Night 29 — Week 4 review: storage, analytics, edge, compute, integration

~2 hr. $0 AWS spend — paper architecture + timed quiz. No new deploy tonight. Synthesizes Nights 23–28 before Week 5 cost labs (S3 lifecycle, Fargate right-sizing, Step Functions capstone).

Plain language first

What Week 4 was about

Week 3 answered “where does data live and how do we reach it?” Week 4 answers “how do we make it fast, query it at scale, route users intelligently, and move signals between services?”

Tonight you connect the dots — one storage decision tree, one full-stack diagram, one quiz, one architecture-notes pass.

Block 1 (~20 min) — Storage class decision tree (exam artifact)

Assignment: Draw the decision tree below from memory. Time yourself: 12 minutes draw, 8 minutes explain aloud using GSA examples for each leaf.

Master tree — S3 + block + file + ephemeral

What are you storing?
│
├─ Objects (files, images, pipeline output, analytics Parquet)
│     │
│     ├─ Frequent access, low latency GET/PUT
│     │     → S3 Standard
│     │     GSA: Iceland GeoJSON output, Night 24 Parquet, static site export
│     │
│     ├─ Unknown or changing access pattern
│     │     → S3 Intelligent-Tiering (small monitoring fee)
│     │
│     ├─ Infrequent access, still need ms retrieval
│     │     → S3 Standard-IA or One Zone-IA (cheaper, AZ risk on One Zone)
│     │
│     ├─ Archive — minutes to hours retrieval OK
│     │     → S3 Glacier Flexible Retrieval or Glacier Deep Archive
│     │
│     └─ Lifecycle automation across tiers
│           → S3 Lifecycle rules (Night 30 lab preview)
│
├─ Block disk for ONE EC2 instance (boot or data volume)
│     │
│     ├─ General SSD — boot, app disks, moderate IOPS
│     │     → EBS gp3 (right-size IOPS/throughput without bigger disk)
│     │
│     ├─ Highest random IOPS — database ON EC2 (not RDS/Aurora)
│     │     → EBS io2 / io2 Block Express
│     │
│     ├─ Big sequential scans — logs, throughput-bound HDD
│     │     → EBS st1 (not for boot/root on standard AMIs)
│     │
│     └─ Cheapest bulk HDD — rarely accessed
│           → EBS sc1
│
├─ Shared files across many EC2/Lambda (NFS, same content)
│     → Amazon EFS (regional NFS; scales with clients)
│     GSA: shared PMTiles build config across Batch workers (exam scenario)
│
├─ Temp scratch on ONE EC2 — OK to lose on stop/terminate
│     → EC2 Instance Store (ephemeral, highest local IOPS)
│     GSA: Iceland batch temp — durable output still goes to S3
│
└─ Managed SQL / NoSQL with ACID
      → Aurora/RDS or DynamoDB — NOT EBS as “database product”
      GSA: resort_stats in Aurora; wiki counters in DynamoDB

Three-layer cache + analytics map (draw second diagram)

                    Users (global)
                         │
                         ▼
              ┌──────────────────────┐
              │ CloudFront (Night 26)│  ← HTTP responses at edge
              │  static HTML cached  │
              │  api/wiki* no cache  │
              └──────────┬───────────┘
                         │
         ┌───────────────┼───────────────┐
         ▼               ▼               ▼
    S3 static       API Gateway      Route 53 (Night 25)
    export          + Lambda wiki    weighted / failover /
                         │           latency policies
                         ▼
              ┌──────────────────────┐
              │ ElastiCache Redis    │  ← app objects in VPC (Night 23)
              │ cache-aside Aurora   │
              └──────────┬───────────┘
                         │ miss
                         ▼
              ┌──────────────────────┐
              │ Aurora OLTP          │  ← source of truth (Nights 16–17)
              └──────────┬───────────┘
                         │ snapshot / ETL
                         ▼
              ┌──────────────────────┐
              │ S3 + Glue + Iceberg  │  ← Athena SQL (Night 24)
              └──────────────────────┘

Side channel — Iceland pipeline (Nights 12–14, 27–28):
  EventBridge cron → ECS Fargate → S3 output
       success → EventBridge → SQS → Lambda uploader
       failure → EventBridge → SNS fan-out

Checklist — labels the SAA exam expects on your diagram

Three sentences to practice out loud

1. “CloudFront caches HTTP at the edge; ElastiCache caches application objects inside the VPC; they solve different problems and stack — not replace each other.”
2. “Aurora serves live resort_stats; Athena queries Parquet/Iceberg snapshots in S3 — OLTP vs analytics, not either/or for the same workload.”
3. “Iceland pipeline durable output is S3; EBS on a batch worker is scratch; EventBridge schedules the job, SQS buffers success handoff, SNS fans out failures.”

