Amazon Web Services Data-Engineer-Associate - AWS Certified Data Engineer - Associate (DEA-C01)

AWS Glue workflows are a feature of AWS Glue that enable you to create and visualize complex ETL pipelines using AWS Glue components, such as crawlers, jobs, triggers, and development endpoints. AWS Glue workflows provide automated orchestration and require minimal manual effort, as they handle dependency resolution, error handling, state management, and resource allocation for your ETL workflows. You can use AWS Glue workflows to ingest data from your operational databases into your Amazon S3 based data lake, and then use AWS Glue and Amazon EMR to process the data in the data lake. This solution will meet the requirements with the least operational overhead, as it leverages the serverless and fully managed nature of AWS Glue, and the scalability and flexibility of Amazon EMR12.

The other options are not optimal for the following reasons:

B. AWS Step Functions tasks. AWS Step Functions is a service that lets you coordinate multiple AWS services into serverless workflows. You can use AWS Step Functions tasks to invoke AWS Glue and Amazon EMR jobs as part of your ETL workflows, and use AWS Step Functions state machines to define the logic and flow of your workflows. However, this option would require more manual effort than AWS Glue workflows, as you would need to write JSON code to define your state machines, handle errors and retries, and monitor the execution history and status of your workflows3.

C. AWS Lambda functions. AWS Lambda is a service that lets you run code without provisioning or managing servers. You can use AWS Lambda functions to trigger AWS Glue and Amazon EMR jobs as part of your ETL workflows, and use AWS Lambda event sources and destinations to orchestrate the flow of your workflows. However, this option would also require more manual effort than AWS Glue workflows, as you would need to write code to implement your business logic, handle errors and retries, and monitor the invocation and execution of your Lambda functions. Moreover, AWS Lambda functions have limitations on the execution time, memory, and concurrency, which may affect the performance and scalability of your ETL workflows.

D. Amazon Managed Workflows for Apache Airflow (Amazon MWAA) workflows. Amazon MWAA is a managed service that makes it easy to run open source Apache Airflow on AWS. Apache Airflow is a popular tool for creating and managing complex ETL pipelines using directed acyclic graphs (DAGs). You can use Amazon MWAA workflows to orchestrate AWS Glue and Amazon EMR jobs as part of your ETL workflows, and use the Airflow web interface to visualize and monitor your workflows. However, this option would have more operational overhead than AWS Glue workflows, as you would need to set up and configure your Amazon MWAA environment, write Python code to define your DAGs, and manage the dependencies and versions of your Airflow plugins and operators.

1: AWS Glue Workflows

2: AWS Glue and Amazon EMR

3: AWS Step Functions

AWS Lambda

Amazon Managed Workflows for Apache Airflow

AWS Glue Workflows can coordinate multiple ETL jobs and triggers. They support parallel execution and sequential dependencies, which is ideal for concurrent data processing followed by correlation steps, all with minimal operational overhead.

“Use AWS Glue Workflows to orchestrate multiple ETL jobs in sequence or in parallel, supporting conditional triggers and dependency management.”

– Ace the AWS Certified Data Engineer - Associate Certification - version 2 - apple.pdf

To address the requirement of masking PII data and ensuring secure access throughout the data pipeline, the combination of AWS Glue DataBrew and IAM provides a low-maintenance solution.

A. AWS Glue DataBrew for Masking:

AWS Glue DataBrew provides a visual tool to perform data transformations, including masking PII data. It allows for easy configuration of data transformation tasks without requiring manual coding, making it ideal for this use case.

Amazon SageMaker ML Lineage Tracking provides a unified mechanism to automatically capture and track the lineage of ML artifacts (datasets, jobs, and models). It supports auditability, reproducibility, and compliance by maintaining metadata for every stage of the ML lifecycle.

“Use Amazon SageMaker Lineage Tracking to automatically record and track the lineage of data, training jobs, models, and endpoints to ensure auditability and reproducibility.”

– Ace the AWS Certified Data Engineer - Associate Certification - version 2 - apple.pdf

 The error message indicates that the AWS Glue job cannot access the Amazon S3 bucket through the VPC endpoint. This could be because the VPC’s route table does not have the necessary routes to direct the traffic to the endpoint. To fix this, the data engineer must verify that the route table has an entry for the Amazon S3 service prefix (com.amazonaws.region.s3) with the target as the VPC endpoint ID. This will allow the AWS Glue job to use the VPC endpoint to access the S3 bucket without going through the internet or a NAT gateway. For more information, see Gateway endpoints. References:

Troubleshoot the AWS Glue error “VPC S3 endpoint validation failed”

Amazon VPC endpoints for Amazon S3

[AWS Certified Data Engineer - Associate DEA-C01 Complete Study Guide]

A sort-merge join generates large shuffle files, leading to “No space left on device” errors when both datasets are large. Using a broadcast join sends a smaller dataset to all executors, avoiding shuffle and disk I/O overhead.

“Broadcast joins reduce shuffle I/O by distributing the smaller dataset to all worker nodes, mitigating disk space and shuffle errors.”

– Ace the AWS Certified Data Engineer - Associate Certification - version 2 - apple.pdf

This is the most cost-effective and direct fix for large shuffle-stage failures.

Step 1: Log Collection (FluentBit and CloudWatch)

Option A suggests using FluentBit to collect logs and OpenTelemetry to collect traces.

