 instructions |
| AVXVNNIINT16 | AVX-VNNI-INT16 instructions |
| CMPCCXADD | CMPCCXADD instructions |
| ENQCMD | Enqueue Command |
| GFNI | Galois Field New Instructions |
| HYPERVISOR | Running under hypervisor |
| MSRLIST | Read/Write List of Model Specific Registers |
| PREFETCHI | PREFETCHIT0/1 instructions |
| VAES | AVX-512 vector AES instructions |
| VPCLMULQDQ | Carry-less multiplication quadword |
| WRMSRNS | Non-Serializing Write to Model Specific Register |
By default, the following CPUID flags have been blacklisted: AVX10 (use AVX10_VERSION instead), BMI1, BMI2, CLMUL, CMOV, CX16, ERMS, F16C, HTT, LZCNT, MMX, MMXEXT, NX, POPCNT, RDRAND, RDSEED, RDTSCP, SGX, SSE, SSE2, SSE3, SSE4, SSE42, SSSE3 and TDX_GUEST. See sources.cpu configuration options to change the behavior.
See the full list in github.com/klauspost/cpuid.
| Attribute | Description |
|---|---|
| AVX10_VERSION | AVX10 vector ISA version (if supported) |
| Flag | Description |
|---|---|
| IDIVA | Integer divide instructions available in ARM mode |
| IDIVT | Integer divide instructions available in Thumb mode |
| THUMB | Thumb instructions |
| FASTMUL | Fast multiplication |
| VFP | Vector floating point instruction extension (VFP) |
| VFPv3 | Vector floating point extension v3 |
| VFPv4 | Vector floating point extension v4 |
| VFPD32 | VFP with 32 D-registers |
| HALF | Half-word loads and stores |
| EDSP | DSP extensions |
| NEON | NEON SIMD instructions |
| LPAE | Large Physical Address Extensions |
| Flag | Description |
|---|---|
| AES | Announcing the Advanced Encryption Standard |
| EVSTRM | Event Stream Frequency Features |
| FPHP | Half Precision(16bit) Floating Point Data Processing Instructions |
| ASIMDHP | Half Precision(16bit) Asimd Data Processing Instructions |
| ATOMICS | Atomic Instructions to the A64 |
| ASIMRDM | Support for Rounding Double Multiply Add/Subtract |
| PMULL | Optional Cryptographic and CRC32 Instructions |
| JSCVT | Perform Conversion to Match Javascript |
| DCPOP | Persistent Memory Support |
| Feature | Value | Description |
|---|---|---|
kernel-config.<option> | true | Kernel config option is enabled (set ‘y' or ‘m'). Default options are NO_HZ, NO_HZ_IDLE, NO_HZ_FULL and PREEMPT |
kernel-selinux.enabled | true | Selinux is enabled on the node |
kernel-version.full | string | Full kernel version as reported by /proc/sys/kernel/osrelease (e.g. ‘4.5.6-7-g123abcde') |
kernel-version.major | string | First component of the kernel version (e.g. ‘4') |
kernel-version.minor | string | Second component of the kernel version (e.g. ‘5') |
kernel-version.revision | string | Third component of the kernel version (e.g. ‘6') |
The kernel label source is configurable, see worker configuration and sources.kernel configuration options for details.
| Feature | Value | Description |
|---|---|---|
memory-numa | true | Multiple memory nodes i.e. NUMA architecture detected |
memory-nv.present | true | NVDIMM device(s) are present |
memory-nv.dax | true | NVDIMM region(s) configured in DAX mode are present |
memory-swap.enabled | true | Swap is enabled on the node |
| Feature | Value | Description |
|---|---|---|
network-sriov.capable | true | Single Root Input/Output Virtualization (SR-IOV) enabled Network Interface Card(s) present |
network-sriov.configured | true | SR-IOV virtual functions have been configured |
| Feature | Value | Description |
|---|---|---|
pci-<device label>.present | true | PCI device is detected |
pci-<device label>.sriov.capable | true | Single Root Input/Output Virtualization (SR-IOV) enabled PCI device present |
<device label> is format is configurable and set to <class>_<vendor> by default. For more more details about configuration of the pci labels, see sources.pci options and worker configuration instructions.
| Feature | Value | Description |
|---|---|---|
usb-<device label>.present | true | USB device is detected |
<device label> is format is configurable and set to <class>_<vendor>_<device> by default. For more more details about configuration of the usb labels, see sources.usb options and worker configuration instructions.
