Workload Autoscaler Helm Values

Use exported CLOUDPILOT_WA_HELM_VALUES YAML to customize the Workload Autoscaler (WA) Helm installation. This input configures the WA controller, certificate job, metrics storage, and NodeAgent. To configure workload recommendations and updates, use WA policies and workloads.

Availability: the selected installation script must support CLOUDPILOT_WA_HELM_VALUES. Older generated installers ignore this variable. Confirm that your target release includes support before using it. For an upgrade sequence with intermediate versions, every WA installer in that sequence must support the input and the supplied fields. An older step can ignore the variable and reset custom settings before a later step reapplies them. Available chart fields and defaults also depend on each selected WA chart version.

Using saved Pre Run settings for a standalone WA installation also requires server support for prepending Pre Run to that command. Older servers omit it; export the YAML directly in your terminal before running the generated command instead.

Set values in Pre Run

For supported AWS, Azure, and GCP clusters, open Settings > Advanced Options > Pre Run Command in the Console, add an exported YAML string, and save. Then obtain and run the cluster’s installation or manual-upgrade command.

Standalone WA installation executes Pre Run before the installation command; it does not execute Post Run. Cluster manual upgrades retain their existing Pre Run and Post Run behavior.

For example, this partial configuration adds a toleration to the WA controller:

export CLOUDPILOT_WA_HELM_VALUES='
tolerations:
  - key: example.com/dedicated
    operator: Equal
    value: platform
    effect: NoSchedule
'

On Azure, retain the installer-generated Spot toleration when replacing a list; see the Azure example. Setting only the root list does not configure the certificate job, metrics storage, or NodeAgent.

You can export the variable directly in the shell before running the generated command. This also applies to Universal clusters, whose Settings page currently does not expose the Pre Run editor. To read an existing file:

export CLOUDPILOT_WA_HELM_VALUES="$(cat ./wa-values.yaml)"

The file must exist in the command’s execution environment. Both release and development installers write the YAML literally to a temporary file. They do not expand ${VARIABLE} placeholders inside it. Export is required because the installer executes in a child shell; the single-quoted examples also prevent shell expansion.

Saving Pre Run alone does not update running resources. Keep the configuration for subsequent shell upgrades. Automatic upgrades do not execute Pre Run or apply newly edited values through these installation scripts; use the supported installation/manual-upgrade command to apply a change.

Empty and partial values

Only include fields you intend to change.

The installers use --reset-values. They do not discover and retain arbitrary historical Helm overrides: omitted fields return to target defaults, subject to installer arguments. Existing storage-class and tracing-mount handling remain exceptions described under installer precedence.

Find all values for your version

The managed Helm release is workload-autoscaler, in namespace cloudpilot. Its chart version can differ from the Agent and NA chart versions. Use the target WA chart version selected by the generated installer (CLOUDPILOT_WORKLOAD_AUTOSCALER_CHART_VERSION), not the NA installer’s CLOUDPILOT_REBALANCE_VERSION.

Replace <target-wa-chart-version> before running:

helm repo add onboard https://onboard.cloudpilot.ai
helm repo update onboard
export WA_CHART_VERSION='<target-wa-chart-version>'
helm show values onboard/workload-autoscaler --version "${WA_CHART_VERSION}" > wa-defaults.yaml

This is the complete defaults document for that chart version, not the complete effective installation configuration. The installer adds provider defaults and explicit arguments. Copy only the overrides you need.

Inspect the installed release separately:

helm list --namespace cloudpilot --filter '^workload-autoscaler$'
helm get values workload-autoscaler --namespace cloudpilot --all --output yaml

Helm can accept unknown keys that no template reads. In the current chart, nodeAgent.nodeSelector is the only exposed node-selector field. Root nodeSelector, certGenerator.nodeSelector, and victoriaMetrics.nodeSelector do not select nodes. The NA controller uses a separate chart and input.

Complete value reference

These paths are relative to the YAML root. Image, tracing, and cold-memory fields require support in the target chart and component versions.

Each of the following image groups supports all four fields: repository, tag, digest, and pullPolicy.

For example, image.controller.repository is the controller repository and image.controller.pullPolicy is its pull policy. Preserve the supported repository/tag/digest combination; changing a tag alone does not override a pinned digest. Release image-source selection can override these repositories.

Installer precedence

The effective order, from lowest to highest priority, is:

1. Target chart defaults.
2. Azure-generated values, when CLOUD_PROVIDER=azure.
3. YAML exported as CLOUDPILOT_WA_HELM_VALUES.
4. Explicit installer --set, --set-json, and release image-source arguments.

To change an existing debugfs-only NodeAgent installation to mount tracefs, use export NODE_AGENT_TRACEFS_ENABLED=true after verifying the selected nodes expose the required filesystem. A YAML true value alone does not override a preserved false installer argument. This retains existing installer behavior.

NodeAgent enablement

The Console’s standalone installation command includes export ENABLE_NODE_AGENT=true. To disable NodeAgent for that installation, change this assignment to export ENABLE_NODE_AGENT=false in the copied command. Setting false earlier in your terminal or in Pre Run does not override the unchanged command’s true assignment. The resulting explicit installer argument also takes precedence over nodeAgent.enable in customer YAML.

