Why decouple?

Tightly coupled systems fail together: a spike or crash in one service cascades to others. Loose coupling absorbs spikes, isolates failures, and lets teams deploy independently — core SAA resilient design theme.

Messaging patterns

Amazon SQS essentials

• Standard queue — at-least-once delivery, best-effort ordering, nearly unlimited throughput.
• FIFO queue — exactly-once processing, strict order, throughput limits (batching + high throughput mode on exam as option).
• Visibility timeout — message hidden while worker processes; extend if job takes longer or message returns to queue.
• Dead-letter queue (DLQ) — after maxReceiveCount failures, message moves to DLQ for inspection (poison pill pattern).
• Long polling — reduces empty receives and cost (WaitTimeSeconds).

Exam pattern: sudden traffic burst → queue between API and workers → Auto Scaling on worker fleet based on queue depth (CloudWatch ApproximateNumberOfMessagesVisible).

Amazon SNS

• Topics publish to multiple subscribers in parallel.
• Common combo: SNS → SQS fan-out (each subscriber queue gets a copy) for durable parallel processing.
• Message filtering — subscribers receive subset based on JSON attributes.
• FIFO topics exist for strict ordering use cases.

Amazon EventBridge

• Event bus — default bus for AWS service events; custom buses for multi-account/org patterns.
• Rules match event patterns → target (Lambda, Step Functions, SQS, SNS, API destinations).
• Schema registry — discover and enforce event shapes.
• Archive & replay — debug and reprocess events.

Prefer EventBridge over SNS when routing many event types from AWS services with content-based filtering.

Idempotency and ordering

• Design consumers idempotent — duplicate deliveries happen (SQS standard, SNS retries).
• Need strict order per customer? FIFO SQS or Kinesis partition key — not standard SQS.

Step Functions (bonus tie-in)

Orchestrate multi-step workflows with built-in retries, catch, and parallel branches — exam mentions when coordinating Lambda + human approval + compensating transactions.

Reference architecture

API Gateway → Lambda → SQS → EC2/ECS workers (Auto Scaling on queue depth)
                      ↘ DLQ → alarm → ops
S3 upload event → EventBridge rule → Lambda thumbnail + SNS notify

Exam traps

• SQS is pull — consumers poll; SNS is push.
• DLQ must be same type (FIFO → FIFO) and configured on source queue.
• "Exactly-once" in exam usually means FIFO SQS + idempotent consumer, not standard queue.

Official reference

SAA-C03 exam guide — scalable and loosely coupled architectures.