Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Migrating Spark Kinesis connector to SDK 2.x for Amazon EMR 7.0

The Amazon SDK provides a rich set of APIs and libraries to interact with Amazon cloud computing services, such as managing credentials, connecting to S3 and Kinesis services. The Spark Kinesis connector is used to consume data from Kinesis Data Streams, and the received data is transformed and processed in Spark’s execution engine. Currently this connector is built on top of 1.x of Amazon SDK and Kinesis-client-library (KCL).

As part of the Amazon SDK 2.x migration, the Spark Kinesis connector is also updated accordingly to run with the SDK 2.x. In the Amazon EMR 7.0 release, Spark contains the SDK 2.x upgrade that is not yet available in the community version of Apache Spark. If you use the Spark Kinesis connector from a release that's lower than 7.0, you must migrate your application codes to run on SDK 2.x before you can migrate to Amazon EMR 7.0.

Migration guides

This section describes the steps to migrate an application to the upgraded Spark Kinesis connector. It includes guides to migrate to the Kinesis Client Library (KCL) 2.x, Amazon credential providers, and Amazon service clients in Amazon SDK 2.x. For reference, it also includes a sample WordCount program that uses the Kinesis connector.

Migrating KCL from 1.x to 2.x

Migrating Amazon credentials providers from Amazon SDK 1.x to 2.x

Credentials providers are used to obtain Amazon credentials for interactions with Amazon. There are several interface and class changes related to the credentials providers in SDK 2.x, which can be found here. Spark Kinesis connector has defined an interface (org.apache.spark.streaming.kinesis.SparkAWSCredentials) and implementation classes that returns 1.x version of Amazon credential providers. These credentials providers are needed when initializing Kinesis clients. For instance, if you are using the method SparkAWSCredentials.provider in the applications, you would need to update codes to consume 2.x version of Amazon credential providers.

Following is an example of using the credential providers in Amazon SDK 1.x:

import org.apache.spark.streaming.kinesis.SparkAWSCredentials
import com.amazonaws.auth.AWSCredentialsProvider
 
val basicSparkCredentials = SparkAWSCredentials.builder
    .basicCredentials("accessKey", "secretKey")
    .build()
                                     
val credentialProvider = basicSparkCredentials.provider
assert(credentialProvider.isInstanceOf[AWSCredentialsProvider], "Type should be AWSCredentialsProvider")

Migrating Amazon service clients from Amazon SDK 1.x to 2.x

Amazon service clients have different package names in 2.x (i.e. software.amazon.awssdk). whereas the SDK 1.x uses com.amazonaws. For more information about the client changes, see here. If you are using these service clients in the codes, you would need to migrate the clients accordingly.

Following is an example of creating a client in SDK 1.x:

import com.amazonaws.services.dynamodbv2.AmazonDynamoDBClient
import com.amazonaws.services.dynamodbv2.document.DynamoDB
 
AmazonDynamoDB ddbClient = AmazonDynamoDBClientBuilder.defaultClient();
AmazonDynamoDBClient ddbClient = new AmazonDynamoDBClient();

Code examples for streaming applications

import java.net.URI
import software.amazon.awssdk.auth.credentials.DefaultCredentialsProvider
import software.amazon.awssdk.http.apache.ApacheHttpClient
import software.amazon.awssdk.services.kinesis.KinesisClient
import software.amazon.awssdk.services.kinesis.model.DescribeStreamRequest
import software.amazon.awssdk.regions.Region
import software.amazon.kinesis.metrics.{MetricsLevel, MetricsUtil}
 
import org.apache.spark.SparkConf
import org.apache.spark.storage.StorageLevel
import org.apache.spark.streaming.{Milliseconds, StreamingContext}
import org.apache.spark.streaming.dstream.DStream.toPairDStreamFunctions
import org.apache.spark.streaming.kinesis.KinesisInitialPositions.Latest
import org.apache.spark.streaming.kinesis.KinesisInputDStream
 
 
object KinesisWordCountASLSDKV2 {
 
  def main(args: Array[String]): Unit = {
    val appName = "demo-app"
    val streamName = "demo-kinesis-test"
    val endpointUrl = "https://kinesis.us-west-2.amazonaws.com"
    val regionName = "us-west-2"
 
    // Determine the number of shards from the stream using the low-level Kinesis Client
    // from the AWS Java SDK.
    val credentialsProvider = DefaultCredentialsProvider.create
    require(credentialsProvider.resolveCredentials() != null,
      "No AWS credentials found. Please specify credentials using one of the methods specified " +
        "in https://docs.aws.amazon.com/sdk-for-java/latest/developer-guide/credentials.html")
    val kinesisClient = KinesisClient.builder()
      .credentialsProvider(credentialsProvider)
      .region(Region.US_WEST_2)
      .endpointOverride(URI.create(endpointUrl))
      .httpClientBuilder(ApacheHttpClient.builder())
      .build()
    val describeStreamRequest = DescribeStreamRequest.builder()
      .streamName(streamName)
      .build()
    val numShards = kinesisClient.describeStream(describeStreamRequest)
      .streamDescription
      .shards
      .size
 
 
    // In this example, we are going to create 1 Kinesis Receiver/input DStream for each shard.
    // This is not a necessity; if there are less receivers/DStreams than the number of shards,
    // then the shards will be automatically distributed among the receivers and each receiver
    // will receive data from multiple shards.
    val numStreams = numShards
 
    // Spark Streaming batch interval
    val batchInterval = Milliseconds(2000)
 
    // Kinesis checkpoint interval is the interval at which the DynamoDB is updated with information
    // on sequence number of records that have been received. Same as batchInterval for this
    // example.
    val kinesisCheckpointInterval = batchInterval
 
    // Setup the SparkConfig and StreamingContext
    val sparkConfig = new SparkConf().setAppName("KinesisWordCountASLSDKV2")
    val ssc = new StreamingContext(sparkConfig, batchInterval)
 
    // Create the Kinesis DStreams
    val kinesisStreams = (0 until numStreams).map { i =>
      KinesisInputDStream.builder
        .streamingContext(ssc)
        .streamName(streamName)
        .endpointUrl(endpointUrl)
        .regionName(regionName)
        .initialPosition(new Latest())
        .checkpointAppName(appName)
        .checkpointInterval(kinesisCheckpointInterval)
        .storageLevel(StorageLevel.MEMORY_AND_DISK_2)
        .metricsLevel(MetricsLevel.DETAILED)
        .metricsEnabledDimensions(Set(MetricsUtil.OPERATION_DIMENSION_NAME, MetricsUtil.SHARD_ID_DIMENSION_NAME))
        .build()
    }
 
    // Union all the streams
    val unionStreams = ssc.union(kinesisStreams)
 
    // Convert each line of Array[Byte] to String, and split into words
    val words = unionStreams.flatMap(byteArray => new String(byteArray).split(" "))
 
    // Map each word to a (word, 1) tuple so we can reduce by key to count the words
    val wordCounts = words.map(word => (word, 1)).reduceByKey(_ + _)
 
    // Print the first 10 wordCounts
    wordCounts.print()
 
    // Start the streaming context and await termination
    ssc.start()
    ssc.awaitTermination()
  }
}

Considerations when using the upgraded Spark Kinesis connector