How would you handle large file uploads to object storage from clients?

Question

Accepted Answer

The key principle is to keep large file bytes out of your application servers. Proxying a 2 gigabyte upload eats their bandwidth, memory and connections, and becomes a bottleneck the moment you have real volume. Instead I have clients upload directly to object storage using a pre-signed URL. My application holds the credentials and generates a time-limited URL, usually valid for about 15 minutes, that grants permission to upload one specific object, and hands that to the client, who uploads straight to the store. The bytes never touch my app, and the client never gets broad storage credentials, just a narrow expiring permission for that one key. For large files I use multipart upload. The file is split into parts, commonly 8 or 16 megabytes each, uploaded in parallel, which is faster and resumable, so if part 37 fails on a flaky connection it is retried on its own rather than restarting the whole 2 gigabyte transfer. On S3 multipart is required past 5 gigabytes anyway. I constrain the pre-signed URL to limit risk, a short expiry, a maximum content length, an allowed content type, so it cannot be abused as open storage. Then I wire the app to react to completion, usually an event notification from the store, to validate the object, scan it if needed, and kick off processing. So it is direct to storage via pre-signed URLs, multipart for large and resumable transfers, constrained and validated, with the app orchestrating rather than carrying the bytes.

Answer

What is a pre-signed URL and why is it safer than putting credentials on the client?

A time-limited, scoped URL granting permission to do one operation on one object. The client never gets broad storage credentials, just a narrow, expiring grant, so a leaked URL exposes only that single object for a short window.

Answer

Why use multipart upload for large files?

It splits the file into parts uploaded in parallel (faster) and resumable, a failed part is retried alone instead of restarting the whole upload. Essential for large files over unreliable connections.

How would you handle large file uploads to object storage from clients?

How would you handle large file uploads to object storage from clients?

The follow-ups they ask next

What the interviewer is listening for

What sinks the answer

If you genuinely do not know

Keep going with storage

Knowing the answer is not the same as recalling it under pressure