Let's just proxy the HTTP requests to the S3 API to the right node for the dot in question, and stream the PUT/GET directly to the filesystem. We should never buffer the files in memory, let alone in etcd.
ACs:
[x] It should be possible to upload arbitrarily large files (in particular, >2MiB files are a good test case to validate that we've avoided etcd limits).
[x] In particular, it should be possible to upload files larger than the server memory.
[x] The system should continue to work afterwards (eg, we don't break something).
Let's just proxy the HTTP requests to the S3 API to the right node for the dot in question, and stream the PUT/GET directly to the filesystem. We should never buffer the files in memory, let alone in etcd.
ACs: