upload_fileobj is always mutlipart

[plotlogic-andrew]

• Async AWS SDK for Python version: 12.4.0
• Python version: 3.12
• Operating System: Fedora 36

Description

upload_fileobj always does multipart uploads. This results in:

• small uploaded file having different etags than if the standard boto3.upload_fileobj was used
• (presumably) 3 x the API calls for small files (create-upload-complete vs just upload)

What I Did

    import boto3
    import aioboto3

    file = 'test.txt'
    with open(file, "wb") as fp:
        pass
    
    bucket, key = ...

    s3_client = boto3.client('s3') 
    s3_client.upload_file(file, bucket, key + '/test-sync.txt')
        
    async with aioboto3.Session().client('s3') as s3_client:
        await s3_client.upload_file(file, bucket, key + '/test-async.txt')

The uploadfile_obj code make reference to multipart_chunksize but doesn't use it.

[terricain]

Yeah I've not looked at the S3 transfer code in years but im sure it should not multipart if the file is small enough. Will look at fixing it at some point

[plotlogic-andrew]

I'd call it a feature request more than a bug :-).

I'm happy to contribute - but after a quick look the code for upload_fileobj coming from botocore seemed opaque at best to me. I'll look again when I have more time to.

And thanks for putting this together! I've always been bothered that boto3 defaults to 10 threads to keep an upload pipe full.

[terricain]

s3.upload_fileobj now respects Config.multipart_threshold and will issue a singular s3.put_object if the file is below the threshold.