Right-sizing recommendations with MCOA and Perses dashboards

Since the general availability of right-sizing recommendations in Red Hat Advanced Cluster Management for Kubernetes 2.16, right-sizing has been available for platform engineers and FinOps teams to leverage Grafana-based dashboards to identify over-provisioned and under-utilized resources across their multicluster environments. This feature has matured through developer preview, technology preview for namespaces and Red Hat OpenShift Virtualization, and ultimately reached GA.

Now, we are excited to announce the developer preview of right-sizing recommendations powered by Perses dashboards delivered through the multicluster observability add-on (MCOA) as part of the Red Hat Advanced Cluster Management for Kubernetes 2.17 release. This represents a process of significant architectural evolution, migrating from Grafana to Perses as the visualization layer, while also moving the right-sizing logic into the MCOA for a cleaner, more modular deployment model. This developer preview covers namespace-level right-sizing and OpenShift Virtualization right-sizing with Perses dashboards.

Why migrate to Perses

Perses is a CNCF project that provides a modern, open source dashboard solution designed specifically for cloud-native observability. The migration from Grafana to Perses brings several advantages:

1. Native Kubernetes integration: Perses dashboards are defined as code using the Perses Go SDK, enabling version-controlled, programmatically generated dashboards that live alongside the application code.
2. Tighter console integration: Perses dashboards embedded directly into the Red Hat Advanced Cluster Management console monitoring experience, providing seamless navigation between cluster management and resource optimization views.
3. Interactive data links: Tables support clickable links that drill down from namespace-level overview to workload-level and pod-level detail views, enabling a fluid top-down investigation flow.
4. Open source sustainability: As a CNCF project, Perses aligns with Red Hat's commitment to open source ecosystems and community-driven innovation.
5. Dashboards as code: Right-sizing dashboards defined in Go using the Perses SDK, making them testable, reviewable, and maintainable through standard software engineering practices.

Namespace right-sizing with Perses

The Red Hat Advanced Cluster Management right-sizing namespace Perses dashboard provides a comprehensive view of resource allocation efficiency across all namespaces in a selected cluster. Start by selecting the cluster you would like to explore, the recommendation profile, and preferred aggregation period from the dropdown at the top of the dashboard (Figure 1).

• Cluster-level summary: At the top, four stat panels show CPU recommendation, usage, request, and utilization for the selected cluster. It presents these four metrics for memory, giving administrators a complete picture of cluster resource efficiency. Utilization is color-coded: green for efficient, red for under-utilized, yellow for over-utilized (Figure 2-3).

• Utilization of top namespaces: A time-series chart displays the top 20 namespaces by CPU and memory utilization ratio over time, enabling quick identification of the most over-provisioned or under-provisioned namespaces (Figure 2-3).

• Namespace quota tables: Detailed tables for both CPU and memory display per-namespace Utilization %, Usage, Request, Recommendation, and Request Hard (Figure 2-3).

OpenShift Virtualization right-sizing with Perses

The Red Hat Advanced Cluster Management right-sizing OpenShift Virtualization Perses dashboard provides purpose-built views for virtual machine resource optimization.

Four stat panels summarize the current state across all VMs in the selected cluster and namespace: Total CPU Overestimation (cores that can be reclaimed, highlighted in red), Total CPU Underestimation (additional cores needed, highlighted in yellow), Total Memory Overestimation, and Total Memory Underestimation. These aggregate views only consider running VMs, ensuring recommendations are based on active workloads (Figure 4-5).

Four interactive tables provide granular per-VM analysis for CPU overestimation: CPU Underestimation, Memory Overestimation, and Memory Underestimation. Each table shows VM Name, Namespace, Utilization %, Usage, Request, Recommendation, and the overestimation/underestimation delta.

Each VM name is a clickable link that navigates to a dedicated detail dashboard showing (Figure 6-9):

• Stat panels for CPU/memory over-estimation or under-estimation, usage, request, and utilization
• Time-series charts depicting CPU and memory utilization over time for the selected VM

We can filter the information by cluster, namespace, and preferred time aggregation (e.g., 5/10/30/60/90 days).

Get started

Now let's get started. This section will demonstrate how to enable right-sizing with Perses dashboards.

Prerequisites:

• Red Hat Advanced Cluster Management for Kubernetes 2.17
• Multicluster observability operator (MCO) installed on the hub
• Multicluster observability add-on (MCOA) deployed
• Observability components (Prometheus, Thanos, Perses) active
• OpenShift Virtualization enabled on managed clusters for virtualization right-sizing

Enable right-sizing with Perses dashboards

To try the enhanced delivery model and Perses dashboards, use a single command to activate the new experience:

kubectl annotate mco observability \
observability.open-cluster-management.io/right-sizing-capable=v1

It switches right-sizing to the modern add-on framework. It cleans up legacy policy resources automatically, Perses dashboards appear in the OpenShift console, and preserves all existing customizations (filters, thresholds, cluster targeting): no downtime, no manual migration.

