AWS S3 with Java: Functional Algebra-based API using Scala


In this tutorial we will see a functional, algebra-based API design on top of AWS S3 Java API using Scala. To achieve this, we will be performing basic S3 operations:

  • Create/delete bucket
  • Upload/download/delete object
  • Versioning

We will also try to isolate our impure code.

Because Scala incorporates capabilities of the FP language, we will build an algebra-based API, which means it is a clear (behaviour-driven), composable (based on Algebraic Data Types), and verifiable (based on Algebraic laws) API. For that we will use Xor and Kleisli monads.

We will use Kleisli monad to sequence operations and Xor monad to do a basic failover. Xor also shows in the signature that the result of the computation can be AmazonServiceException or a child of it. We will get Xor from Try using implicit conversion. Try is used to try/catch exceptions.

type S3OperationResult[T] = AmazonClientException Xor T

implicit def toXor[T](tr: Try[T]): S3OperationResult[T] = {
  tr match {
    case Success(bucket) => Xor.Right(bucket)
    case Failure(exc: AmazonClientException) => Xor.Left(exc)

Other Details

Log File is a temporary created file with “Hello AWS S3!” text.

val logFile = {
  val logFilePrefix = "tmp-"
  val logFileSuffix = ".log"
  val logFile: File = File.createTempFile(logFilePrefix, logFileSuffix)

  val writer = new PrintWriter(logFile)
  writer.write("Hello AWS S3!")




val s3client: AmazonS3 = new AmazonS3Client(new ProfileCredentialsProvider())

Create Bucket

Create bucket and enable versioning.

val createBucket = Kleisli[S3OperationResult, String, Bucket] {
bucketName: String =>
  Try {
    println(s"S3 create bucket: ${bucketName}")
    val bucket = s3client.createBucket(
      new CreateBucketRequest(bucketName))
    s3client.setBucketVersioningConfiguration(new SetBucketVersioningConfigurationRequest(
      bucketName, new BucketVersioningConfiguration(BucketVersioningConfiguration.ENABLED)))

Upload Log File

val uploadObject = { logFile: File =>
  Kleisli[S3OperationResult, String, PutObjectResult] { bucketName: String =>
    Try {
      println(s"S3 upload log file: ${logFile.getName}")
        new PutObjectRequest(bucketName, logFile.getName, logFile))

Delete Log File

If versioning is enabled then delete will put a Delete Marker on the last version of the file.

val deleteObject = { logFileName: String =>
Kleisli[S3OperationResult, String, Unit] { bucketName: String =>
Try {
println(s"S3 delete log file: ${logFileName}")
new DeleteObjectRequest(bucketName, logFileName))

Download Log File

If versioning is enabled then specific version of the file can be downloaded (even after deletion).

val downloadObject = { versionId: String =>
  Kleisli[S3OperationResult, String, S3Object] { bucketName: String =>
    Try {
      println(s"S3 get object with version: ${versionId}")
        new GetObjectRequest(bucketName, logFile.getName)

Empty And Delete Bucket

Bucket can be deleted only if it’s empty (if versioning is enabled then all file versions should be deleted as well).

val emptyAndDeleteBucket = Kleisli[S3OperationResult, String, Unit] {
bucketName: String =>
  Try {
    println(s"S3 delete bucket: ${bucketName}")
    val versionListing = s3client.listVersions(
      new ListVersionsRequest()

    import scala.collection.JavaConverters._

    for (s3VersionSummary <- versionListing.getVersionSummaries.iterator().asScala) {
      s3client.deleteVersion(bucketName, s3VersionSummary.getKey(), s3VersionSummary.getVersionId())

Sequence of Operations

Since Kleisli is a Monad, we can put our operations in a sequence.

val test = for {
  bucket <- createBucket
  putObjectResult <- uploadObject(logFile)
  _ <- deleteObject(logFile.getName)
  s3Obj <- downloadObject(putObjectResult.getVersionId)
  _ <- emptyAndDeleteBucket
} yield ()

Sequence of Operations Application

All operations need a bucket name, that’s why our Kleisli arrows take it as input (aligned on it).

val bucketName = "logs-" + UUID.randomUUID
val res = test(bucketName)

Exceptions and Recovery

AWS Api can generally throw two types of Exceptions:

  1. AmazonClientException
  2. AmazonServiceException

Test sequence will exit whenever one of the operations breaks and then prints the detailed issue description.

res recover {
  case exc: AmazonServiceException =>
  case exc: AmazonClientException =>



  1. AWS credentials file with permissions to access S3.
  2. Docker
  3. Java 8
  4. SBT


You can find an SRC code on [Bitbucket] (
Since S3 App is running in a Docker Container, you need to build a Docker Image.

Here’s how to create a Dockerfile (creds system property defines path to your AWS credentials file):

sbt -Dcreds="path/to/credentials" clean docker

Now to build a Docker Image from the Dockerfile (this will take some time to download the Java image from Docker Hub):

$ cd target/docker
$ docker build .

To check if image is created (lists images with the label aws.s3 that has a value s3):

docker images --filter "label=aws.api=s3"

To run S3App in Docker Container:

docker run default/s3


Firstly we saw how to use an AWS S3 Java API with Scala programming language. We implemented creation, deletion, and emptying of the bucket. Object upload and download was also described in detail.

Secondly we saw how Xor monad can be used to encapsulate an error effect (Left indicates the problem and Right indicates the success). And we also saw how Kleisli monad is used to delay the moment of the function sequence execution before some configuration is passed (bucketName in our case or in more complex cases - db/connection configuration). Here, the nature of Kleisli is very similar to Reader monad (Reader monad is basically represented through Kleisli).

Now that you’ve seen the advantages of using Scala (clear, composable, and verifiable), especially its functional capabilities, you should now be able to use Xor and Kleisli as base for Algebra-based API.