将 CloudFormation 与 Managed Service for Apache Flink 结合使用 - Managed Service for Apache Flink
Amazon Web Services 文档中描述的 Amazon Web Services 服务或功能可能因区域而异。要查看适用于中国区域的差异,请参阅 中国的 Amazon Web Services 服务入门 (PDF)

Amazon Managed Service for Apache Flink 之前称为 Amazon Kinesis Data Analytics for Apache Flink。

本文属于机器翻译版本。若本译文内容与英语原文存在差异,则一律以英文原文为准。

将 CloudFormation 与 Managed Service for Apache Flink 结合使用

以下练习演示如何启动通过在同一堆栈中Amazon CloudFormation使用 Lambda 函数创建的 Flink 应用程序。

开始前的准备工作

在开始本练习之前,请按照使用 Amazon CloudFormation AWS:: KinesisAnalytics:: Application 创建 Flink 应用程序的步骤进行操作。

编写 Lambda 函数

要在创建或更新后启动 Flink 应用程序,我们使用 kinesisanalyticsv2 启动应用程序 API。Flink 应用程序创建后,该调用将由Amazon CloudFormation事件触发。在本练习的后面部分,我们将讨论如何设置堆栈以触发 Lambda 函数,但首先我们将重点介绍 Lambda 函数声明及其代码。我们在这个例子中使用Python3.8运行时。

StartApplicationLambda: Type: AWS::Lambda::Function DependsOn: StartApplicationLambdaRole Properties: Description: Starts an application when invoked. Runtime: python3.8 Role: !GetAtt StartApplicationLambdaRole.Arn Handler: index.lambda_handler Timeout: 30 Code: ZipFile: | import logging import cfnresponse import boto3 logger = logging.getLogger() logger.setLevel(logging.INFO) def lambda_handler(event, context): logger.info('Incoming CFN event {}'.format(event)) try: application_name = event['ResourceProperties']['ApplicationName'] # filter out events other than Create or Update, # you can also omit Update in order to start an application on Create only. if event['RequestType'] not in ["Create", "Update"]: logger.info('No-op for Application {} because CFN RequestType {} is filtered'.format(application_name, event['RequestType'])) cfnresponse.send(event, context, cfnresponse.SUCCESS, {}) return # use kinesisanalyticsv2 API to start an application. client_kda = boto3.client('kinesisanalyticsv2', region_name=event['ResourceProperties']['Region']) # get application status. describe_response = client_kda.describe_application(ApplicationName=application_name) application_status = describe_response['ApplicationDetail']['ApplicationStatus'] # an application can be started from 'READY' status only. if application_status != 'READY': logger.info('No-op for Application {} because ApplicationStatus {} is filtered'.format(application_name, application_status)) cfnresponse.send(event, context, cfnresponse.SUCCESS, {}) return # create RunConfiguration. run_configuration = { 'ApplicationRestoreConfiguration': { 'ApplicationRestoreType': 'RESTORE_FROM_LATEST_SNAPSHOT', } } logger.info('RunConfiguration for Application {}: {}'.format(application_name, run_configuration)) # this call doesn't wait for an application to transfer to 'RUNNING' state. client_kda.start_application(ApplicationName=application_name, RunConfiguration=run_configuration) logger.info('Started Application: {}'.format(application_name)) cfnresponse.send(event, context, cfnresponse.SUCCESS, {}) except Exception as err: logger.error(err) cfnresponse.send(event,context, cfnresponse.FAILED, {"Data": str(err)})

在前面的代码中,Lambda 将处理传入Amazon CloudFormation的事件,过滤掉除此之外的所有内容UpdateCreate然后获取应用程序状态,如果状态为READY,则启动它。为了获取应用程序状态,您需要创建 Lambda 角色,如下所示:

创建 Lambda 角色

您为 Lambda 创建一个角色以成功地与应用程序 “通信” 并写入日志。此角色将使用默认托管策略,但您可能需要使用自定义策略来缩小其范围。

StartApplicationLambdaRole: Type: AWS::IAM::Role DependsOn: TestFlinkApplication Properties: Description: A role for lambda to use while interacting with an application. AssumeRolePolicyDocument: Version: '2012-10-17' Statement: - Effect: Allow Principal: Service: - lambda.amazonaws.com Action: - sts:AssumeRole ManagedPolicyArns: - arn:aws:iam::aws:policy/Amazonmanaged-flinkFullAccess - arn:aws:iam::aws:policy/CloudWatchLogsFullAccess Path: /

