Groundcover has opened a beta for its Autonomous Infrastructure feature, which continuously tunes CPU and memory requests on Kubernetes Deployments based on real‑world telemetry collected by its eBPF sensor. Engineers can now see request‑vs‑usage metrics, under‑provisioned workload counts, and projected monthly savings, turning a traditionally manual, periodic task into an automated, data‑driven process.
Kubernetes rightsizing automation in practice
The beta leverages historical usage data to recommend request values for each Deployment. New Deployments start in a learning mode until sufficient data is gathered. Unlike reactive approaches that wait for a spike before adjusting resources, the engine analyses usage patterns and contention trends to anticipate future demand, allowing it to increase or decrease requests proactively. Engineers retain control through policy settings that define acceptable bounds for adjustments.
Operational considerations
Because request changes can affect pod scheduling, the system integrates with existing node autoscalers such as Karpenter, which base node provisioning on the sum of pod requests. More accurate requests give the autoscaler a better signal, potentially reducing excess node capacity. Some adjustments may require pod restarts, depending on whether the underlying container runtime permits in‑place resizing. Teams should monitor the beta’s recommendation dashboard to validate changes before they are applied, especially in environments with strict SLAs.
Implications for reliability and security
Over‑provisioning is a common safety net, but it inflates infrastructure spend. Under‑provisioning, on the other hand, can lead to CPU throttling, memory OOM kills, and pod eviction under node pressure—each of which can degrade service availability and expose workloads to denial‑of‑service conditions. By keeping request values aligned with actual consumption, the beta helps maintain a balance between cost and stability, which is a core concern for SRE and security engineers alike.
Related CloudNinjas coverage: hands-on guides.
What This Means For Practitioners
Adopt the beta in a controlled namespace to evaluate recommendation accuracy against your own cost model. Verify that any automatic restarts align with your deployment pipelines, and adjust policy bounds to reflect acceptable risk thresholds. Keep an eye on node autoscaler behavior after request changes, and compare projected savings with actual billing data. Finally, incorporate the recommendation dashboard into regular observability reviews to ensure the system continues to meet both performance and security expectations.
