Why this matters
- Wrong store choices surface as rewrite projects: hot partitions, missing joins, or six-figure managed bills.
- System design interviews expect a justified recommendation, not a technology laundry list.
- VaultCommerce runs five engines today because each solves one problem well — the skill is knowing which problem you're solving.
SQL (Postgres)
ACID transactions
Complex joins
Fixed schema
NoSQL
Flexible schema
Horizontal scale
Specialised access
The five-question framework
Work through these dimensions before committing:
Selection checklist
- Data shape — relational rows, nested documents, time-series events, graph edges?
- Access pattern — point lookups, range scans, full-text search, aggregations?
- Consistency — must checkout be strongly consistent, or is eventual OK for a feed?
- Scale trajectory — write QPS, storage growth, multi-region needs in three years?
- Team and ops — managed service, existing expertise, on-call capacity?
# VaultCommerce checkout service sketch
workload: order_placement
entities: [users, carts, inventory, payments]
hot_query: "deduct inventory and charge card atomically"
read_write_ratio: 1:3
consistency: strong
scale_3yr: 50k orders/min peak
recommendation: postgres
The rest of this guide is part of Pro
Free covers every fresher guide in full. Pro unlocks Senior, Staff and Architect guides across every track.
Still to come · 3 sections
- Mapping VaultCommerce workloads
- Managed vs self-hosted
- Red flags that force a rethink
Test yourself
Answer these before moving on — recall is what makes it stick.