[feature] Add caching mechanism

[jszendre]

At Intuit we are considering adding independent s3 proxy servers for various AWS roles. One of the use cases would be for fielding requests from Spark's S3A connector and returning a more consistent view of the data after create / rename / delete / update operations.

Could we add a pattern for users to specify a caching mechanism? A user could specify a struct that implements an interface for interacting with the cache. One example cache could be with a minio backend.

Thanks

[jawher]

Hi @jszendre,

I'm not sure I understood your request.

But here's, from my understanding, how I see it:

This repository is a proxy which makes it possible for apps to interact with S3 without needing the AWS SDK nor any credentials.

Here's a diagram showcasing how this works in the Mirakl (my company) context:

sequenceDiagram
	participant app as your app
	participant lib as lib-xfiles
	participant xfiles as s3proxy
	participant s3 as S3 Bucket

	app ->> lib: XFilesService.fetchBlob(bucket, key)
	lib ->> xfiles: GET /api/v1/presigned/url/<bucket>/<key>
	xfiles -->> lib: 200 OK {url: "https://google/...."}
	lib ->> s3: GET <url>
	s3 -->> lib: OK <data>
	lib -->> app: OK <data>

As you can see, this is a custom flow we "invented", and we also have an in-house library lib-xfiles to handle this flow.

However, in your use case, i.e. Spark S3 Adapter: this uses the AWS SDK directly to interact with the S3 API. And as such, it cannot transparently use s3proxy.

But maybe I misunderstood your request entirely, in which case, do no hesitate to explain it more plainly ?

Thanks.