Advanced Cluster Management 5 moves fleet right-sizing to MCOA and Perses

Red Hat Advanced Cluster Management 5 has made its newer multicluster right-sizing architecture generally available, shifting recommendation delivery to the multicluster observability add-on and making Perses the primary dashboard experience. The change matters to platform and FinOps teams because it alters how CPU and memory recommendations reach managed clusters while preserving the existing configuration model. (Red Hat Developer)

What changed

The release replaces the previous governance-policy path with an MCOA flow that generates resources per cluster and distributes PrometheusRule objects through ManifestWork. Aggregated metrics return to hub Thanos and feed Perses dashboards integrated into the OpenShift console. Namespace and OpenShift Virtualization right-sizing remain configured through the MultiClusterObservability resource and the existing rs-namespace-config and rs-virt-config ConfigMaps. (Red Hat Developer)

For upgrades from Advanced Cluster Management 2.16 or 2.17, Red Hat says the standard migration requires no manual cleanup. The controller removes legacy Policy, Placement and PlacementBinding resources, creates the required MCOA resources and Perses dashboards, and preserves existing right-sizing customizations. The old preview opt-in annotation is ignored after upgrade, while placementConfiguration now determines which managed clusters receive the rules. Teams that previously relied on a custom namespaceBinding should verify their targeting after migration. (Red Hat Developer)

Profiles add a tuning decision

Advanced Cluster Management 5 also introduces independent CPU and memory aggregation profiles. The defaults are maximum observed usage, P99 and P95; administrators can define additional integer percentiles from P1 through P99 in the right-sizing ConfigMaps. Recommendations use five-minute usage series over a rolling one-day window, then apply the configured recommendation percentage, which defaults to 110 percent. (Red Hat Developer)

That flexibility is useful but not automatic optimization. A lower percentile reduces the influence of rare peaks and can produce tighter requests, while a higher percentile retains more headroom. Red Hat explicitly recommends validating profile changes against workload service-level objectives and applying resource adjustments gradually. (Red Hat Developer)

What platform teams should check

After installing or upgrading, teams should confirm that multicluster observability and the required right-sizing modes are enabled, then inspect the namespace and virtualization dashboards under Observe → Dashboards (Perses). Historical data is not backfilled, and recommendations can take roughly 30 minutes to appear after enablement. The system supports both MCOA-managed platform-metrics federation and the older observability allowlist pipeline, allowing teams to adopt the new dashboard and delivery model without immediately replacing their collection architecture. (Red Hat Developer)