You can definitely get read errors and your application should be aware of this and handle this use case. Amazon's own S3 is not immune to this and after years

You can definitely get read errors and your application should be aware of this and handle this use case. Amazon's own S3 is not immune to this and after years of running spark jobs which shard output across thousands of files on s3 (and load from the same) you'll see these underlying http errors on a daily scale even intra-region. Even if you're doing multi region s3 replication you'll run into this for external clients semi-occasionally.