Is it safe to host multiple PyPI Cloud servers using the same S3 backend for high availability?

[oschlueter]

The docs link to an Ansible playbook that mentions hosting multiple PyPI Cloud servers to ensure high availability of the service but the docs themselves never mention this. What I would like to do is create multiple AWS EC2 instances in different availability zones and create a single DNS record pointing to all of these instances, for example:

A 10 192.0.2.35
A 10 192.1.2.35
A 10 192.2.2.35

Would this work? Do I need to consider anything or be aware of anything?

[stevearc]

The webservers can live anywhere since they are stateless. The tricky part for multiple availability zones is that the webservers all have to be pointed at the same cache backend or they can get out of sync (e.g. a new package uploaded to one won't show up in the other). I haven't used AWS for many years now, but I think DynamoDB now has tables that span multiple availability zones. If not, you could look at this experimental support for synchronizing multiple cache backends using S3 notifications I added a while back.