To opt out with instant rollback:

kubectl annotate mco observability \
observability.open-cluster-management.io/right-sizing-capable-

This removes the annotation and reverts to the previous delivery model immediately: no data loss, no disruption.

By default, the configuration to your MultiClusterObservability CR on the hub cluster will be enabled for both namespace and virtualization right-sizing, and it will automatically:

1. Deploy Prometheus recording rules to managed clusters via ManifestWorks
2. Configure metric forwarding through the observability pipeline
3. Register Perses dashboards with the hub's Perses server

Once it is done, access the dashboards through the Red Hat Advanced Cluster Management console under Observability →Dashboards or navigate directly to Red Hat Advanced Cluster Management right-sizing namespace and Red Hat Advanced Cluster Management right-sizing OpenShift Virtualization.

Metrics collections

Right-sizing generates Prometheus recording rules on each managed cluster and federates the resulting metrics back to the hub. The full end-to-end experience includes, recording rules deployed automatically, metrics federated seamlessly, and Perses dashboards populated immediately. We recommend enabling platform metrics collection via the observability add-on.

Enable platform metrics collection. Configure platform metrics collection through the capabilities section of the MultiClusterObservability CR. To enable it, ensure the following is set on the hub cluster:

apiVersion: observability.open-cluster-management.io/v1beta2
kind: MultiClusterObservability
metadata:
  annotations: 
    observability.open-cluster-management.io/right-sizing-capable: v1 
  name: observability
spec:
  capabilities:
    platform:
      analytics:
        namespaceRightSizingRecommendation:
          enabled: true
        virtualizationRightSizingRecommendation:
          enabled: true
      metrics:
        default:
          enabled: true

Or patch an existing CR:

oc patch MultiClusterObservability observability --type=merge -p '{
  "spec": {
    "capabilities": {
      "platform": {
        "metrics": {
          "default": {
            "enabled": true
          }
        }
      }
    }
  }
}'

With platform metrics collection enabled, the observability add-on manages the full metrics pipeline end to end. Enabling right-sizing automatically deploys three resources to each matching managed cluster:

• PrometheusRules in openshift-monitoring: Recording rules that aggregate raw platform metrics (CPU, memory, quotas, and for virtualization: KubeVirt vCPU and memory) into the acm_rs:* and acm_rs_vm:* metric families at 5-minute and 1-day granularity.

• ScrapeConfig platform-metrics-right-sizing in open-cluster-management-agent-addon: Instructs the PrometheusAgent to federate the aggregated right-sizing metrics from the platform Prometheus /federate endpoint.

• PrometheusAgent (platform-metrics-collector): Deployed by the add-on for platform metrics. It discovers the right-sizing ScrapeConfig via label selectors, scrapes the federated metrics, and remote-writes them to the hub Thanos receives endpoint.

This requires no additional configuration. The metrics appear in hub Thanos and are immediately available for the Perses dashboards. You can verify the ScrapeConfig was picked up by checking the PrometheusAgent's generated configuration on a managed cluster.

oc exec -n open-cluster-management-agent-addon \
prom-agent-platform-metrics-collector-0 -- \
cat /etc/prometheus/config_out/prometheus.env.yaml \
| grep -A10 "right-sizing"

Expected output:

- job_name: scrapeConfig/open-cluster-management-agent-addon/platform-metrics-right-sizing
metrics_path: /federate
params:
match[]:
- '{__name__="acm_rs:namespace:cpu_request_hard"}'
- '{__name__="acm_rs:namespace:cpu_request"}'
...

Without platform metrics collection

Right-sizing also works when you do not enable the platform metrics collection. In this mode, only the PrometheusRules deploy to managed clusters. The recording rules still run on each cluster's platform Prometheus and produce the acm_rs:* / acm_rs_vm:* metrics.

The existing observability pipeline automatically handles the Federation to the hub. Right-sizing metrics are pre-configured in the observability metrics allowlist (observability-metrics-allowlist), so the existing metrics collector on each managed cluster picks up the right-sizing metrics and forwards them to the hub without any additional user configuration.

Disclaimer: Since this is a developer preview, the Perses dashboards are under active development and may change in future releases. For more information and known limitations, refer to the official product documentation.

What's next

This developer preview marks the beginning of the transition from Grafana to Perses for right-sizing visualization in Red Hat Advanced Cluster Management with general availability (GA) targeted for the next release. In addition, upcoming Red Hat Advanced Cluster Management releases will introduce expanded right-sizing capabilities in developer preview, including workload- and pod-level recommendation support, helping users further optimize resource utilization across their clusters.

We value your feedback, which is crucial for shaping the future of right-sizing in Red Hat Advanced Cluster Management. Share your questions and recommendations with us using the Red Hat OpenShift feedback form.