Cloud Databases & Database-as-a-Service

10. Practice Problems

Decision-style problems — there is sometimes more than one defensible answer. Use the decision tree from §8 as your reference.

1. In two sentences, distinguish RDS from Aurora.
   RDS এবং Aurora-র পার্থক্য দুই বাক্যে লিখুন।
   ✨ Show Answer

   Answer: RDS runs the standard upstream Postgres or MySQL engine on managed EC2 hosts. Aurora keeps the same wire-protocols and SQL but replaces the storage engine with AWS's own distributed, multi-AZ-replicated layer — yielding higher throughput and durability at higher cost.

   RDS = standard Postgres/MySQL managed। Aurora = একই wire-protocol, কিন্তু AWS-এর নিজের লেখা distributed storage — দ্রুততর ও বেশি durable, দাম একটু বেশি।

2. A Bangladeshi e-learning startup with spiky traffic (exam nights) needs Postgres. Which DBaaS is most cost-effective and why?
   পরীক্ষার রাতে ট্রাফিক ১০× বেড়ে যায় এমন একটি e-learning startup-এর জন্য কোন DBaaS সবচেয়ে cost-effective?
   ✨ Show Answer

   Answer: A serverless Postgres — Neon, Supabase or Aurora Serverless v2. They scale up automatically during exam-night spikes and scale almost to zero in the off-hours, so the startup pays for actual usage rather than a 24×7 large instance. Neon's branch feature also makes preview environments per pull request trivial.

   Serverless Postgres (Neon / Aurora Serverless v2) — peak-এ auto-scale up, off-peak-এ প্রায় শূন্য, তাই bill কম।

3. Why is DynamoDB sometimes cheaper than Postgres at very high QPS?
   খুব বেশি QPS-এ DynamoDB কখনও কখনও Postgres-এর চেয়ে সস্তা কেন?
   ✨ Show Answer

   Answer: A Postgres instance has a fixed cost regardless of QPS, and you must over-provision for peak load. DynamoDB charges per request and scales horizontally with no instance to provision, so at predictable, very high QPS the per-request price often beats the cost of a similarly sized Postgres cluster — and it never wakes you at 3 a.m.

   Postgres instance fixed খরচ + peak-এর জন্য over-provision; DynamoDB per-request — তাই উচ্চ QPS-এ DynamoDB সস্তা হতে পারে।

4. Write the AWS CLI command to create an encrypted, multi-AZ Postgres 16 RDS instance with 7-day backups.
   Encrypted multi-AZ Postgres 16 RDS instance তৈরি করার AWS CLI কমান্ড লিখুন।
   ✨ Show Answer

   ans4.sh

   aws rds create-db-instance \
     --db-instance-identifier myapp-prod \
     --engine postgres --engine-version 16.3 \
     --db-instance-class db.t4g.medium \
     --allocated-storage 20 \
     --master-username dbadmin --master-user-password "$DB_PWD" \
     --multi-az --storage-encrypted \
     --backup-retention-period 7

5. Your CTO suggests Spanner for a Bangladeshi food-delivery startup running entirely in one country. Why might that be overkill?
   শুধু বাংলাদেশে চলা food-delivery startup-এর জন্য Spanner কেন বেশি?
   ✨ Show Answer

   Answer: Spanner's value is global, externally-consistent multi-region writes — paid for with higher latency and far higher cost. A single-country app gets the same correctness from a regional Postgres at a fraction of the price. Spanner only pays off when active-active multi-region writes are a hard requirement.

   Spanner-এর মূল মূল্য বহু-region active-active write — শুধু বাংলাদেশে চললে regional Postgres-ই যথেষ্ট, খরচ অনেক কম।

6. Name two distinct ways BigQuery's pricing can surprise a careless engineer.
   BigQuery-র pricing-এ কোন দুটি অপ্রত্যাশিত কারণে bill বেড়ে যেতে পারে?
   ✨ Show Answer

   Answer: (1) SELECT * on a wide table scans every column, billing per byte — even though the user only needed two columns. (2) Querying without partition pruning (WHERE _PARTITIONTIME > …) re-scans the entire historical dataset every time, multiplying cost by years of accumulated data.

   (১) SELECT * — সব column scan হয়, bill বাড়ে। (২) Partition filter ছাড়া query — পুরো history scan হয়, খরচ বহুগুণ।

7. Which managed database would you pick for a Cloudflare-Workers app that must respond in < 50 ms from Singapore, Frankfurt and Sao Paulo?
   Singapore, Frankfurt, Sao Paulo — সব region থেকে < 50 ms response চাইলে কোন DB?
   ✨ Show Answer

   Answer: Turso (libSQL) or Cloudflare D1 — both replicate SQLite to many edge regions, giving sub-50-ms reads worldwide. For writes only one primary region exists, but for read-heavy edge apps that is the right trade.

   Turso অথবা Cloudflare D1 — SQLite globally replicate করে, edge থেকে দ্রুত read পাওয়া যায়।

8. Explain "egress fee" in one paragraph, with an example.
   Egress fee এক অনুচ্ছেদে উদাহরণসহ বুঝিয়ে বলুন।
   ✨ Show Answer

   Answer: Cloud providers charge for data leaving their network — typically 5–9 cents per GB. Hosting your database on AWS in Mumbai but serving an analytics dashboard from a Singapore VPS means every query result crosses regions and racks up egress fees. Often this single line item is bigger than the database compute cost itself, and is the most common "Why is our cloud bill so high?" answer.

   Cloud-এর বাইরে data নেওয়ার fee — সাধারণত প্রতি GB-তে ৫-৯ সেন্ট। DB এক region-এ, app অন্য region-এ থাকলে এই fee অনেক বেড়ে যায়।

9. A Bangladesh-Bank-regulated mobile-banking workload requires data to stay inside Bangladesh. How does that change your DBaaS choice?
   Bangladesh Bank-নিয়ন্ত্রিত workload — DBaaS পছন্দে এটি কীভাবে প্রভাব ফেলে?
   ✨ Show Answer

   Answer: Compliance overrides convenience. None of the major hyperscalers run a region inside Bangladesh today, so a fully managed AWS/GCP/Azure database is usually disqualified. The realistic options are: self-hosted Postgres on a local data centre, a regulated on-shore cloud (such as a local provider's managed offering), or Oracle/Microsoft solutions deployed on premises with a vendor support contract. Always check the latest BB Cyber Security Guidelines before committing.

   Compliance > convenience। বাংলাদেশে major cloud-এর region নেই, তাই self-hosted Postgres বা local data-center-ভিত্তিক managed service বেছে নিতে হবে। সর্বশেষ BB-র Cyber Security Guidelines যাচাই করা আবশ্যক।

10. In one sentence: when should you NOT use a serverless database?
    এক বাক্যে — কখন serverless DB ব্যবহার করা ঠিক নয়?
    ✨ Show Answer

    Answer: When traffic is high and steady — at constant heavy load, a provisioned reserved instance is usually cheaper, and you avoid the cold-start latency some serverless engines incur after idle periods.

    যখন traffic ধারাবাহিকভাবে বেশি — তখন provisioned reserved instance সস্তা ও cold-start-হীন।