Groundcover has released Autonomous Infrastructure in beta. The feature automatically adjusts CPU and memory requests for Kubernetes workloads based on actual IT usage in production. The technology comes from Wand Cloud, which Groundcover acquired in September.
The problem Groundcover aims to solve is a familiar one for platform teams. Kubernetes requires engineers to configure requests in advance for each workload. This scaling question must therefore be answered before actual traffic begins flowing. According to Groundcover, most teams rarely review these settings anymore. The consequences range from out-of-memory errors to unused capacity
Groundcover offers a Kubernetes-focused observability platform. It recently expanded its agent with an Agent Mode for incident investigation. That investigation can be conducted in natural language via an LLM. The telemetry collected by the platform now forms the basis for rightsizing capabilities.
The data is already there
Since the Groundcover sensor has already built up a production history, customers running a supported version do not need to install anything. Requested CPU and memory appear alongside actual usage, along with an estimated monthly savings and the number of workloads that are under-resourced. A graph shows how much of that savings has already been realized, allowing users to quantify the value of the Groundcover offering.
Note that in the beta, the amounts are based on prices per CPU and memory unit that customers enter themselves. Public cloud prices will follow, then billing data such as the AWS Cost and Usage Report.
Stability first
Most optimization tools determine the size of each workload individually, according to Groundcover. A request that works for one service might leave neighboring workloads short on resources on a full node. Autonomous Infrastructure therefore first checks how full the node is before increasing a request.
CEO Shahar Azulay emphasizes that customers can first assess how much their clusters can save before allowing the software to make any changes. The feature is disabled by default. An engineer chooses which deployments, namespaces, or clusters are included. During the beta, the software only affects requests; limits remain untouched. Each policy has minimum and maximum limits, and an audit log records who linked which workload to a policy and when. Groundcover recommends starting with development namespaces and, for example, leaving a production database alone for the time being.
See also: Groundcover makes its observability agent more versatile