请注意,Lambda 资源将在同一个堆栈中创建 Flink 应用程序之后创建,因为它们依赖于该堆栈。

调用 Lambda 函数

现在剩下的就是调用 Lambda 函数。这是使用自定义资源完成的。

StartApplicationLambdaInvoke: Description: Invokes StartApplicationLambda to start an application. Type: AWS::CloudFormation::CustomResource DependsOn: StartApplicationLambda Version: "1.0" Properties: ServiceToken: !GetAtt StartApplicationLambda.Arn Region: !Ref AWS::Region ApplicationName: !Ref TestFlinkApplication

这就是使用 Lambda 启动您的 Flink 应用程序所需要的全部。现在,您可以创建自己的堆栈了,或者使用下面的完整示例来看看所有这些步骤在实践中是如何工作的。

以下示例是上述步骤的稍微扩展版本,并通过模板参数进行了额外的RunConfiguration调整。这是一个可供您尝试的工作堆栈。请务必阅读随附的注意事项:

stack.yaml

Description: 'kinesisanalyticsv2 CloudFormation Test Application' Parameters: ApplicationRestoreType: Description: ApplicationRestoreConfiguration option, can be SKIP_RESTORE_FROM_SNAPSHOT, RESTORE_FROM_LATEST_SNAPSHOT or RESTORE_FROM_CUSTOM_SNAPSHOT. Type: String Default: SKIP_RESTORE_FROM_SNAPSHOT AllowedValues: [ SKIP_RESTORE_FROM_SNAPSHOT, RESTORE_FROM_LATEST_SNAPSHOT, RESTORE_FROM_CUSTOM_SNAPSHOT ] SnapshotName: Description: ApplicationRestoreConfiguration option, name of a snapshot to restore to, used with RESTORE_FROM_CUSTOM_SNAPSHOT ApplicationRestoreType. Type: String Default: '' AllowNonRestoredState: Description: FlinkRunConfiguration option, can be true or false. Default: true Type: String AllowedValues: [ true, false ] CodeContentBucketArn: Description: ARN of a bucket with application code. Type: String CodeContentFileKey: Description: A jar filename with an application code inside a bucket. Type: String Conditions: IsSnapshotNameEmpty: !Equals [ !Ref SnapshotName, '' ] Resources: TestServiceExecutionRole: Type: AWS::IAM::Role Properties: AssumeRolePolicyDocument: Version: '2012-10-17' Statement: - Effect: Allow Principal: Service: - kinesisanlaytics.amazonaws.com Action: sts:AssumeRole ManagedPolicyArns: - arn:aws:iam::aws:policy/AmazonKinesisFullAccess - arn:aws:iam::aws:policy/AmazonS3FullAccess Path: / InputKinesisStream: Type: AWS::Kinesis::Stream Properties: ShardCount: 1 OutputKinesisStream: Type: AWS::Kinesis::Stream Properties: ShardCount: 1 TestFlinkApplication: Type: 'AWS::kinesisanalyticsv2::Application' Properties: ApplicationName: 'CFNTestFlinkApplication' ApplicationDescription: 'Test Flink Application' RuntimeEnvironment: 'FLINK-1_15' ServiceExecutionRole: !GetAtt TestServiceExecutionRole.Arn ApplicationConfiguration: EnvironmentProperties: PropertyGroups: - PropertyGroupId: 'KinesisStreams' PropertyMap: INPUT_STREAM_NAME: !Ref InputKinesisStream OUTPUT_STREAM_NAME: !Ref OutputKinesisStream AWS_REGION: !Ref AWS::Region FlinkApplicationConfiguration: CheckpointConfiguration: ConfigurationType: 'CUSTOM' CheckpointingEnabled: True CheckpointInterval: 1500 MinPauseBetweenCheckpoints: 500 MonitoringConfiguration: ConfigurationType: 'CUSTOM' MetricsLevel: 'APPLICATION' LogLevel: 'INFO' ParallelismConfiguration: ConfigurationType: 'CUSTOM' Parallelism: 1 ParallelismPerKPU: 1 AutoScalingEnabled: True ApplicationSnapshotConfiguration: SnapshotsEnabled: True ApplicationCodeConfiguration: CodeContent: S3ContentLocation: BucketARN: !Ref CodeContentBucketArn FileKey: !Ref CodeContentFileKey CodeContentType: 'ZIPFILE' StartApplicationLambdaRole: Type: AWS::IAM::Role DependsOn: TestFlinkApplication Properties: Description: A role for lambda to use while interacting with an application. AssumeRolePolicyDocument: Version: '2012-10-17' Statement: - Effect: Allow Principal: Service: - lambda.amazonaws.com Action: - sts:AssumeRole ManagedPolicyArns: - arn:aws:iam::aws:policy/Amazonmanaged-flinkFullAccess - arn:aws:iam::aws:policy/CloudWatchLogsFullAccess Path: / StartApplicationLambda: Type: AWS::Lambda::Function DependsOn: StartApplicationLambdaRole Properties: Description: Starts an application when invoked. Runtime: python3.8 Role: !GetAtt StartApplicationLambdaRole.Arn Handler: index.lambda_handler Timeout: 30 Code: ZipFile: | import logging import cfnresponse import boto3 logger = logging.getLogger() logger.setLevel(logging.INFO) def lambda_handler(event, context): logger.info('Incoming CFN event {}'.format(event)) try: application_name = event['ResourceProperties']['ApplicationName'] # filter out events other than Create or Update, # you can also omit Update in order to start an application on Create only. if event['RequestType'] not in ["Create", "Update"]: logger.info('No-op for Application {} because CFN RequestType {} is filtered'.format(application_name, event['RequestType'])) cfnresponse.send(event, context, cfnresponse.SUCCESS, {}) return # use kinesisanalyticsv2 API to start an application. client_kda = boto3.client('kinesisanalyticsv2', region_name=event['ResourceProperties']['Region']) # get application status. describe_response = client_kda.describe_application(ApplicationName=application_name) application_status = describe_response['ApplicationDetail']['ApplicationStatus'] # an application can be started from 'READY' status only. if application_status != 'READY': logger.info('No-op for Application {} because ApplicationStatus {} is filtered'.format(application_name, application_status)) cfnresponse.send(event, context, cfnresponse.SUCCESS, {}) return # create RunConfiguration from passed parameters. run_configuration = { 'FlinkRunConfiguration': { 'AllowNonRestoredState': event['ResourceProperties']['AllowNonRestoredState'] == 'true' }, 'ApplicationRestoreConfiguration': { 'ApplicationRestoreType': event['ResourceProperties']['ApplicationRestoreType'], } } # add SnapshotName to RunConfiguration if specified. if event['ResourceProperties']['SnapshotName'] != '': run_configuration['ApplicationRestoreConfiguration']['SnapshotName'] = event['ResourceProperties']['SnapshotName'] logger.info('RunConfiguration for Application {}: {}'.format(application_name, run_configuration)) # this call doesn't wait for an application to transfer to 'RUNNING' state. client_kda.start_application(ApplicationName=application_name, RunConfiguration=run_configuration) logger.info('Started Application: {}'.format(application_name)) cfnresponse.send(event, context, cfnresponse.SUCCESS, {}) except Exception as err: logger.error(err) cfnresponse.send(event,context, cfnresponse.FAILED, {"Data": str(err)}) StartApplicationLambdaInvoke: Description: Invokes StartApplicationLambda to start an application. Type: AWS::CloudFormation::CustomResource DependsOn: StartApplicationLambda Version: "1.0" Properties: ServiceToken: !GetAtt StartApplicationLambda.Arn Region: !Ref AWS::Region ApplicationName: !Ref TestFlinkApplication ApplicationRestoreType: !Ref ApplicationRestoreType SnapshotName: !Ref SnapshotName AllowNonRestoredState: !Ref AllowNonRestoredState

