IaaS: VMs, storage, networking (EC2, GCE, Azure VMs)
PaaS: Runtime, managed (Elastic Beanstalk, App Engine, Heroku)
SaaS: End-user apps (Gmail, Slack, Salesforce)
FaaS/Serverless: Functions (Lambda, Cloud Functions, Azure Functions)
Category Service Use
Compute EC2 Virtual machines Compute Lambda Serverless functions Compute ECS/EKS Container orchestration Storage S3 Object storage Storage EBS Block storage Storage EFS File storage Database RDS Managed SQL Database DynamoDB NoSQL Database ElastiCache Redis/Memcached Database Aurora MySQL/PostgreSQL compatible Network VPC Virtual network Network ELB Load balancer Network CloudFront CDN Network Route 53 DNS Messaging SQS Queue Messaging SNS Pub/sub Messaging Kinesis Stream processing Analytics Athena Query S3 with SQL ML SageMaker ML platform
# Core Objects
Pod: Smallest deployable unit (1+ containers)
Deployment: Manages ReplicaSets, rolling updates
Service: Stable network endpoint (ClusterIP, NodePort, LoadBalancer)
ConfigMap/Secret: Configuration management
Ingress: HTTP routing, TLS termination
PV/PVC: Persistent storage
Namespace: Resource isolation
# Key Commands
kubectl get pods -n <namespace>
kubectl logs <pod> -f
kubectl exec -it <pod> -- /bin/bash
kubectl apply -f manifest.yaml
kubectl rollout undo deployment/<name>
Container VM
Startup Seconds Minutes Size MBs GBs Isolation Process-level Hardware-level OS Shared kernel Full OS Density 100s per host 10s per host
Code → Build → Test → Stage → Deploy → Monitor
Build: Compile, lint, unit tests
Test: Integration, e2e, security scan
Stage: Canary, blue-green, rolling
Deploy: Progressive rollout
Monitor: Metrics, logs, alerts
Strategy Downtime Risk Rollback Speed
Rolling Zero Medium Fast Blue-Green Zero Low Instant Canary Zero Low Fast Recreate Yes High Slow
Metrics: Numbers over time (CPU, memory, request rate, error rate)
→ Prometheus, Grafana, CloudWatch
Logging: Structured events
→ ELK Stack, Loki, CloudWatch Logs
Tracing: Request flow across services
→ Jaeger, Zipkin, OpenTelemetry
Alerting: Threshold-based notifications
→ PagerDuty, OpsGenie, Alertmanager
SLI: Service Level Indicator (measured metric, e.g., latency p99)
SLO: Service Level Objective (target, e.g., p99 < 200ms)
SLA: Service Level Agreement (contract with penalties)
Error Budget: 1 - SLO = allowed failure rate
Git as single source of truth
Declarative infrastructure (Terraform, Pulumi)
Automated reconciliation (ArgoCD, Flux)
Pull-based deployment (cluster pulls from git)
Tool Type Language
Terraform Declarative HCL Pulumi Imperative/Declarative Python/TS/Go CloudFormation Declarative YAML/JSON Ansible Procedural YAML
Right-sizing instances
Reserved/spot instances
Auto-scaling policies
Storage tiering (S3 Standard → Glacier)
Delete unused resources
Tag everything for cost allocation
Know the trade-offs between services (RDS vs DynamoDB, SQS vs Kinesis)
Design for failure (multi-AZ, multi-region)
Security: least privilege, encrypt at rest and in transit
Cost matters: mention reserved capacity, spot instances
Monitoring is not optional: discuss what to measure