Datadog outlines three deployment paths for Pod Autoscaler to cut idle Kubernetes cost

Datadog explains how its Pod Autoscaler and supporting tools help platform teams reduce idle Kubernetes cost by rightsizing requests, limits and replicas across clusters through an in‑app rollout, GitOps cluster profiles, and AI‑assisted pull requests.

Datadog describes how its Pod Autoscaler and related tooling let platform teams apply workload‑level autoscaling recommendations across Kubernetes fleets to reduce idle cost. The blog post identifies the main causes of wasted capacity — overprovisioned CPU and memory requests, idle replicas preserved for headroom, and obsolete Horizontal Pod Autoscalers — and positions workload rightsizing as the largest savings opportunity beyond node autoscaling. That framing matters because it shifts teams’ focus from nodes to per‑workload CPU/memory and replica settings, where the biggest idle cost reductions can be achieved.

One deployment path is an in‑app setup that gives operators a centralized UI to inspect workloads across clusters and manage rollout in bulk. From the autoscaling setup page, teams can see which workloads are ready for activation and which require existing autoscaling settings to be carried over, then enable autoscaling across many workloads without editing YAML. The UI generates and deploys DatadogPodAutoscaler objects, surfaces estimated idle cost for prioritization, and removes manual manifest authoring as a gating step for large‑scale rollouts.

For GitOps environments, Datadog supports cluster profiles implemented as policy‑as‑code. Teams can choose from three standard workload scaling profiles or author a DatadogPodAutoscalerClusterProfile custom resource and apply it by labeling namespaces; the Datadog Cluster Agent detects those labels and automatically creates DatadogPodAutoscaler resources for eligible workloads, including Deployments and Argo Rollouts. AI can generate manifests and pull requests to speed adoption, while vertical resizing lets teams adjust resource requests (CPU/memory) in place rather than only changing replica counts, enabling combined recommendations and implementations for both replicas and resource requests.

Taken together, the in‑app UI, GitOps cluster profiles and AI‑generated PRs provide multiple operational models for platform teams: guided rollouts from a central console, policy enforcement via code, or automated onboarding to accelerate adoption. For builders and SREs the practical outcome is a shorter path from recommendation to rollout, with prioritization driven by estimated cost impact and mechanisms designed to apply changes safely at scale. By automatically creating DatadogPodAutoscaler resources and supporting both replica autoscaling and vertical request adjustments, the tooling aims to reduce idle cost across fleets while preserving existing operational controls.