同样,您可能需要调整 Lambda 以及应用程序本身的角色。

在创建上面的堆栈之前,请不要忘记指定您的参数。

参数.json

[ { "ParameterKey": "CodeContentBucketArn", "ParameterValue": "YOUR_BUCKET_ARN" }, { "ParameterKey": "CodeContentFileKey", "ParameterValue": "YOUR_JAR" }, { "ParameterKey": "ApplicationRestoreType", "ParameterValue": "SKIP_RESTORE_FROM_SNAPSHOT" }, { "ParameterKey": "AllowNonRestoredState", "ParameterValue": "true" } ]

YOUR_JAR用您的特定要求替换YOUR_BUCKET_ARN和。您可以按照本指南创建 Amazon S3 存储桶和应用程序 jar。

现在创建堆栈(将 YOUR_REGION 替换为您选择的区域,例如 us-east-1):

aws cloudformation create-stack --region YOUR_REGION --template-body "file://stack.yaml" --parameters "file://parameters.json" --stack-name "TestManaged Service for Apache FlinkStack" --capabilities CAPABILITY_NAMED_IAM

您现在可以导航到 https://console.aws.amazon.com/cloudformation 查看进度。创建后,您应该会看到您的 Flink 应用程序处于Starting状态。可能需要几分钟时间来运行Running

有关更多信息,请参阅下列内容: