Cloudflare Exporter underreports requests compare to Cloudflare dashboard

[soontobemil]

Description

The Cloudflare exporter on prometheus is consistently reporting ~25% fewer requests compared to the Cloudflare Dashboard for the same time periods. This appears to be caused by multiple architectural issues in how the exporter queries and processes metrics.

Environment

• Cloudflare Plan: Pro/Enterprise
• Scrape Interval: 60 seconds
• Scrape Delay: 300 seconds (5 minutes)

Expected Behavior

Prometheus metrics from cloudflare_zone_requests_total should match the "Total Requests" shown in the Cloudflare Dashboard for the same time period (accounting for the 5-minute scrape delay).

Actual Behavior

Prometheus consistently shows ~25% fewer requests than the Cloudflare Dashboard.

Example:

• Cloudflare Dashboard (30-minute period): 192-215 requests
• Prometheus (same period): 150-180 requests
• Discrepancy: ~25-30% missing data

Screenshots

• Total Request of 22.03M between 9:00-9:30

• Total Request of 18.8M-ish between 9:00-9:30 (Via thanos/prometheus)
• sum(increase(cloudflare_zone_requests_total{zone="example.com"}[30m]))

*Both data was captured at 10:30 - so this should have given enough time for the scraped metrics to settle in.

Reproduction Steps

1. Deploy cloudflare-exporter with default configuration (60s interval, 300s delay)
2. Let it run for at least 1 hour
3. Compare Prometheus metrics with Cloudflare Dashboard for the same time period:
   sum(increase(cloudflare_zone_requests_total{zone="example.com"}[30m]))
4. Check Cloudflare Dashboard "Total Requests" for the same 30-minute window (accounting for 5-minute offset)
5. Observe ~20-30% discrepancy

Please advise, thank you!

[soontobemil]

Anyone prob has an idea or answer on this?

[niharikagade-dot]

Hi , I was also seeing discrepancies between the exporter’s request counts (cloudflare_zone_requests_total) and the Cloudflare dashboard “Total Requests” for the same time period.

From my testing, the Cloudflare Analytics GraphQL API supports a requestSource filter, e.g.:

httpRequestsAdaptiveGroups(
  filter: {
    datetime_geq: $start,
    datetime_lt: $end,
    requestSource: "eyeball"
  }
) {
  sum {
    count
  }
}

Using requestSource: "eyeball" ensures that only end-user (eyeball) traffic is counted. This aligns much more closely with what the Cloudflare dashboard shows and excludes internal or automated requests that the dashboard doesn’t count. This is documented in Cloudflare’s Analytics API filtering features: [https://developers.cloudflare.com/analytics/graphql-api/features/filtering/]

Would it be possible to update the exporter’s GraphQL queries to optionally include this filter? That should help reduce the ~25% gap currently seen between Prometheus metrics and dashboard totals.