Investigating Persistent GeoIP Database Discrepancies Across Multiple CDN PoPs and Public IP Resolvers

Hey everyone,

Building on the recent discussion about inconsistent IP geolocation, I'm facing a more complex issue with our SaaS infrastructure, particularly concerning our CDN setup.

We're using a major CDN provider, and while the primary goal is global distribution, the GeoIP database discrepancies we're observing for our edge IPs are causing headaches. For a single public IP address served by our CDN, different public IP lookup tools (and even programmatic API calls to various GeoIP providers) are returning wildly different physical locations. This isn't just a minor city difference; we're talking continent-level discrepancies sometimes.

Here's what I've tried so far:

• DNS Verification: Confirmed our domain is resolving to the CDN's Anycast IPs. We've checked DNS propagation from multiple global locations.
• CDN PoP Logic: Reviewed CDN configuration to understand their PoP selection algorithms. It seems standard, based on lowest latency, but the reported IP locations don't always align with the expected PoP.
• Multiple GeoIP Sources: Used MaxMind GeoLite2, IPinfo.io, and a custom database from a premium provider. All show variations for the same IP.
• Traceroute/MTR Analysis: Ran traceroutes from various vantage points (using services like RIPE Atlas, Catchpoint) to our CDN edge IPs. The network paths do generally terminate at the expected regional PoP, yet the GeoIP data for the final hop IP often points elsewhere.
• BGP Routing: Looked at BGP announcements for the CDN's IP ranges. They seem properly announced globally, but again, the GeoIP mapping doesn't always reflect the BGP origin or expected peering location.

The most frustrating part is when I query an IP that's clearly serving from a European PoP (based on traceroute and CDN logs), but a GeoIP tool reports it in North America. Here's a simplified example of the conflicting output:

# Querying CDN Edge IP: 192.0.2.1
# Using IPinfo.io
{
  "ip": "192.0.2.1",
  "city": "Frankfurt",
  "region": "Hesse",
  "country": "DE",
  "loc": "50.1109,8.6821",
  "org": "ASxxxx CDN Provider Inc."
}

# Using MaxMind GeoLite2 (local lookup)
{
  "ip_address": "192.0.2.1",
  "country": { "iso_code": "US", "name": "United States" },
  "city": { "name": "Ashburn" },
  "location": { "latitude": 39.0437, "longitude": -77.4875 }
}

# Using Another Public IP Tool (e.g., ip-api.com)
{
  "status": "success",
  "country": "Germany",
  "countryCode": "DE",
  "region": "HE",
  "regionName": "Hesse",
  "city": "Frankfurt am Main",
  "lat": 50.1188,
  "lon": 8.6843,
  "isp": "CDN Provider Inc."
}

This level of inconsistency makes it hard to accurately segment users by region or even troubleshoot latency issues effectively. Is this a common problem with Anycast IPs and certain GeoIP providers, or am I missing a fundamental aspect of how these databases are updated and maintained for dynamic CDN infrastructures?

Anyone faced this before?