2.6 - Event-driven architecture and cost-aware design

1. A batch pipeline can tolerate interruptions and restart from checkpoints. It runs for hours each night on hundreds of instances. What lowers compute cost the most?

• On-Demand instances sized precisely for the nightly job window
• A 3-year Reserved Instance purchased for every instance in the fleet
• Dedicated Hosts reserved to guarantee physical capacity every night
• EC2 Spot Instances with checkpointing and diversified instance types (correct answer)

An interruptible, checkpointed batch job is ideal for Spot, which can be up to ~90% cheaper; diversifying instance types reduces interruption impact. Reservations suit steady 24/7 usage (not nightly bursts), and Dedicated Hosts are the most expensive option.

2. A SaaS platform emits many event types and wants an event bus where each new subscriber consumes only the events it cares about, routed by attributes, and where a third-party monitoring SaaS can receive selected events directly. Which TWO EventBridge capabilities enable this? (Choose TWO.)

• Content-based event rules that match on the event payload and route to targets (correct answer)
• A single shared SQS queue that every subscriber polls and then filters on the client side
• API destinations that deliver matched events to external HTTP/SaaS endpoints (correct answer)
• A Kinesis shard per event type that consumers must poll continuously

EventBridge rules match on event content and route only relevant events to each target, so new subscribers add a rule without touching producers, and API destinations deliver matched events to external HTTP or SaaS endpoints with auth and rate control. A shared SQS queue forces client-side filtering, and per-type Kinesis shards add polling and management overhead.

3. An image-processing service must notify exactly one worker fleet that pulls and processes jobs at its own pace, absorbing bursts without losing work if all workers are briefly down. Which service is the right buffer between producer and workers?

• Amazon SNS, pushing each job to the worker fleet as a fan-out notification
• Amazon EventBridge, matching each job against rules before delivery to workers
• AWS Step Functions, orchestrating each job through a sequence of worker states
• Amazon SQS, holding jobs until a worker polls and deletes each message (correct answer)

SQS is a pull-based buffer: workers poll and process at their own pace, and messages persist if workers are temporarily down, giving natural burst absorption and load leveling. SNS pushes and would drop messages with no available subscriber acknowledgment path, EventBridge is for routing, and Step Functions orchestrates rather than buffers a single worker pool.

4. A telemetry platform ingests millions of ordered records per second and needs multiple independent applications to replay and re-read the same stream over a 7-day window for different analytics. Which service is the best backbone?

• Amazon SQS standard queues, one per consuming analytics application
• Amazon SNS topics fanning every incoming record out to each of the analytics applications
• Amazon Kinesis Data Streams with multiple consumers reading the retained stream (correct answer)
• AWS Step Functions coordinating the analytics applications per record

Kinesis Data Streams retains ordered records (up to 7 days, or longer with extended retention) and lets multiple independent consumers read and replay the same shard data, which is exactly the multi-reader, ordered, high-throughput need. SQS deletes messages once consumed (no shared replay), SNS does not retain for replay, and Step Functions orchestrates rather than streams.

5. An event-driven platform ingests events from an SQS source, must lightly transform and filter them, and route the results to different targets with minimal glue code, while producers and consumers agree on evolving event structures safely. Which TWO EventBridge capabilities help most? (Choose TWO.)

• A fleet of always-on EC2 workers that poll SQS and transform each event by hand
• EventBridge Pipes to connect the SQS source through filtering and enrichment to a target (correct answer)
• The EventBridge Schema Registry so consumers generate code bindings for evolving events (correct answer)
• A single monolithic Lambda function that hardcodes every source-to-target mapping in one place

EventBridge Pipes wires a source (like SQS) to a target with built-in filtering and optional enrichment, removing custom glue code, and the Schema Registry lets teams discover and generate typed bindings so event structures can evolve safely. Always-on EC2 pollers and a monolithic mapping Lambda are exactly the undifferentiated glue code Pipes is meant to replace.

6. An S3 bucket stores thumbnails that are easily regenerated from originals and accessed infrequently. Losing a copy is acceptable because they can be rebuilt. Which storage class minimizes cost for this data?

• S3 Standard for maximum durability across three Availability Zones
• S3 One Zone-Infrequent Access (correct answer)
• S3 Glacier Flexible Retrieval with expedited restores
• S3 Standard-Infrequent Access with cross-Region replication

One Zone-IA stores infrequently accessed data in a single AZ at lower cost, ideal for reproducible data where losing the AZ copy is tolerable because it can be regenerated. Standard and Standard-IA cost more for multi-AZ durability, and Glacier adds retrieval latency and fees for data that is still read on demand.