I know DNS will fall back if it can't reach a service, but would a 502 trigger that?
This has a few drawbacks like making sure your Route53 configuration is identical to your CF config, ensuring your origin servers can cope with the additional load if CF caching isn't available and the DNS propagation time required for the Name Servers to update.
During the last outage, we were able to get into the CF dashboard and simply disable the proxy which allowed our clients to access our origin servers directly but this time we can't even get into the Dashboard.
Ideally, I'd want something where if Cloudflare goes down, I don't have to change anything, but... 502 isn't going to trigger that without some work on my part.
Meh.
Backup DNS entries won't help, what you need is to use multiple DNS providers and add them all to your NS records
Since I'm mostly thinking about static sites, I could also have something local that pings the site, and if it goes down, it could update my nameservers to point elsewhere on its own.
Probably more trouble than its worth though.
I'm not sure if the browser receives all A/AAAA records from the syscall, or just one. I guess that if the browser has the whole list, and the error is in the 500 range, it could retry a different IP but I'm not sure if browsers do this.
It is possible to have multiple A/AAAA records pointing to different loadbalancers, but I don't know how browsers would deal with this.
But it does make it possible for a browser to retry a different IP if a 500 range error, or timeout occurs.
But like I said in my other post: I don't know if browsers actually do this.
If there are errors, the site would need to be modified to not respond if broken and unable to proxy to a working origin. Perhaps CF have not coded their proxies in this manor.