[PERF] Optimize PublicAsset create by using S3 copy object api

[vttranlina]

Why?

ref: #1061

Requirement

linagora/james-project#5222

How

Introduce BlobStoreToolBox interface

public interface BlobStoreToolBox {
    Mono<BlobId> copyBlob(BlobId blobId, BucketName sourceBucket, BucketName targetBucket);
}

• The S3 implement looks like:

public class S3BlobStoreToolBox implements BlobStoreToolBox {
    private final S3AsyncClient s3Client;
    @Override
    public Mono<BlobId> copyBlob(BlobId blobId, BucketName sourceBucket, BucketName targetBucket) {
        return Mono.fromFuture(s3Client.copyObject(CopyObjectRequest.builder()
                .sourceBucket(sourceBucket.asString())
                .sourceKey(blobId.asString())
                .destinationBucket(targetBucket.asString())
                .destinationKey(blobId.asString())
                .build()))
            .publishOn(Schedulers.parallel())
            .thenReturn(blobId);
    }
}

• For another implement

public class NaiveBlobStoreToolbox implements BlobStoreToolBox {
    private final BlobStore blobStore;
    @Override
    public Mono<BlobId> copyBlob(BlobId blobId, BucketName sourceBucket, BucketName targetBucket) {
        return Mono.from(blobStore.readBytes(sourceBucket, blobId))
            .flatMap(bytes -> Mono.from(blobStore.save(targetBucket, bytes, BlobStore.StoragePolicy.LOW_COST)));
    }
}

Location: tmail-backend/blob/blob-extensions

Adapt PublicAssetSetMethod

Currently, we are using blobResolvers to resolve creationRequest.blobId. I propose that we use UploadRepository directly to retrieve UploadMetaData.

Impact scope:

• The blobId in the client request must have the prefix uploads-.
• We can add a new API Publisher<UploadMetaData> listUploads(Username user, UploadId id) to UploadRepository, or use listUploads directly and filter UploadMetaData (in case the upload size for each user is not large).
• Update the PublicAssetCreationRequest POJO by adding blobId and bucketName field, and removing content field:

case class PublicAssetCreationRequest(size: Size,
                                      contentType: ImageContentType,
                                      identityIds: Seq[IdentityId] = Seq.empty,
                                      blobId: BlobId,
                                      bucketName: BucketName)

Inject BlobStoreToolBox into PublicAssetRepository and use the copyBlob API.

Guice binding

• using S3BlobStoreToolBox for S3blob store chooser
• using NaiveBlobStoreToolbox for other remain

[chibenwa]

I think creating public asset will be very rare.
I am expecting the gains of such an optimization to be low.
Also it breaks encapsulation and hard code the use of the blobStore which in terms of design do not seem desirable.

[vttranlina]

honestly, when writing a ticket, I have a bit similar thought

[vttranlina]

any thing other idea from team? if not, I will close it