workload-variant-autoscaler: Dataflow¶

Controller Watches¶

Kubernetes resources this controller monitors for changes. Each watch triggers reconciliation when the watched resource is created, updated, or deleted.

Type	GVK	Source
For	/v1/ConfigMap	`internal/controller/configmap_reconciler.go:97`
For	api/v1/InferencePool	`internal/controller/inferencepool_reconciler.go:113`
For	api/v1alpha1/VariantAutoscaling	`internal/controller/variantautoscaling_controller.go:362`
For	apix/v1alpha2/InferencePool	`internal/controller/inferencepool_reconciler.go:109`

Reconciliation Flow¶

How the controller interacts with the Kubernetes API during reconciliation.

sequenceDiagram
    %% Static dataflow for workload-variant-autoscaler

    participant KubernetesAPI as Kubernetes API
    participant controller_manager as controller-manager

    KubernetesAPI->>+controller_manager: Watch ConfigMap (reconcile)
    KubernetesAPI->>+controller_manager: Watch InferencePool (reconcile)
    KubernetesAPI->>+controller_manager: Watch VariantAutoscaling (reconcile)
    KubernetesAPI->>+controller_manager: Watch InferencePool (reconcile)

    Note over KubernetesAPI: Defined CRDs
    Note right of KubernetesAPI: VariantAutoscaling (llmd.ai/v1alpha1)

Configuration¶

ConfigMaps and Helm values that control this component's runtime behavior.

ConfigMaps¶

Name	Data Keys	Source
saturation-scaling-config	default	`config/manager/configmap-saturation-scaling.yaml`
service-classes-config	freemium.yaml, premium.yaml	`deploy/configmap-serviceclass.yaml`
wva-queueing-model-config	default	`deploy/configmap-queueing-model.yaml`
wva-saturation-scaling-config	default	`deploy/configmap-saturation-scaling.yaml`
wva-variantautoscaling-config	GLOBAL_OPT_INTERVAL, PROMETHEUS_BASE_URL, PROMETHEUS_METRICS_CACHE_CLEANUP_INTERVAL, PROMETHEUS_METRICS_CACHE_FETCH_INTERVAL, PROMETHEUS_METRICS_CACHE_FRESH_THRESHOLD, PROMETHEUS_METRICS_CACHE_MAX_SIZE, PROMETHEUS_METRICS_CACHE_STALE_THRESHOLD, PROMETHEUS_METRICS_CACHE_TTL, PROMETHEUS_METRICS_CACHE_UNAVAILABLE_THRESHOLD, PROMETHEUS_TLS_INSECURE_SKIP_VERIFY, WVA_LIMITED_MODE, WVA_NODE_SELECTOR, WVA_SCALE_TO_ZERO	`config/manager/configmap.yaml`
wva-variantautoscaling-config	GLOBAL_OPT_INTERVAL, PROMETHEUS_BASE_URL, PROMETHEUS_METRICS_CACHE_CLEANUP_INTERVAL, PROMETHEUS_METRICS_CACHE_FETCH_INTERVAL, PROMETHEUS_METRICS_CACHE_FRESH_THRESHOLD, PROMETHEUS_METRICS_CACHE_STALE_THRESHOLD, PROMETHEUS_METRICS_CACHE_TTL, PROMETHEUS_METRICS_CACHE_UNAVAILABLE_THRESHOLD, PROMETHEUS_TLS_INSECURE_SKIP_VERIFY	`config/openshift/configmap-patch.yaml`

Helm¶

Chart: workload-variant-autoscaler v0.5.1