Memory hook: Queue work, broadcast notifications, route events, retain streams, then analyse the data.
Must remember
- SQS buffers messages for workers; standard queues require consumers that tolerate possible duplicates. SNS publishes to subscribers for fan-out. EventBridge routes events by rules and integrates event-driven applications. Step Functions coordinates workflow steps and state.
- Kinesis Data Streams supports streaming records and replay within retention. Amazon Data Firehose delivers streaming data to supported destinations with managed buffering. A queue of jobs and a replayable stream have different consumption models.
- Athena runs SQL over supported data sources, commonly S3. Glue provides cataloguing and data integration/ETL capabilities. EMR supports big-data frameworks. Redshift is analytical warehousing; Quick Sight is business intelligence/dashboarding. Store, catalogue, transform, query and visualise are distinct stages.
- SageMaker AI supports building, training and deploying ML models. Bedrock provides managed foundation-model application capabilities. Use Rekognition for supported image/video analysis, Textract for document text and structure, Comprehend for language insights, Translate for translation, Transcribe for speech-to-text, Polly for text-to-speech and Lex for conversational interfaces.
- Connect supports cloud contact centres; SES sends application email. IoT Core connects and manages messaging for devices. WorkSpaces provides virtual desktops; AppStream 2.0 provides application streaming; WorkSpaces Secure Browser provides managed browser access. Course names can lag product branding.
- Amplify supports frontend/mobile application development and hosting workflows. CodeBuild runs builds and tests; CodePipeline coordinates delivery stages; X-Ray helps trace distributed requests. These are different parts of application delivery.
- Prefer an appropriate managed task service over building a custom ML model when the requirement is already supported. AI output still requires evaluation, privacy controls and human review where the consequence demands it.
Choose under exam pressure
| Requirement | Choice and reason |
|---|---|
| One event must reach several independent consumers | SNS fan-out or EventBridge routing, depending on filtering/integration needs. |
| Query S3 files using SQL | Athena. |
| Turn a written announcement into speech | Polly. |
Traps
- Transcribe and Polly point in opposite directions.
- SNS fan-out does not replace a durable work queue for every subscriber.
- Glue metadata does not itself store all of the source data.
Active recall
1. Which service finds document forms and tables?
Textract for supported document extraction.
2. Which service coordinates a multi-step process with state?
Step Functions.
3. A data analyst wants charts rather than ETL code. Which category fits?
Business intelligence, such as Quick Sight.
4. Which tool records a request across distributed components?
X-Ray tracing, with suitable instrumentation.
5. Does a custom ML model always beat a managed AI API?
No. Consider fit, quality, operating effort, privacy and total cost before custom training.