Effective today the R2 Metrics page includes a toggle that switches the Class A and Class B operation charts from a simple request count to a requests‑per‑second (RPS) view. The change is limited to the visualisation layer; the underlying data collection remains the same.
New toggle for operation charts
On the Metrics page each chart now offers a Request count and a Requests/second mode. Selecting Request count displays the total number of operations recorded in each time bucket, which is useful for estimating billable usage. Selecting Requests/second converts the same data into a rate, showing how many operations occur per second over the same interval. The toggle applies independently to both Class A and Class B charts.
Why the rate view matters
For AI workloads, data pipelines, or any service that reads or writes large objects, understanding the peak operation rate can expose performance bottlenecks that raw counts hide. A sudden spike in RPS may indicate a burst of inference requests, a batch job, or an unexpected client behavior. Conversely, a low RPS with a high cumulative count suggests steady, predictable traffic. Both patterns have different implications for scaling, throttling, and cost forecasting.
Operational considerations
- Integrate the RPS chart into existing monitoring dashboards to set alerts on abnormal spikes or sustained high rates.
- Correlate RPS data with downstream services (e.g., compute or database) to verify that downstream capacity can handle peak loads.
- Use the request‑count view when reviewing monthly billing statements, as it directly reflects the units that Cloudflare charges.
- Periodically compare the two views to detect changes in traffic patterns that may warrant architectural adjustments, such as adding caching layers or adjusting client retry logic.
Related CloudNinjas coverage: hands-on guides.
What This Means For Practitioners
Practitioners should start using the Requests/second view to define realistic scaling thresholds and to spot traffic anomalies early. Pair the rate data with existing alerting pipelines, and keep the Request count view for cost analysis. Regularly review both metrics after any deployment that could affect I/O patterns, and adjust capacity or throttling policies accordingly.