| Feature | Value | Description |
|---|---|---|
storage-nonrotationaldisk | true | Non-rotational disk, like SSD, is present in the node |
| Feature | Value | Description |
|---|---|---|
system-os_release.ID | string | Operating system identifier |
system-os_release.VERSION_ID | string | Operating system version identifier (e.g. ‘6.7') |
system-os_release.VERSION_ID.major | string | First component of the OS version id (e.g. ‘6') |
system-os_release.VERSION_ID.minor | string | Second component of the OS version id (e.g. ‘7') |
The custom label source is designed for creating user defined labels. However, it has a few statically defined built-in labels:
| Feature | Value | Description |
|---|---|---|
custom-rdma.capable | true | The node has an RDMA capable Network adapter |
custom-rdma.enabled | true | The node has the needed RDMA modules loaded to run RDMA traffic |
NFD has many extension points for creating vendor and application specific labels. See the customization guide for detailed documentation.
NFD is able to create extended resources, see the NodeFeatureRule CRD and its extendedResources field for more details.
Note that NFD is not a replacement for the usage of device plugins.
An example use-case for extended resources could be based on custom feature (created e.g. with feature files that exposes the node SGX EPC memory section size. This value will then be turned into an extended resource of the node, allowing PODs to request that resource and the Kubernetes scheduler to schedule such PODs to only those nodes which have a sufficient capacity of said resource left.
Usage instructions.
Usage instructions.
Developer Preview This feature is currently in developer preview and subject to change. It is not recommended to use it in production environments.
The kubectl plugin kubectl nfd can be used to validate/dryrun and test NodeFeatureRule objects. It can be installed with the following command:
git clone https://github.com/kubernetes-sigs/node-feature-discovery
+ Kubectl plugin · Node Feature Discovery
Kubectl plugin
Table of contents
Developer Preview This feature is currently in developer preview and subject to change. It is not recommended to use it in production environments.
Overview
The kubectl plugin kubectl nfd can be used to validate/dryrun and test NodeFeatureRule objects. It can be installed with the following command:
git clone https://github.com/kubernetes-sigs/node-feature-discovery
cd node-feature-discovery
make build-kubectl-nfd
KUBECTL_PATH=/usr/local/bin/
@@ -13,4 +13,4 @@ Processing rule: my sample rule
*** Labels ***
vendor.io/my-sample-feature=true
NodeFeatureRule "examples/nodefeaturerule.yaml" is valid for NodeFeature "examples/nodefeature.yaml"
-
Node Feature Discovery master
\ No newline at end of file
+NFD-GC (NFD Garbage-Collector) is preferably run as a Kubernetes deployment with one replica. It makes sure that all NodeFeature and NodeResourceTopology objects have corresponding nodes and removes stale objects for non-existent nodes.
The daemon watches for Node deletion events and removes NodeFeature and NodeResourceTopology objects upon them. It also runs periodically to make sure no node delete event was missed and to remove any NodeFeature or NodeResourceTopology objects that were created without corresponding node. The default garbage collector interval is set to 1h which is the value when no -gc-interval is specified.
In Helm deployments (see garbage collector parameters) NFD-GC will only be deployed when featureGates.NodeFeatureAPI or topologyUpdater.enable is set to true.
NFD-GC (NFD Garbage-Collector) is preferably run as a Kubernetes deployment with one replica. It makes sure that all NodeFeature and NodeResourceTopology objects have corresponding nodes and removes stale objects for non-existent nodes.
The daemon watches for Node deletion events and removes NodeFeature and NodeResourceTopology objects upon them. It also runs periodically to make sure no node delete event was missed and to remove any NodeFeature or NodeResourceTopology objects that were created without corresponding node. The default garbage collector interval is set to 1h which is the value when no -gc-interval is specified.
In Helm deployments (see garbage collector parameters) NFD-GC will only be deployed when featureGates.NodeFeatureAPI or topologyUpdater.enable is set to true.
NFD-Master is responsible for connecting to the Kubernetes API server and updating node objects. More specifically, it modifies node labels, taints and extended resources based on requests from nfd-workers and 3rd party extensions.
The NodeFeature Controller uses NodeFeature objects as the input for the NodeFeatureRule processing pipeline. In addition, any labels listed in the NodeFeature object are created on the node (note the allowed label namespaces are controlled).
NFD-Master acts as the controller for NodeFeatureRule objects. It applies the rules specified in NodeFeatureRule objects on raw feature data and creates node labels accordingly. The feature data used as the input is received from nfd-worker instances through NodeFeature objects.
