The module lifecycle stage: Experimental
The module has requirements for installation
The adaptive-resource-management module is intended for cluster administrators and platform engineers who want to automate the selection of resource requests and limits for workloads. It eliminates the need to manually tune CPU and memory parameters by leveraging VPA (Vertical Pod Autoscaler) recommendations.
The module deploys the AutoVPA controller based on Goldilocks, which automatically creates VPA objects for workloads in selected namespaces and provides resource recommendations.
Main Features
- Automatic creation of VPA objects for Deployments, StatefulSets, DaemonSets, and Jobs in managed namespaces.
- Flexible namespace selection: manage all namespaces, only namespaces with a specific label, or namespaces matching a label selector.
- Combined selection mode: use a label selector together with the special label
autovpa.deckhouse.io/enabledto form a union of matching namespaces. - No changes to existing workload manifests are required; VPA objects are created and maintained automatically.
- Minimal resource footprint: the controller runs as a single replica with low CPU and memory requirements.
Limitations
- The module requires the
vertical-pod-autoscalermodule to be enabled in the cluster for VPA objects to function. - AutoVPA creates VPA objects in
Offmode (recommendation-only); it does not apply resource changes to running pods automatically.