Night 39 — Keywords: compute, networking, integration
~2 hr. $0 AWS spend. Force words about time, path, and change feed decide these questions.
Plain language first
within seconds → Lambda (often + API Gateway)
minutes / Auto Scaling group → EC2 Auto Scaling family
predictable 9–5 → scheduled scaling
no NAT gateway + AWS APIs private → VPC endpoints
on-prem dedicated link → Direct Connect (not “private S3 endpoint”)
IPv6 outbound only → egress-only internet gateway
CloudFront auth near users + cheap → Lambda@Edge
many Regions for latency (expensive) → only if stem allows cost
Block 1 (~30 min) — Maps
Compute / scale
| Force words | Pick | Kill |
|---|---|---|
| bursts within seconds; API frontend | API Gateway + Lambda | EC2/ECS/Beanstalk Auto Scaling |
| protect API backends from spikes | API Gateway throttling (+ cache sometimes) | “do nothing”; Multi-AZ API Gateway |
| predictable office-hours load; morning cold | scheduled scaling | wait for dynamic CPU |
| mixed instance types/sizes; known schedule | scheduled over predictive | predictive (assumes same instance family) |
| need N after an Availability Zone dies | Auto Scaling group min ≥ 2N across Availability Zones | min = N with 1 per Availability Zone |
| default scale-in terminate preference | oldest launch template (after Availability Zone balance) | “oldest instance” / random first |
| memory / disk space on EC2 | CloudWatch agent | default CloudWatch metrics |
| edge auth; don’t redeploy whole stack multi-Region | Lambda@Edge | multi-Region + latency Route 53 for cost stems |
| CloudFront 504s | origin failover / origin group | ignore high availability |
Network / hybrid path
| Force words | Pick | Kill |
|---|---|---|
| private AWS service access, no internet gateway / NAT gateway | VPC endpoint | Direct Connect, VPN CloudHub, encryption |
| dedicated on-prem ↔ AWS bandwidth | Direct Connect | “Direct Connect as S3 private endpoint” inside the VPC |
| interconnect many VPCs/accounts, share resources | Organizations + Resource Access Manager (when sharing) | ParallelCluster |
Integration / streaming
| Force words | Pick | Kill |
|---|---|---|
| DynamoDB item change → notify | Streams + Lambda (+ SNS) | DynamoDB Accelerator; Adapter without enable Streams |
| anonymize before durable NoSQL | Kinesis → Lambda → DynamoDB | land in S3/Dynamo first |
| S3 event → many team queues | SNS fan-out → SQS (S3 → one destination only) | two SQS queues directly on S3 |
| S3 PUT → run Fargate/ECS task least effort | EventBridge → ECS task | Lambda wrapper unless needed |
Block 2 (~60 min) — Stem drills
- Private subnet; S3+DynamoDB; no NAT gateway; no public Internet.
- Portal expects global spike; protect backends beyond network ACLs.
- Auto Scaling group slow every morning 9am; known schedule; mixed instance sizes.
- Fault tolerant: always ≥2 after Availability Zone loss; peak 6.
- Memory + disk utilization on Linux and Windows EC2.
- Long CloudFront logins + occasional 504; cost-effective.
- Real-time personal data; anonymize before NoSQL store.
- DynamoDB write capacity hot partitions / uneven keys.
- Default Auto Scaling scale-in; which instance preference after Availability Zone balance?
- SFTP least operational work was Night 36 — tonight: API bursts in seconds.
- VPC endpoints
- API Gateway throttling
- Scheduled scaling
- Min 4 / max 6
- CloudWatch agent
- Lambda@Edge + origin failover
- Kinesis → Lambda → DynamoDB
- High-cardinality keys
- Oldest launch template
- API Gateway + Lambda
Block 3 (~30 min) — Quiz
night-39-quiz.json (20 questions).
Tomorrow (Night 40)
Timed Practice Exam 2 (target ≥75%). For every miss: write the force words you missed — not a new service chapter.