NOTE: when gRPC (DEPRECATED) is used for communicating the features (by setting the flag
-feature-gates NodeFeatureAPI=falseon both nfd-master and nfd-worker, or via Helm values.featureGates.NodeFeatureAPI=false), (re-)labelling only happens when a request is received from nfd-worker. That is, in practice rules are evaluated and labels for each node are created on intervals specified by thecore.sleepIntervalconfiguration option of nfd-worker instances. This means that modification or creation of NodeFeatureRule objects does not instantly cause the node labels to be updated. Instead, the changes only come visible in node labels as nfd-worker instances send their labelling requests. This limitation is not present when gRPC interface is disabled and NodeFeature API is used.
NFD-Master supports dynamic configuration through a configuration file. The default location is /etc/kubernetes/node-feature-discovery/nfd-master.conf, but, this can be changed by specifying the-config command line flag. Configuration file is re-read whenever it is modified which makes run-time re-configuration of nfd-master straightforward.
Master configuration file is read inside the container, and thus, Volumes and VolumeMounts are needed to make your configuration available for NFD. The preferred method is to use a ConfigMap which provides easy deployment and re-configurability.
The provided nfd-master deployment templates create an empty configmap and mount it inside the nfd-master containers. In kustomize deployments, configuration can be edited with:
kubectl -n ${NFD_NS} edit configmap nfd-master-conf
-In Helm deployments, Master pod parameter master.config can be used to edit the respective configuration.
See nfd-master configuration file reference for more details. The (empty-by-default) example config contains all available configuration options and can be used as a reference for creating a configuration.
NFD-Master runs as a deployment, by default it prefers running on the cluster's master nodes but will run on worker nodes if no master nodes are found.
For High Availability, you should increase the replica count of the deployment object. You should also look into adding inter-pod affinity to prevent masters from running on the same node. However note that inter-pod affinity is costly and is not recommended in bigger clusters.
Note: When NFD-Master is intended to run with more than one replica, it is advised to use
-enable-leader-electionflag. This flag turns on leader election for NFD-Master and let only one replica to act on changes in NodeFeature and NodeFeatureRule objects.
If you have RBAC authorization enabled (as is the default e.g. with clusters initialized with kubeadm) you need to configure the appropriate ClusterRoles, ClusterRoleBindings and a ServiceAccount for NFD to create node labels. The provided template will configure these for you.
NFD-Master is responsible for connecting to the Kubernetes API server and updating node objects. More specifically, it modifies node labels, taints and extended resources based on requests from nfd-workers and 3rd party extensions.
The NodeFeature Controller uses NodeFeature objects as the input for the NodeFeatureRule processing pipeline. In addition, any labels listed in the NodeFeature object are created on the node (note the allowed label namespaces are controlled).
NFD-Master acts as the controller for NodeFeatureRule objects. It applies the rules specified in NodeFeatureRule objects on raw feature data and creates node labels accordingly. The feature data used as the input is received from nfd-worker instances through NodeFeature objects.
NOTE: when gRPC (DEPRECATED) is used for communicating the features (by setting the flag
-feature-gates NodeFeatureAPI=falseon both nfd-master and nfd-worker, or via Helm values.featureGates.NodeFeatureAPI=false), (re-)labelling only happens when a request is received from nfd-worker. That is, in practice rules are evaluated and labels for each node are created on intervals specified by thecore.sleepIntervalconfiguration option of nfd-worker instances. This means that modification or creation of NodeFeatureRule objects does not instantly cause the node labels to be updated. Instead, the changes only come visible in node labels as nfd-worker instances send their labelling requests. This limitation is not present when gRPC interface is disabled and NodeFeature API is used.
NFD-Master supports dynamic configuration through a configuration file. The default location is /etc/kubernetes/node-feature-discovery/nfd-master.conf, but, this can be changed by specifying the-config command line flag. Configuration file is re-read whenever it is modified which makes run-time re-configuration of nfd-master straightforward.
Master configuration file is read inside the container, and thus, Volumes and VolumeMounts are needed to make your configuration available for NFD. The preferred method is to use a ConfigMap which provides easy deployment and re-configurability.
The provided nfd-master deployment templates create an empty configmap and mount it inside the nfd-master containers. In kustomize deployments, configuration can be edited with:
kubectl -n ${NFD_NS} edit configmap nfd-master-conf
+In Helm deployments, Master pod parameter master.config can be used to edit the respective configuration.
See nfd-master configuration file reference for more details. The (empty-by-default) example config contains all available configuration options and can be used as a reference for creating a configuration.
NFD-Master runs as a deployment, by default it prefers running on the cluster's master nodes but will run on worker nodes if no master nodes are found.
For High Availability, you should increase the replica count of the deployment object. You should also look into adding inter-pod affinity to prevent masters from running on the same node. However note that inter-pod affinity is costly and is not recommended in bigger clusters.