FluentBit is a lightweight log processor that integrates with Amazon EKS to collect and forward logs from Kubernetes clusters. It is widely used with minimal overhead, making it an ideal choice for log collection in this scenario. FluentBit is also natively compatible with AWS services.

OpenTelemetry is a popular framework to collect traces from distributed applications. It provides observability, making it easier to monitor microservices.

This combination allows you to effectively gather both logs and traces with minimal setup and configuration, aligning with the goal of least development effort.

CloudWatch can be used to monitor logs (Option B and C). However, for applications that need more custom and fine-grained control over logging mechanisms, FluentBit and OpenTelemetry are the preferred choice in microservice environments.

Step 2: Log and Trace Correlation (Amazon OpenSearch)

Option D (Amazon OpenSearch) is specifically designed to search, analyze, and visualize logs, metrics, and traces in real-time. OpenSearch allows you to correlate logs and traces effectively.

With Amazon OpenSearch, you can set up dashboards that help in visualizing both logs and traces together, which assists in identifying any failure points across the entire request flow.

It offers integrations with FluentBit and OpenTelemetry, ensuring that both logs from the EKS cluster and application traces are centrally collected, stored, and correlated without additional heavy development.

Step 3: Why Other Options Are Not Suitable

Option B (Amazon Kinesis) is designed for real-time data streaming and analytics but is not as well-suited for tracing microservice requests and logs correlation compared to OpenSearch.

Option C (Amazon MSK) provides a managed Kafka streaming service, but this adds complexity when trying to integrate and correlate logs and traces from a microservice environment. Setting up Kafka requires more development effort compared to using FluentBit and OpenTelemetry.

Option E (AWS Glue) is primarily an ETL (Extract, Transform, Load) service. While Glue is powerful for data processing, it is not a native tool for log and trace correlation, and using it would add unnecessary complexity for this use case.

Conclusion:

To meet the requirements with the least development effort:

Use FluentBit for log collection and OpenTelemetry for tracing (Option A).

Correlate logs and traces using Amazon OpenSearch (Option D).

This approach leverages AWS-native services designed for seamless integration with microservices hosted on Amazon EKS and ensures effective monitoring with minimal overhead.

For exactly-once delivery and processing in Amazon Kinesis Data Streams, the best approach is to design the application so that it handles idempotency. By embedding a unique ID in each record, the application can identify and remove duplicate records during processing.

Exactly-Once Processing:

Kinesis Data Streams does not natively support exactly-once processing. Therefore, idempotency should be designed into the application, ensuring that each record has a unique identifier so that the same event is processed only once, even if it is ingested multiple times.

This pattern is widely used for achieving exactly-once semantics in distributed systems.

Option B is the best solution to meet the requirements with the least operational overhead because S3 Select is a feature that allows you to retrieve only a subset of data from an S3 object by using simple SQL expressions. S3 Select works on objects stored in CSV, JSON, or Parquet format. By using S3 Select, you can avoid the need to download and process the entire S3 object, which reduces the amount of data transferred and the computation time. S3 Select is also easy to use and does not require any additional services or resources.

Option A is not a good solution because it involves writing custom code and configuring an AWS Lambda function to load data from the S3 bucket into a pandas dataframe and query the required column. This option adds complexity and latency to the data retrieval process and requires additional resources and configuration. Moreover, AWS Lambda has limitations on the execution time, memory, and concurrency, which may affect the performance and reliability of the data retrieval process.

Option C is not a good solution because it involves creating and running an AWS Glue DataBrew project to consume the S3 objects and query the required column. AWS Glue DataBrew is a visual data preparation tool that allows you to clean, normalize, and transform data without writing code. However, in this scenario, the data is already in Parquet format, which is a columnar storage format that is optimized for analytics. Therefore, there is no need to use AWS Glue DataBrew to prepare the data. Moreover, AWS Glue DataBrew adds extra time and cost to the data retrieval process and requires additional resources and configuration.

Option D is not a good solution because it involves running an AWS Glue crawler on the S3 objects and using a SQL SELECT statement in Amazon Athena to query the required column. An AWS Glue crawler is a service that can scan data sources and create metadata tables in the AWS Glue Data Catalog. The Data Catalog is a central repository that stores information about the data sources, such as schema, format, and location. Amazon Athena is a serverless interactive query service that allows you to analyze data in S3 using standard SQL. However, in this scenario, the schema and format of the data are already known and fixed, so there is no need to run a crawler to discover them. Moreover, running a crawler and using Amazon Athena adds extra time and cost to the data retrieval process and requires additional services and configuration.

AWS Certified Data Engineer - Associate DEA-C01 Complete Study Guide

S3 Select and Glacier Select - Amazon Simple Storage Service

AWS Lambda - FAQs

What Is AWS Glue DataBrew? - AWS Glue DataBrew

Populating the AWS Glue Data Catalog - AWS Glue

What is Amazon Athena? - Amazon Athena

The company's Apache Spark ETL job on Amazon EMR uses high CPU but low memory, meaning that compute-optimized EC2 instances would be the most cost-effective choice. These instances are designed for high-performance compute applications, where CPU usage is high, but memory needs are minimal, which is exactly the case here.

Compute Optimized Instances:

Compute-optimized instances, such as the C5 series, provide a higher ratio of CPU to memory, which is more suitable for jobs with high CPU usage and relatively low memory consumption.

Switching from general-purpose EC2 instances to compute-optimized instances can reduce costs while improving performance, as these instances are optimized for workloads like Spark jobs that perform a lot of computation.