Skip to main content
Vantage performance depends on both workload autoscaling and the underlying cluster’s ability to supply schedulable capacity. KEDA can change replica counts in response to CPU utilization and application metrics, but it does not add nodes, change quotas, or make nodes eligible for TechCore workloads.

Autoscaling with KEDA

Understand trigger strategies, enable KEDA, verify generated ScaledObjects, and plan capacity.

Monitoring with Prometheus

Configure the metrics pipeline used by Prometheus-based autoscaling triggers.

Sizing

CPU, memory, and pod-count requirements per node pool for minimal and production installs.

Additional workloads metadata

Apply advanced per-chart overrides, such as replica counts, to individual Vantage workloads.

Prerequisites

Label and taint TechCore and training worker nodes.

Troubleshooting

Diagnose unhealthy ScaledObjects, unavailable metrics, and scheduling failures.

​
Capacity principles

  • Treat the request totals on the Sizing page as an installation floor, not a production sizing guarantee.
  • Test with representative documents, skills, concurrency, and latency requirements.
  • Monitor HPA desired replicas and pending pods together so trigger problems are not confused with cluster-capacity problems.
  • Size node pools, quotas, registries, and storage for simultaneous scaling across multiple services.
  • Keep the vantage-operator and vantage-selfhosted charts on the same version so their autoscaling configuration remains compatible.