Opscraft
AWS core services · Lesson 4 of 8 · 3 min read

Databases and messaging, the DevOps view

RDS and Aurora

  • Multi-AZ (synchronous standby, automatic failover ~1–2 min via DNS change) is for availability; read replicas (asynchronous) are for read scaling (and can be promoted for DR, including cross-region).
  • Aurora: storage replicated 6 ways across 3 AZs, up to 15 low-lag replicas, failover typically under 30 s, cluster (writer) and reader endpoints, Global Database for cross-region DR (~1 s replication lag), Serverless v2 for variable load.
  • Backups: automated (point-in-time restore within retention, up to 35 days) + manual snapshots; restore always creates a new instance — plan DNS/config switching.
  • Operations: parameter groups, maintenance windows, blue/green deployments for major upgrades, Performance Insights, RDS Proxy for connection pooling (Lambda, bursty apps), IAM database authentication, Secrets Manager rotation.
  • Connection storms from many pods → RDS Proxy or app pooling; keep DBs in private data subnets with SG references from app SGs.
Enlarge diagramMULTI-AZREAD REPLICASprimaryAZ astandbyAZ b · notrafficprimarywritesreplica 1readsreplica 2readssyncFailover in ~1-2 min by DNS.async: may lag behind
Figure 1. Multi-AZ is for availability; read replicas are for scaling reads. They solve different problems.
Swipe sideways to see all of it, or tap to open it full screen.

DynamoDB

Key-value/document, single-digit-ms latency at any scale. Partition key (+ optional sort key) design is everything — avoid hot partitions. On-demand vs provisioned (with auto scaling), GSIs/LSIs, TTL for expiry, Streams for change data capture, PITR backups, global tables for multi-region active-active. Common DevOps use: Terraform state locking (historically), feature flags, idempotency keys.

Messaging and events

Service Model Use
SQS Queue (pull); Standard (at-least-once, best-effort order) or FIFO (ordering, exactly-once processing) Decoupling, buffering, work queues; visibility timeout; DLQs
SNS Pub/sub push to SQS, Lambda, HTTP, email/SMS Fan-out
EventBridge Event bus with content-based rules, schedules, cross-account AWS service events (e.g. ECR scan complete → Lambda), SaaS events, cron
Kinesis / MSK Streams High-throughput ordered streams; Kafka

Pattern: SNS → multiple SQS queues for fan-out; KEDA scales Kubernetes consumers on SQS queue depth.

Enlarge diagramorders servicepublishesOrderPlacedSNS topicSQS: billingbilling workerSQS: emailemail workerSQS:analyticsanalytics job
Figure 2. Fan-out with SNS and SQS: one event, many independent consumers, each with its own queue so a slow one never blocks the rest.
Swipe sideways to see all of it, or tap to open it full screen.