TensorFusion DocsFor System Admins
Helm Install Values
Every key in the values.yaml of the tensor-fusion Helm chart, with its default and meaning
This page lists every key in values.yaml of the tensor-fusion Helm chart, with its type, default, and effect. It matches chart 1.8.8 (appVersion 2.17.3).
Pass overrides with -f my-values.yaml or --set key=value:
helm upgrade --install tensor-fusion-sys tensor-fusion \
--repo https://download.tensor-fusion.ai \
--namespace tensor-fusion-sys --create-namespace \
-f my-values.yamlHelm replaces lists as a whole. When you override a list such as dynamicConfig.alertRules, providerConfigs.nvidia.hardwareMetadata or controller.command, copy the default list first and edit the copy. Otherwise the default entries are dropped.
Overlay values files
The chart ships three overlay files next to values.yaml. Add them with -f.
| File | What it changes | When to use |
|---|---|---|
values-production.yaml | 2 controller replicas, 3 Alertmanager replicas with a 5Gi volume, higher resource requests, greptime.installStandalone: false | Production. You must also set greptime.host and related keys. Available at https://download.tensor-fusion.ai/values-production.yaml. |
values-cn.yaml | Image repositories for the controller, agent, GreptimeDB, vector, Alertmanager and the webhook cert job point to mirrors reachable from mainland China | Clusters that cannot pull from Docker Hub, quay.io or registry.k8s.io. Available at https://download.tensor-fusion.ai/values-cn.yaml. |
values-multi-vendor.yaml | initialGpuNodeLabelSelector: "" | Clusters with accelerators from more than one vendor. In the chart source only. |
Configuration Properties
| Property | Type | Description |
|---|---|---|
| imagePullSecrets | array | Image pull secrets added to the controller Pod (default: []) |
| nameOverride | string | Overrides the chart name used in resource names and labels (default: "") |
| fullnameOverride | string | Overrides the full resource name prefix. Unset, the prefix is the Helm release name (default: "") |
| namespaceOverride | string | Namespace for the chart's namespaced resources. Unset, the release namespace is used (default: "") |
| serviceAccount | object | ServiceAccount of the controller |
| initialGpuNodeLabelSelector | string | Label selector of the nodes the controller watches as GPU nodes. Set it to "" to watch every node (default: nvidia.com/gpu.present=true) |
| controller | object | The controller Deployment: CRD controllers, Pod webhook and the embedded scheduler |
| greptime | object | GreptimeDB, the store for metrics and system logs |
| agent | object | cluster-agent sidecar that connects the cluster to the TensorFusion Console |
| cloudVendorCredentials | object | Access key pair for pools in Provisioned mode |
| providerConfigs | object | Default ProviderConfig resources created by the chart. Only nvidia exists |
| schedulingConfigTemplate | object | Default SchedulingConfigTemplate created by the chart |
| cluster | object | Default TensorFusionCluster with one NVIDIA GPU pool |
| alert | object | Alertmanager StatefulSet deployed by the chart |
| schedulerConfig | object | KubeSchedulerConfiguration of the embedded scheduler |
| dynamicConfig | object | Global settings the controller reloads at runtime, including alert rules |