APNIC’s core registry services are whois, Registration Data Access Protocol (RDAP), Resource Public Key Infrastructure (RPKI), Internet Routing Registry (IRR), and Reverse Domain Name System (rDNS). The availability for these APNIC core services for Q4 2025 is shown in Table 1, along with the previous quarters and combined for the year.
| Service | Q1 availability | Q2 availability | Q3 availability | Q4 availability | 2025 availability |
| whois and IRR | 99.991% | 99.996% | 99.999% | 99.993% | 99.995% |
| RDAP | 99.998% | 99.998% | 99.997% | 99.989% | 99.996% |
| RPKI | 99.999% | 99.983% | 99.994% | 99.984% | 99.990% |
| rDNS | 99.998% | 99.999% | 100% | 100% | 99.999% |
Refining measurement to avoid double-counting
As noted in the Q2 report, one of our key objectives for 2025 was to evolve how we calculate availability by combining two distinct vantage points:
- Third-party monitoring: External probes that send queries every minute to detect downtime
- User-perspective metrics: Analysis of access logs to calculate the ratio of successful requests to total requests (the error rate)
During Q4, the Cloudflare outage on 18 November 2025 (detailed in their post-mortem) served as a practical test for this combined methodology.
Because APNIC uses Cloudflare for load-balancing and caching for the RDAP and RPKI RRDP core services, the outage led to elevated error rates for our users. At the same time, our third-party monitoring service detected and recorded downtime for these services.
From a data perspective, this created a challenge. If we were to simply aggregate the downtime from the third-party probes and the user-perspective error rate from the access logs for the same period, we would be double-counting, and our availability figure would be lower than it should be.
Applying data exclusion for accuracy
When an outage is clearly identified and accounted for by third-party monitoring probes, we exclude that specific time window when calculating the user-perspective error rates.
In the case of the 18 November incident, we:
- Recorded the outage duration based on the monitoring probes (contributing to the 75% weighted availability)
- Excluded the outage period from the user-perspective error-rate calculation (the 25% weighted perspective)
This prevents an outage from being counted twice in the final percentage, while still ensuring the impact on users is reflected in the overall availability score.
Incident response and future resilience
As detailed in the Q3 summary, the Cloudflare event highlighted the importance of incident response and failover processes. During the peak of the 5xx errors, APNC successfully updated DNS for the RRDP service to route clients directly to our on-premises servers, bypassing the affected proxy.
In Q4 2025, user-perspective availability metrics were added to the whois and rDNS services. In 2026, these metrics will continue to be refined to provide the community with a more nuanced and accurate view of APNIC service performance. Incident response processes will also be further improved to ensure response teams have clear, well-practised procedures to restore services quickly and confidently during incidents.
The views expressed by the authors of this blog are their own and do not necessarily reflect the views of APNIC. Please note a Code of Conduct applies to this blog.