Back to All Sheets
AWSCurrent AWS PlatformINTERMEDIATE

Amazon Web Services (AWS)

AWS reference covering accounts, regions, IAM, networking, compute, storage, databases, serverless, messaging, observability, security, reliability, cost, infrastructure as code, and incident readiness.

78 min read
Category: Cloud & DevOps

Accounts, Regions & Identity

Organizations, Accounts, Regions & AZs

Use multiple accounts as strong isolation boundaries, choose regions for users/compliance/services, and spread production across Availability Zones.

text
Organization
├─ Security / Log Archive
├─ Shared Services
├─ Production account(s)
└─ Non-production account(s)

IAM Roles, Policies & Federation

Prefer short-lived federated sessions and workload roles. Evaluate identity policy, resource policy, boundaries, SCPs, and explicit denies together.

AWS Identity Mechanisms
MechanismUseCredential Model
IAM Identity CenterHuman workforce accessFederated short-lived sessions
IAM roleAWS workload/service accessTemporary STS credentials
Resource policyCross-account/service delegationAttached to resource
IAM userExceptional legacy/service caseLong-lived unless federated; minimize
🔒
Explicit Deny Wins

Least privilege requires both narrow allows and organization guardrails. Never embed long-lived access keys in code, images, or repositories.

Networking & Compute

VPC, Subnets, Routing & Endpoints

Routes decide reachability; security groups are stateful resource firewalls; NACLs are stateless subnet filters. Private subnets still need explicit egress/service paths.

text
Internet -> ALB (public subnets, multi-AZ)
-> ECS/EC2/Lambda ENIs (private subnets)
-> RDS/ElastiCache (isolated subnets)
Private service traffic -> VPC endpoints where supported

EC2, ECS, EKS & Lambda

Select compute by control needs, workload shape, team maturity, startup/latency, portability, and operational burden.

AWS Compute Selection
ServiceBest FitYou Operate
LambdaEvent-driven, bursty, short-lived functionsCode, limits, dependencies
ECS/FargateContainer services without node managementTask/service definitions and app
EKSKubernetes ecosystem and platform needsClusters, add-ons, workloads
EC2OS/kernel/control or specialized hostsInstances, patching, scaling

Storage, Databases & Messaging

S3, EBS & EFS

S3 is regional object storage; EBS is block storage attached within an AZ; EFS is managed shared POSIX file storage.

bash
aws s3 cp artifact.zip s3://example-artifacts/releases/artifact.zip
aws s3api head-object --bucket example-artifacts --key releases/artifact.zip
⚠️
S3 Is Not a Filesystem

Design around whole-object operations, keys, versioning, lifecycle, checksums, multipart upload, and request semantics.

RDS/Aurora, DynamoDB, ElastiCache, SQS & SNS

Choose relational databases for constraints/joins, DynamoDB for known key access at scale, ElastiCache for disposable acceleration, SQS for durable queues, and SNS/EventBridge for fan-out/routing.

🔁
Queue Consumers Must Be Idempotent

At-least-once delivery means duplicates are normal. Define visibility timeouts, retries, poison-message handling, and DLQ redrive.

Reliability, Security, Cost & Delivery

CloudWatch, CloudTrail, Backups & Resilience

Centralize metrics/logs/traces, retain audit trails, encrypt data, test restore, design multi-AZ first, and choose multi-region only from explicit recovery objectives.

text
SLI/SLO -> alarms -> runbook -> owner
RPO/RTO -> backup frequency -> restore test -> failover drill
CloudTrail + config/change events -> central immutable archive

Infrastructure as Code, Cost & Production Checklist

Use reviewed IaC and progressive delivery; tag ownership; budget and anomaly-alert costs; remove idle resources; model quotas and regional dependencies.

Production Checklist

Separate accounts, federate access, enforce least privilege, use multi-AZ, encrypt, centralize logs, test backups, set budgets/quotas, scan IaC, pin artifacts by digest, and rehearse incidents.

Related References

Recommended Next Sheets