Harvester Auto-Scaling: Threshold evaluation and interval behavior

Resource Optimization for Low-Traffic Workloads

Harvester provides an Auto-Scaling feature designed to reduce operational costs by adjusting worker instance counts based on request-rate thresholds. To optimize a low-traffic environment, it is necessary to define precise minimum and maximum instance bounds to ensure availability while minimizing idle compute spend.

Configuration Uncertainty

While the documentation specifies that scaling is triggered by request rates, there is a lack of clarity regarding the scope of these metrics. It is unclear if the system evaluates thresholds on a per-node basis or as a global aggregate across the entire cluster. Additionally, the default scaling interval—the frequency at which the system evaluates these thresholds—is not explicitly defined for recent releases.

These gaps make it difficult to predict the exact timing of scale-down events and the resulting impact on billing cycles.

• Is the request-rate threshold evaluated per instance or globally across the cluster?
• What is the default scaling interval used to evaluate these thresholds?