The useful observability change in VMware Cloud Foundation 9.1 is not another dashboard. Broadcom says VCF Operations can collect real-time metrics across vCenter, ESXi, NSX, and vSAN, query them with Prometheus Query Language, expose them through standard APIs, and pull them into third-party dashboards such as Grafana. That creates an opportunity to connect the virtualization layer to the same operational view used for applications, networks, and services.
Why it matters in production
Cross-layer visibility matters because many infrastructure incidents do not respect product boundaries. A slow application can begin with storage latency, host contention, an NSX path, or a vCenter control-plane problem. When each component keeps its own timeline and query language, the team spends the first part of an incident aligning clocks and screenshots. Granular metrics from the core VCF components are most valuable when they shorten that correlation step.
PromQL makes the data more portable, but it does not create an observability strategy by itself. Teams still have to define metric ownership, labels, cardinality limits, retention, scrape or query intervals, and the alert rules that deserve an operational response. Exporting every available series to Grafana can reproduce the same silo in a different interface while increasing storage cost and noise.
The right design separates three needs. VCF-native views should support platform specialists during detailed diagnosis. A central observability platform should receive the small set of infrastructure signals needed to correlate service health across domains. Service dashboards should translate those signals into workload impact. PromQL and standard APIs provide the connection points, but the team must decide which data crosses each boundary and who responds when it changes.
A controlled rollout should start with one production service and its dependency map. Select a few ESXi, vSAN, NSX, and vCenter indicators that explain known failure modes, export them to the existing monitoring platform, and compare timestamps with application telemetry during a test or real incident. Validate query cost, data freshness, label stability, access control, and behavior when VCF Operations itself is impaired before treating the integration as a new source of truth.
Practical takeaway
The practical conclusion is that VCF 9.1 can reduce monitoring isolation only if teams use its open query surface deliberately. The goal is not more VMware metrics. It is a shorter path from a user-visible symptom to the responsible infrastructure layer, with one timeline, clear ownership, and enough retained evidence to explain what happened after service is restored.