Storage class

For a new installation, set the supported installation variable before running the command:

export STORAGE_CLASS='example-wait-for-first-consumer'

The named class must exist and use volumeBindingMode: WaitForFirstConsumer. When STORAGE_CLASS is omitted, the installer checks the cluster’s default class. The StorageClass preflight does not read customer YAML, so setting only victoriaMetrics.storageClass can fail before Helm or validate a different class. Use STORAGE_CLASS for this choice and omit victoriaMetrics.storageClass from custom YAML.

For an existing metrics StatefulSet, its nonempty storage-class setting takes precedence over STORAGE_CLASS and YAML. If its template omits the field, avoid adding a new class through YAML: volume claim templates cannot be changed in place. Keep the existing storage configuration; this input is not a storage migration mechanism. SKIP_SC_VERIFICATION=true only skips the new-install binding-mode check and does not resolve storage-class conflicts.

Tolerations examples

WA has four independent toleration paths. The root tolerations list does not propagate to the certificate job, metrics StatefulSet, or NodeAgent. A toleration permits a matching taint; it does not select nodes.

AWS, GCP, and other non-Azure installations

This example replaces each configurable list with an illustrative dedicated-node toleration:

export CLOUDPILOT_WA_HELM_VALUES='
tolerations:
  - key: example.com/dedicated
    operator: Equal
    value: platform
    effect: NoSchedule
certGenerator:
  tolerations:
    - key: example.com/dedicated
      operator: Equal
      value: platform
      effect: NoSchedule
victoriaMetrics:
  tolerations:
    - key: example.com/dedicated
      operator: Equal
      value: platform
      effect: NoSchedule
nodeAgent:
  tolerations:
    - key: example.com/dedicated
      operator: Equal
      value: platform
      effect: NoSchedule
'

The three cluster-wide component lists default to empty on non-Azure installs. NodeAgent still has its built-in operator: Exists toleration, which tolerates all taints. Adding a narrower NodeAgent toleration does not narrow its placement. To limit NodeAgent coverage, use nodeAgent.nodeSelector and verify the selected nodes still cover the workloads whose runtime metrics you need.

Azure / AKS

The installer adds kubernetes.azure.com/scalesetpriority=spot:NoSchedule to all four lists. When replacing them, retain that entry wherever Spot eligibility is required:

export CLOUDPILOT_WA_HELM_VALUES='
tolerations:
  - key: kubernetes.azure.com/scalesetpriority
    operator: Equal
    value: spot
    effect: NoSchedule
  - key: example.com/dedicated
    operator: Equal
    value: platform
    effect: NoSchedule
certGenerator:
  tolerations:
    - key: kubernetes.azure.com/scalesetpriority
      operator: Equal
      value: spot
      effect: NoSchedule
    - key: example.com/dedicated
      operator: Equal
      value: platform
      effect: NoSchedule
victoriaMetrics:
  tolerations:
    - key: kubernetes.azure.com/scalesetpriority
      operator: Equal
      value: spot
      effect: NoSchedule
    - key: example.com/dedicated
      operator: Equal
      value: platform
      effect: NoSchedule
nodeAgent:
  tolerations:
    - key: kubernetes.azure.com/scalesetpriority
      operator: Equal
      value: spot
      effect: NoSchedule
    - key: example.com/dedicated
      operator: Equal
      value: platform
      effect: NoSchedule
'

The built-in NodeAgent toleration remains in addition to its configured list. Replace illustrative taints with those on your nodes and include every other required toleration when replacing a list.

Configure NA and WA independently

CLOUDPILOT_NA_HELM_VALUES configures only the NA controller chart. CLOUDPILOT_WA_HELM_VALUES configures only the WA chart. You can export both in one Pre Run Command; neither variable is automatically copied into the other chart.

For example, this partial configuration adds an AWS/Azure NA node selector and limits WA NodeAgent coverage independently, without replacing toleration lists:

export CLOUDPILOT_NA_HELM_VALUES='
nodeSelector:
  example.com/node-pool: platform
'
export CLOUDPILOT_WA_HELM_VALUES='
nodeAgent:
  nodeSelector:
    example.com/runtime-metrics: "true"
'

For GCP NA, put its selector under controller as described in Node Autoscaler Helm values. The WA NodeAgent path is the same across providers. Existing explicit selector arguments, if supplied, retain precedence. These selectors restrict placement; ensure eligible nodes exist before applying them.

Verify and recover

After installation or a shell upgrade, inspect both the Helm values and all WA components:

helm get values workload-autoscaler --namespace cloudpilot --all --output yaml
kubectl get deployment cloudpilot-workload-autoscaler --namespace cloudpilot
kubectl get statefulset cloudpilot-victoria-metrics --namespace cloudpilot
kubectl get daemonset cloudpilot-node-agent --namespace cloudpilot
kubectl get jobs,pods --namespace cloudpilot --output wide
kubectl get events --namespace cloudpilot --sort-by=.metadata.creationTimestamp

The NodeAgent DaemonSet is absent when disabled. Helm runs the certificate Job as a post-install/post-upgrade hook; successful Jobs can be deleted by its hook policy. The installation wrapper’s existing readiness wait covers the WA controller Pods, not metrics storage or NodeAgent. Inspect those components and use kubectl describe pod <pod-name> --namespace cloudpilot for scheduling failures.

Malformed YAML fails in Helm, potentially after the installer has already applied CRDs. Correct the values and rerun the supported command. For Pending components, restore the last known working tolerations or selector and check storage binding. An empty custom input returns to target defaults plus installer behavior on the next run; it does not recover arbitrary earlier custom values.