Surely the repeated logins can't be helping the situation. I suppose it is entirely auth related across all Meta products. The repeated strain could pose a cold start problem for example.
If you're okay with load balancing DNS queries across multiple providers you could do 2x Cloudflare primary NS, 2x GCP for example where each provider is in sync with each other.
If not, a manual swap at the registrar level would be good enough.
I should also mention this approach sort of breaks with Cloudflare's proxied records which dynamically assign anycast IPs for records placed on their CDN. So if using this approach the failover NS provider would probably need to also use a different CDN, preferably one that just gives you a CNAME.