Note: When NFD-Master is intended to run with more than one replica, it is advised to use
-enable-leader-electionflag. This flag turns on leader election for NFD-Master and let only one replica to act on changes in NodeFeature and NodeFeatureRule objects.
If you have RBAC authorization enabled (as is the default e.g. with clusters initialized with kubeadm) you need to configure the appropriate ClusterRoles, ClusterRoleBindings and a ServiceAccount for NFD to create node labels. The provided template will configure these for you.
NFD-Topology-Updater is preferably run as a Kubernetes DaemonSet. This assures re-examination on regular intervals and/or per pod life-cycle events, capturing changes in the allocated resources and hence the allocatable resources on a per-zone basis by updating NodeResourceTopology custom resources. It makes sure that new NodeResourceTopology instances are created for each new nodes that get added to the cluster.
Because of the design and implementation of Kubernetes, only resources exclusively allocated to Guaranteed Quality of Service pods will be accounted. This includes CPU cores, memory and devices.
When run as a daemonset, nodes are re-examined for the allocated resources (to determine the information of the allocatable resources on a per-zone basis where a zone can be a NUMA node) at an interval specified using the -sleep-interval option. The default sleep interval is set to 60s which is the value when no -sleep-interval is specified. The re-examination can be disabled by setting the sleep-interval to 0.
Another option is to configure the updater to update the allocated resources per pod life-cycle events. The updater will monitor the checkpoint file stated in -kubelet-state-dir and triggers an update for every change occurs in the files.
In addition, it can avoid examining specific allocated resources given a configuration of resources to exclude via -excludeList
Kubelet PodResource API with the GetAllocatableResources functionality enabled is a prerequisite for nfd-topology-updater to be able to run (i.e. Kubernetes v1.21 or later is required).
Preceding Kubernetes v1.23, the kubelet must be started with --feature-gates=KubeletPodResourcesGetAllocatable=true.
Starting from Kubernetes v1.23, the KubeletPodResourcesGetAllocatable feature gate. is enabled by default
NFD-Topology-Updater supports configuration through a configuration file. The default location is /etc/kubernetes/node-feature-discovery/topology-updater.conf, but, this can be changed by specifying the-config command line flag.
NOTE: unlike nfd-worker, dynamic configuration updates are not supported.
Topology-Updater configuration file is read inside the container, and thus, Volumes and VolumeMounts are needed to make your configuration available for NFD. The preferred method is to use a ConfigMap which provides easy deployment and re-configurability.
The provided nfd-topology-updater deployment templates create an empty configmap and mount it inside the nfd-topology-updater containers. In kustomize deployments, configuration can be edited with:
kubectl -n ${NFD_NS} edit configmap nfd-topology-updater-conf
-In Helm deployments, Topology Updater parameters toplogyUpdater.config can be used to edit the respective configuration.
See nfd-topology-updater configuration file reference for more details. The (empty-by-default) example config contains all available configuration options and can be used as a reference for creating a configuration.
NFD-Topology-Updater is preferably run as a Kubernetes DaemonSet. This assures re-examination on regular intervals and/or per pod life-cycle events, capturing changes in the allocated resources and hence the allocatable resources on a per-zone basis by updating NodeResourceTopology custom resources. It makes sure that new NodeResourceTopology instances are created for each new nodes that get added to the cluster.
Because of the design and implementation of Kubernetes, only resources exclusively allocated to Guaranteed Quality of Service pods will be accounted. This includes CPU cores, memory and devices.
When run as a daemonset, nodes are re-examined for the allocated resources (to determine the information of the allocatable resources on a per-zone basis where a zone can be a NUMA node) at an interval specified using the -sleep-interval option. The default sleep interval is set to 60s which is the value when no -sleep-interval is specified. The re-examination can be disabled by setting the sleep-interval to 0.
Another option is to configure the updater to update the allocated resources per pod life-cycle events. The updater will monitor the checkpoint file stated in -kubelet-state-dir and triggers an update for every change occurs in the files.
In addition, it can avoid examining specific allocated resources given a configuration of resources to exclude via -excludeList
Kubelet PodResource API with the GetAllocatableResources functionality enabled is a prerequisite for nfd-topology-updater to be able to run (i.e. Kubernetes v1.21 or later is required).
Preceding Kubernetes v1.23, the kubelet must be started with --feature-gates=KubeletPodResourcesGetAllocatable=true.
Starting from Kubernetes v1.23, the KubeletPodResourcesGetAllocatable feature gate. is enabled by default
NFD-Topology-Updater supports configuration through a configuration file. The default location is /etc/kubernetes/node-feature-discovery/topology-updater.conf, but, this can be changed by specifying the-config command line flag.
