Database-per-Service

Introduction

Database-per-service gives each microservice private data storage — no shared tables, no cross-service SQL joins. Integration happens via APIs and events, not foreign keys across databases.

The story

A "microservices" rewrite kept one Postgres schema — every service queried orders JOIN users. Deployments still coupled; it was a distributed monolith. Splitting schemas forced proper APIs and cut accidental coupling.

The business problem

Teams that skip disciplined Database-per-Service thinking pay in coupling, outages, and failed microservices interviews:

• Distributed ops: no tracing or health checks — hours to debug one checkout failure.
• Coupling: shared databases and libraries recreate monolith pain across the network.
• Deploy blast radius: one bad service deploy takes down unrelated domains.
• Team boundaries: services aligned to tech layers instead of business capabilities.

The problem teams faced

This lesson addresses:

• When and why Database-per-Service matters in decomposed architectures.
• How to define service boundaries and contracts under interview time pressure.
• Trade-offs vs alternatives — what staff engineers articulate in architecture reviews.
• Production patterns and failure modes in distributed systems.

Understanding the topic

Core idea: Database-per-Service in microservices architecture.

• Problem — distributed ops and coupling this topic addresses.
• Service design — boundaries, contracts, and data ownership.
• Trade-offs — sync vs async, consistency, deploy blast radius.
• Interview — how this appears in decomposition loops.

Internal architecture

Database-per-Service — microservices view:

Order Service → orders_db (Postgres)
User Service   → users_db (Postgres)
Catalog Svc    → catalog_db (Postgres + ES index)
   ↓
Order needs user name → call User API or cache denormalized snapshot
   ↓
Never: SELECT * FROM users.orders JOIN shared schema

Visual explanation

Three diagrams: service architecture, decomposition process, and operational lens:

Informative example

Example — Database-per-Service:

OrderPlaced event carries: orderId, userId, lineItems
Payment service stores payment records locally
Reporting: CDC → warehouse for cross-service analytics
Cross-service query at runtime = design smell → fix with CQRS read model

Execution workflow

Real-world use

Netflix pioneered microservices at scale with hundreds of services and chaos engineering. Amazon's two-pizza teams and service-oriented architecture shaped modern decomposition. Uber migrated from monolith to domain-aligned services for independent deploys. Spotify uses squad-aligned microservices with internal platform tooling. These journeys inform every pattern in this course.

Production case study

A "microservices" rewrite kept one Postgres schema — every service queried orders JOIN users. Deployments still coupled; it was a distributed monolith. Splitting schemas forced proper APIs and cut acc…

• Context: monolith pain or decomposition scenario from this lesson.
• Decision: service boundary and communication choices explained.
• Outcome: deploy frequency, incident isolation, or latency impact.

Trade-offs

• Pro: team autonomy and isolated failure domains.
• Con: distributed complexity — tracing, sagas, contract tests.
• Con: premature decomposition costs more than a modular monolith.

Decision framework

• Start modular monolith; extract when bounded context and team force are proven.
• Database-per-service — integrate via API and events, not shared schema.
• Document rejected alternatives — ADR or interview closing statement.

Best practices

• Align services to business capabilities, not technical layers.
• Draw sync vs async paths; propagate trace context on every hop.
• Health checks + readiness gates before traffic shift.

Anti-patterns to avoid

• Distributed monolith — shared DB, coupled deploys, synchronous chains everywhere.
• Nano-services — operational overhead exceeds team benefit.
• Shared libraries hiding domain coupling between teams.

Common mistakes

• Splitting before bounded contexts are clear — endless refactor.
• Cross-service transactions without saga or idempotency.

Debugging tips

• Follow one trace_id through the decomposition diagram.
• Ask "what couples these services?" for every sync call.

Optimization strategies

• Replace sync chains with domain events where latency allows.
• BFF to aggregate calls — don't make clients orchestrate services.

Common misconceptions

• Microservices ≠ always better — monolith wins for small teams and unclear domains.
• Interviews test decomposition judgment — not memorizing Netflix's service count.

Advanced interview questions

Summary

You can explain Database-per-Service in microservices interviews and production RFCs — with bounded contexts, data ownership, and resilience patterns. Teach it back without notes.