Domain 2 of 4 · Chapter 4 of 6

Reliability for new solutions

Unlock the complete study guide + 1,040 practice questions across 16 full exams.

Bundled into the existing AWS Certified Solutions Architect - Professional premium course — no separate purchase.

Included in this chapter:

  • Multi-AZ self-healing: ELB, Auto Scaling, and health checks
  • Decoupling: SQS, SNS, EventBridge, and dead-letter queues
  • Idempotency, retries with backoff and jitter, and circuit breakers
  • Throttling, graceful degradation, and static stability
  • Data and session tiers, and the Well-Architected Reliability pillar

Choosing an application-integration service for decoupling (SAP-C02)

DimensionAmazon SQSAmazon SNSAmazon EventBridgeKinesis Data Streams
PatternPoint-to-point queue (pull)Pub/sub fan-out (push)Event bus, routing + filtering (push)Ordered streaming log (pull)
Consumers per messageOne consumer deletes itEvery subscriber gets a copyEvery matching rule target gets a copyMany consumers, independent offsets
Ordering and replayFIFO queues for strict order; no replayFIFO topics for order; no replayNo ordering; archive plus replay supportedPer-shard order; replay within retention
Buffering / load levelingYes: core strengthNo (immediate delivery)No (immediate routing)Yes: retains records in the stream
Failure handlingmaxReceiveCount then DLQPer-subscription DLQRetry policy then DLQ; some errors go straight to DLQBisect-on-error; on-failure destination
Best fitDecouple a producer from a worker fleet, buffer spikesNotify many systems of one eventAWS/SaaS service events, content-based routingHigh-throughput ordered analytics, multiple readers

Decision tree

One message to many independent consumers? No (one worker) Need to buffer spikes / load-level a worker fleet? Yes (fan-out) Route by event content / from a SaaS/AWS source? Yes Amazon SQS queue-based load leveling; FIFO for strict order/exactly-once No, ordered + replay Kinesis Data Streams high-throughput ordered log; many readers, replay in retention Yes Amazon EventBridge content-filtered routing, schema + SaaS/AWS event sources No, simple fan-out Amazon SNS pub/sub to every subscriber; durable as SNS-to-SQS fan-out Always: attach a dead-letter queue (maxReceiveCount on SQS; retry policy then DLQ on EventBridge), make handlers idempotent, and retry transient errors with exponential backoff + jitter behind a circuit breaker.
Cheat sheet

Cheat sheet

  • An AZ is the fault boundary: one AZ is not HA
  • ALB cross-zone always on; NLB off by default
  • Use ELB health-check type on the Auto Scaling group
  • Target tracking is the default scaling policy
  • Warm pools cut scale-out latency for slow-booting instances
  • Lifecycle hooks gate warm-pool readiness
  • SQS queue-based load leveling absorbs spikes
  • SQS DLQ triggers on maxReceiveCount; same queue type
  • redrive-allow-policy byQueue allows up to 10 sources
  • SNS fan-out vs SNS-to-SQS durability
  • EventBridge routes by content; some errors skip retries
  • At-least-once delivery demands idempotent handlers
  • Exponential backoff with jitter beats fixed-interval retries
  • Asynchronous Lambda retries function errors twice by default
  • Circuit breakers fail fast; bulkheads isolate resources
  • API Gateway throttling: 10,000 RPS / 5,000 burst default
  • API Gateway usage plans + API keys for per-client limits
  • Graceful degradation keeps the critical path alive
  • Static stability avoids control-plane dependency at failover
  • Service quotas are a silent single point of failure
  • Stateless app tier + externalized session state
  • RDS Multi-AZ ~60–120s; Aurora often under 30s
  • Connect to the cluster endpoint; RDS Proxy speeds failover
  • FIFO source queue needs a FIFO dead-letter queue
  • Combine alarms with CloudWatch composite alarms and an actions suppressor

Unlock with Premium — includes all practice exams and the complete study guide.

Also tested in

References

  1. How Elastic Load Balancing works (cross-zone load balancing)
  2. Using dead-letter queues in Amazon SQS
  3. Understanding retry behavior in Lambda
  4. Throttle requests to your REST APIs in API Gateway
  5. Amazon API Gateway quotas
  6. RDS Multi-AZ deployments with one standby instance
  7. High availability for Amazon Aurora
  8. AWS service quotas
  9. Health checks for instances in an Auto Scaling group
  10. Warm pools for Amazon EC2 Auto Scaling
  11. Amazon EC2 Auto Scaling lifecycle hooks
  12. Using dead-letter queues to process undelivered events in EventBridge
  13. What is Amazon ElastiCache?
  14. What is Amazon DynamoDB?
  15. Reliability pillar - AWS Well-Architected Framework Well-Architected