fs2-aws

fs2 Streaming utilities for interacting with AWS

Scope of the project

fs2-aws provides an fs2 interface to AWS services

The design goals are the same as fs2:

compositionality, expressiveness, resource safety, and speed

Using:

Find the latest release version and add the following dependency:

libraryDependencies +=  "io.laserdisc" %% "fs2-aws" % "VERSION"

S3

The module fs2-aws-s3 provides a purely functional API to operate with the AWS-S3 API. It defines four functions:

trait S3[F[_]] {
  def delete(bucket: BucketName, key: FileKey): F[Unit]
  def uploadFile(bucket: BucketName, key: FileKey): Pipe[F, Byte, ETag]
  def uploadFileMultipart(bucket: BucketName, key: FileKey, partSize: PartSizeMB): Pipe[F, Byte, ETag]
  def readFile(bucket: BucketName, key: FileKey): Stream[F, Byte]
  def readFileMultipart(bucket: BucketName, key: FileKey, partSize: PartSizeMB): Stream[F, Byte]
}

You can find out more in the scaladocs for each function, but as a rule of thumb for:

• Small files: use readFile and uploadFile.
• Big files: use readFileMultipart and uploadFileMultipart.

You can also combine them as you see fit. For example, use uploadFileMultipart and then read it in one shot using readFile.

Getting started with the S3 module

In order to create an instance of S3 we need to first create an S3Client, as well as a cats.effect.Blocker. Here's an example of the former:

import cats.effect._
import java.net.URI
import software.amazon.awssdk.services.s3.S3Client

val mkS3Client: Resource[IO, S3Client] =
  Resource.fromAutoCloseable(
    IO(S3Client.builder().endpointOverride(URI.create("http://localhost:9000")).build())
  )

A Blocker can be easily created using its apply method and then share it. You should only create a single instance. Now we can create our S3[IO] instance:

import fs2.aws.s3._

S3.create[IO](client, blocker).flatMap { s3 =>
  // do stuff with s3 here (or just share it with other functions)
}

Create it once and share it as an argument, as any other resource.

Reading a file from S3

The simple way:

s3.readFile(BucketName("test"), FileKey("foo"))
  .through(fs2.text.utf8Decode)
  .through(fs2.text.lines)
  .evalMap(line => IO(println(line)))

The streaming way in a multipart fashion (part size is indicated in MBs and must be 5 or higher):

s3.readFileMultipart(BucketName("test"), FileKey("foo"), partSize = 5)
  .through(fs2.text.utf8Decode)
  .through(fs2.text.lines)
  .evalMap(line => IO(println(line)))

Writing to a file in S3

The simple way:

Stream.emits("test data".getBytes("UTF-8"))
  .through(s3.uploadFile(BucketName("foo"), FileKey("bar"), partSize = 5))
  .evalMap(t => IO(println(s"eTag: $t")))

The streaming way in a multipart fashion. Again, part size is indicated in MBs and must be 5 or higher.

Stream.emits("test data".getBytes("UTF-8"))
  .through(s3.uploadFileMultipart(BucketName("foo"), FileKey("bar"), partSize = 5))
  .evalMap(t => IO(println(s"eTag: $t")))

Deleting a file in S3

There is a simple function to delete a file.

s3.delete(BucketName("foo"), FileKey("bar"))

Kinesis

Streaming records from Kinesis with KCL

Example using IO for effects (any monad F <: ConcurrentEffect can be used):

val stream: Stream[IO, CommittableRecord] = readFromKinesisStream[IO]("appName", "streamName")

There are a number of other stream constructors available where you can provide more specific configuration for the KCL worker.

Testing

TODO: Implement better test consumer

For now, you can stubbed CommitableRecord and create a fs2.Stream to emit these records:

val record = new Record()
  .withApproximateArrivalTimestamp(new Date())
  .withEncryptionType("encryption")
  .withPartitionKey("partitionKey")
  .withSequenceNumber("sequenceNum")
  .withData(ByteBuffer.wrap("test".getBytes))

val testRecord = CommittableRecord(
  "shardId0",
  mock[ExtendedSequenceNumber],
  0L,
  record,
  mock[RecordProcessor],
  mock[IRecordProcessorCheckpointer])

Checkpointing records

Records must be checkpointed in Kinesis to keep track of which messages each consumer has received. Checkpointing a record in the KCL will automatically checkpoint all records upto that record. To checkpoint records, a Pipe and Sink are available. To help distinguish whether a record has been checkpointed or not, a CommittableRecord class exists to denote a record that hasn't been checkpointed, while the base Record class denotes a commited record.

readFromKinesisStream[IO]("appName", "streamName")
  .through(someProcessingPipeline)
  .to(checkpointRecords_[IO]())

Publishing records to Kinesis with KPL

A Pipe and Sink allow for writing a stream of tuple2 (paritionKey, ByteBuffer) to a Kinesis stream.

Example:

Stream("testData")
  .map { d => ("partitionKey", ByteBuffer.wrap(d.getBytes))}
  .to(writeToKinesis_[IO]("streamName"))

AWS credential chain and region can be configured by overriding the respective fields in the KinesisProducerClient parameter to writeToKinesis. Defaults to using the default AWS credentials chain and us-east-1 for region.

Kinesis Firehose

TODO: Stream get data, Stream send data

SQS

Example

implicit val messageDecoder: Message => Either[Throwable, Quote] = { sqs_msg =>
    io.circe.parser.decode[Quote](sqs_msg.asInstanceOf[TextMessage].getText)
}
fs2.aws
      .sqsStream[IO, Quote](
        sqsConfig,
        (config, callback) => SQSConsumerBuilder(config, callback))
      .through(...)
      .compile
      .drain
      .as(ExitCode.Success)

Testing

//create stream for testing
def stream(deferedListener: Deferred[IO, MessageListener]) =
            aws.testkit
              .sqsStream[IO, Quote](deferedListener)
              .through(...)
              .take(2)
              .compile
              .toList

//create the program for testing the stream
import io.circe.syntax._
import io.circe.generic.auto._
val quote = Quote(...)
val program : IO[List[(Quote, MessageListener)]] = for {
            d <- Deferred[IO, MessageListener]
            r <- IO.racePair(stream(d), d.get).flatMap {
              case Right((streamFiber, listener)) =>
                //simulate SQS stream fan-in here
                listener.onMessage(new SQSTextMessage(Printer.noSpaces.pretty(quote.asJson)))
                streamFiber.join
              case _ => IO(Nil)
            }
          } yield r

//Assert results
val result = program
            .unsafeRunSync()
result should be(...)

TODO: Stream send SQS messages