Block 2 (~40 min) — Week 4 domain recap

Caching ladder (Night 23 + 26)

Need to cache HTTP GET responses globally?
  └─ CloudFront (path behavior + cache policy + origin Cache-Control)

Need to cache DB query results near Lambda in VPC?
  └─ ElastiCache Redis — cache-aside (app populates on miss)

Need microsecond DynamoDB reads at scale?
  └─ DAX (not Redis — DynamoDB-specific)

Memcached vs Redis?
  └─ Redis → sessions, data structures, replication failover
  └─ Memcached → simplest horizontal string cache

Analytics ladder (Night 24)

App needs millisecond UPSERT with ACID?
  └─ Aurora/RDS (Nights 16–17)

Ad hoc SQL on S3 data lake, unpredictable volume?
  └─ Athena + Glue catalog

Petabyte warehouse, heavy BI concurrency?
  └─ Redshift

Live streaming ingest to S3 for later Athena?
  └─ Kinesis Data Firehose (awareness — GSA uses batch ECS today)

Need time travel / schema evolution on lake tables?
  └─ Iceberg format on S3

DNS + edge ladder (Nights 25–26)

Compute + storage ladder (Night 27)

Integration ladder (Night 28)

Block 3 (~60 min) — Timed quiz

File: night-29-quiz.json — 30 scenario questions (~48 min at 96 sec each, or ~40 min if warmed up).

Topics span all of Week 4 — ElastiCache, Athena/Iceberg, Route 53, CloudFront/API GW, EC2/EBS/EFS/Spot, integration services, plus storage-class synthesis. Answer without night-29-quiz-answers.json first.

Score on the study site: /aws-solutions-architect-study/quiz/29/ (restart npm run dev after adding new quiz files).

Target: ≥24/30 (80%) before starting Night 30 S3 lifecycle lab.

Block 4 (~15 min) — Update architecture notes

Add one sentence per pillar to your portfolio notes (or append to existing study-lab docs):

Well-Architected framing (exam + portfolio):

• Performance: CloudFront + ElastiCache layers; Athena partitions
• Reliability: Route 53 failover; SQS DLQ; Redis replication group
• Cost: gp3 over io2 when IOPS not needed; Athena scan minimization; Spot for batch
• Operational excellence: Step Functions visibility (Night 32 preview); CloudWatch dashboards (Night 26)

Week 4 master map

Teardown checklist (Week 4 labs)

Keep running if you plan Night 30–32 labs that reuse Night 9 VPC, Route 53 records, or integration stacks.

Tear down in order when pausing study for a week:

Cost reminders:

• ElastiCache cache.t4g.micro bills while cluster exists — Night 23 teardown stops it.
• Route 53 health check ~$0.50/mo while enabled — Night 25 teardown removes lab check.
• EBS gp3 demo volume — pennies/day; Night 27 teardown deletes it.
• NAT Gateway (Night 9) still the biggest idle cost if VPC stays up — see Night 22 teardown order.

Do not tear down (production / other weeks):

• globalskiatlas.com CloudFront, wiki API, Cognito
• Night 12–14 EventBridge / SQS / SNS if continuing to Night 32 capstone
• Night 16 Aurora if Week 5 labs still reference it

5 flashcards (write tonight)

1. Storage tree — Iceland pipeline final GeoJSON lands where? EBS scratch vs S3?
   (Durable output → S3 Standard; EBS/Instance Store = temp local scratch only.)

2. Cache layers — Wiki static HTML vs resort_stats JSON in VPC?
   (CloudFront edge for HTTP static; ElastiCache Redis cache-aside for DB rows.)

3. OLTP vs analytics — Live UPSERT vs “resorts per country” report?
   (Aurora OLTP; Athena on S3/Iceberg snapshot.)

4. Route 53 failover — PRIMARY ALB unhealthy — what must exist on PRIMARY record?
   (Health check on PRIMARY; low TTL helps clients flip faster.)

5. Integration trio — Monthly cron vs slow worker buffer vs email three teams?
   (EventBridge schedule; SQS; SNS fan-out to SQS queues.)

Night 30 preview

Lab 4A — S3 lifecycle + cost — lifecycle rules on old pipeline prefixes, Cost Explorer top services, AWS Budget alert. Applies the storage class tree from Block 1 to real GSA buckets.

Reference links

• S3 storage classes
• EBS volume types
• Amazon EFS
• Athena best practices
• Route 53 routing policies
• CloudFront cache behaviors
• Amazon ElastiCache best practices