NOTE: unlike nfd-worker, dynamic configuration updates are not supported.
Topology-Updater configuration file is read inside the container, and thus, Volumes and VolumeMounts are needed to make your configuration available for NFD. The preferred method is to use a ConfigMap which provides easy deployment and re-configurability.
The provided nfd-topology-updater deployment templates create an empty configmap and mount it inside the nfd-topology-updater containers. In kustomize deployments, configuration can be edited with:
kubectl -n ${NFD_NS} edit configmap nfd-topology-updater-conf
+In Helm deployments, Topology Updater parameters toplogyUpdater.config can be used to edit the respective configuration.
See nfd-topology-updater configuration file reference for more details. The (empty-by-default) example config contains all available configuration options and can be used as a reference for creating a configuration.
NFD-Worker is preferably run as a Kubernetes DaemonSet. This assures re-labeling on regular intervals capturing changes in the system configuration and makes sure that new nodes are labeled as they are added to the cluster. Worker connects to the nfd-master service to advertise hardware features.
When run as a daemonset, nodes are re-labeled at an default interval of 60s. This can be changed by using the core.sleepInterval config option.
The worker configuration file is watched and re-read on every change which provides a mechanism of dynamic run-time reconfiguration. See worker configuration for more details.
NFD-Worker supports dynamic configuration through a configuration file. The default location is /etc/kubernetes/node-feature-discovery/nfd-worker.conf, but, this can be changed by specifying the-config command line flag. Configuration file is re-read whenever it is modified which makes run-time re-configuration of nfd-worker straightforward.
Worker configuration file is read inside the container, and thus, Volumes and VolumeMounts are needed to make your configuration available for NFD. The preferred method is to use a ConfigMap which provides easy deployment and re-configurability.
The provided nfd-worker deployment templates create an empty configmap and mount it inside the nfd-worker containers. In kustomize deployments, configuration can be edited with:
kubectl -n ${NFD_NS} edit configmap nfd-worker-conf
+ NFD-Worker · Node Feature Discovery
NFD-Worker
NFD-Worker is preferably run as a Kubernetes DaemonSet. This assures re-labeling on regular intervals capturing changes in the system configuration and makes sure that new nodes are labeled as they are added to the cluster. Worker connects to the nfd-master service to advertise hardware features.
When run as a daemonset, nodes are re-labeled at an default interval of 60s. This can be changed by using the core.sleepInterval config option.
The worker configuration file is watched and re-read on every change which provides a mechanism of dynamic run-time reconfiguration. See worker configuration for more details.
Worker configuration
NFD-Worker supports dynamic configuration through a configuration file. The default location is /etc/kubernetes/node-feature-discovery/nfd-worker.conf, but, this can be changed by specifying the-config command line flag. Configuration file is re-read whenever it is modified which makes run-time re-configuration of nfd-worker straightforward.
Worker configuration file is read inside the container, and thus, Volumes and VolumeMounts are needed to make your configuration available for NFD. The preferred method is to use a ConfigMap which provides easy deployment and re-configurability.
The provided nfd-worker deployment templates create an empty configmap and mount it inside the nfd-worker containers. In kustomize deployments, configuration can be edited with:
kubectl -n ${NFD_NS} edit configmap nfd-worker-conf
In Helm deployments, Worker pod parameter worker.config can be used to edit the respective configuration.
See nfd-worker configuration file reference for more details. The (empty-by-default) example config contains all available configuration options and can be used as a reference for creating a configuration.
Configuration options can also be specified via the -options command line flag, in which case no mounts need to be used. The same format as in the config file must be used, i.e. JSON (or YAML). For example:
-options='{"sources": { "pci": { "deviceClassWhitelist": ["12"] } } }'
-
Configuration options specified from the command line will override those read from the config file.
Node Feature Discovery master
\ No newline at end of file
+Configuration options specified from the command line will override those read from the config file.
Nodes with specific features can be targeted using the nodeSelector field. The following example shows how to target nodes with Intel TurboBoost enabled.
apiVersion: v1
+ Using node labels · Node Feature Discovery
Using node labels
Nodes with specific features can be targeted using the nodeSelector field. The following example shows how to target nodes with Intel TurboBoost enabled.
apiVersion: v1
kind: Pod
metadata:
labels:
@@ -10,4 +10,4 @@
name: go1
nodeSelector:
feature.node.kubernetes.io/cpu-pstate.turbo: 'true'
-
For more details on targeting nodes, see node selection.
Node Feature Discovery master
\ No newline at end of file
+For more details on targeting nodes, see node selection.