Cloudflare identified the root cause and have shared the following incident report with us:
Cloudflare was performing scheduled maintenance in our Sydney facility. Before Cloudflare can perform maintenance, engineers must remove the datacenter from production by disabling anycast. As part of disabling anycast, a series of flags are set which notify all products and services that the facility is going into maintenance and the datacenter should not receive traffic.
Due to a bug in the automation to disable anycast, the flags to notify our Tiered Cache and Argo products were not set accordingly. As a result, when Sydney was removed from service, any request that had Sydney as an upper tier would continue to send traffic to Sydney despite the fact that the facility was not supposed to receive traffic. This caused increases in origin connectivity failures for surrounding locations.
We’re confident that they will take some steps to mitigate this same problem in the future.