When S3 launched in 2006 as AWS's very first service, external developers found it revolutionary just because "you could store a file over HTTP." But 17 years on, S3 is no longer a simple file store. As a distributed object storage system, it runs on the design principles of Dynamo (Amazon's internal key-value DB), and in 2020 it even completely overhauled its consistency model.
This article covers how S3 works internally (how data is distributed and replicated), why its consistency model only achieved strong consistency in 2020, the exact API sequence of multipart upload, the internals of versioning, and the signing mechanism behind Presigned URLs.