การอัปโหลดไปยัง S3 โดยตรงด้วย URL ที่ลงลายเซ็นล่วงหน้า

When a client uploads a large file, the naive design sends the bytes to your NestJS API, which then forwards them to object storage. This makes your server a bottleneck.

• Memory & CPU pressure — every upload occupies a request thread and buffers/streams through your process.
• Doubled bandwidth — bytes travel client → API → S3, so you pay for the same data twice.
• Request timeouts — load balancers (e.g. ALB, Nginx) cap request duration; multi-GB uploads stall.

The fix: let the browser upload directly to S3. Your API only issues a short-lived, signed URL that grants permission for one specific operation.

A presigned URL is a normal S3 object URL with extra query parameters that encode a temporary, cryptographically-signed grant. Anyone holding the URL can perform exactly one operation (e.g. PutObject) on exactly one key, until it expires.

• Signed with your AWS credentials, but the credentials are never exposed — only the signature is.
• Scoped to a single HTTP method, bucket, and object key.
• Has a hard expiry (seconds), after which S3 rejects it with 403.

Because S3 validates the signature itself, your API does not touch the file bytes at all.

The end-to-end flow has three actors: the browser, your NestJS API, and S3.

• 1. Request: Browser asks the API, "I want to upload avatar.png, 240 KB, image/png."
• 2. Sign: API validates the request, generates a unique key, and returns a presigned PUT URL.
• 3. Upload: Browser PUTs the raw bytes straight to that URL on S3.
• 4. Confirm: Browser tells the API the upload succeeded; API persists the key in the database.

Your API stays fast and stateless — it never proxies the payload.

Use the AWS SDK v3 modular packages. Create a single S3Client instance and share it via a NestJS provider so credentials and region are configured in one place.

Credentials come from environment variables (or, in production, an IAM role). Never hardcode them.

The @aws-sdk/s3-request-presigner package signs a command without executing it. You build a PutObjectCommand describing the target key and content type, then call getSignedUrl with an expiresIn.

• ContentType in the command is enforced: the browser must send a matching Content-Type header.
• expiresIn is in seconds — keep it short (60–300s) so leaked URLs die quickly.

Never trust the client's filename as the S3 key. A user could send ../../etc/passwd or collide with another user's file. Generate the key server-side.

• Namespace by owner: uploads/{userId}/... so authorization is easy to reason about.
• Use a random UUID to guarantee uniqueness.
• Preserve only a sanitized extension for content-type hints and tooling.

Expose a guarded endpoint that takes the file metadata, validates it with a DTO, and returns the presigned URL plus the final key. The client needs the key later to confirm or to build the public/read URL.

Authentication matters here: the signing endpoint is your access-control gate. S3 itself trusts any valid signature, so all checks (who, what size, what type) must happen before you sign.

Validate metadata before signing. Reject disallowed MIME types and oversized files at the API layer — but remember the client could lie, so this is a first line of defense, not the last.

• Whitelist contentType against an allow-list.
• Cap the declared size to fail fast on obviously huge uploads.

The DTO size check is advisory — the browser still controls how many bytes it actually PUTs. To make S3 itself reject oversized uploads, bind a content-length range into the signature.

For a single PUT, sign ContentLength so S3 enforces an exact byte count. For more flexible limits (a min/max range), use a presigned POST policy instead, which supports content-length-range conditions.

With the presigned PUT URL in hand, the browser uploads with a plain fetch. There is no SDK and no AWS credentials on the client — just the bytes and a matching Content-Type.

• The header must equal the ContentType you signed, or S3 returns 403 SignatureDoesNotMatch.
• Do not send Authorization — the signature lives in the query string.

Because S3 doesn't notify your API, the client calls back after a successful PUT so you can persist the key. For extra safety, the API can HeadObject to verify the object truly exists and check its real size/type before trusting it.

To serve the file later, either keep the bucket private and issue a presigned GET URL on demand, or (for public assets) store and return the public URL.

You learned how to offload heavy uploads to S3 using presigned URLs:

• Why: proxying bytes through the API wastes bandwidth, memory, and hits request timeouts.
• How: the API signs a short-lived PutObjectCommand with getSignedUrl; the browser PUTs straight to S3.
• Keys: always generate them server-side (UUID + namespaced by user); never trust client filenames.
• Security: the presign endpoint is the access-control gate — authenticate, whitelist MIME types, and cap size there. Use presigned POST content-length-range to let S3 enforce size.
• After upload: confirm with HeadObject, persist the key, and serve later via presigned GetObjectCommand URLs.