Database decomposition strategy in Microservices - System Design Exercise

Design: Database Decomposition Strategy for Microservices

Focus on strategies to split a monolithic database into multiple databases owned by microservices. Out of scope: detailed microservice business logic, UI design, or deployment infrastructure.

Functional Requirements

FR1: Support multiple microservices each owning its own data

FR2: Ensure data consistency within each microservice

FR3: Allow microservices to evolve independently

FR4: Enable efficient querying within each microservice

FR5: Support communication between microservices for cross-service data needs

Non-Functional Requirements

NFR1: Handle up to 10,000 concurrent requests across services

NFR2: API response latency p99 under 300ms

NFR3: Availability target of 99.9% uptime

NFR4: Data consistency within a microservice must be strong; eventual consistency allowed across services

NFR5: Data storage must be scalable and maintainable

Think Before You Design

Questions to Ask

❓ Question 1

❓ Question 2

❓ Question 3

❓ Question 4

❓ Question 5

Key Components

Service-specific databases

API gateways or service mesh for communication

Event-driven messaging or asynchronous communication

Data replication or caching for cross-service queries

Database per service pattern

Design Patterns

Database per service

Shared database anti-pattern

Event sourcing and CQRS for cross-service data

Saga pattern for distributed transactions

API composition for aggregating data

Reference Architecture

                +---------------------+
                |   API Gateway       |
                +----------+----------+
                           |
        +------------------+------------------+
        |                                     |
+-------v-------+                     +-------v-------+
| User Service  |                     | Order Service |
| +-----------+|                     | +-----------+|
| | User DB   ||                     | | Order DB  ||
| +-----------+|                     | +-----------+|
+---------------+                   +---------------+
        |                                     |
        +------------------+------------------+
                           |
                +----------v----------+
                | Event Bus / Message  |
                | Queue               |
                +---------------------+

Components

API Gateway

Nginx or Kong

Route client requests to appropriate microservices

User Service

Node.js/Java/Spring Boot

Handles user-related business logic and owns User database

Order Service

Node.js/Java/Spring Boot

Handles order-related business logic and owns Order database

User DB

PostgreSQL or MongoDB

Stores user data, owned exclusively by User Service

Order DB

PostgreSQL or MongoDB

Stores order data, owned exclusively by Order Service

Event Bus / Message Queue

Kafka or RabbitMQ

Facilitates asynchronous communication and eventual consistency between services

Request Flow

1. Client sends request to API Gateway

2. API Gateway routes request to appropriate microservice

3. Microservice processes request using its own database

4. If data needed by another service, microservice publishes event to Event Bus

5. Other microservices subscribe to relevant events and update their own data or caches

6. Microservices respond back to API Gateway

7. API Gateway sends response to client

Database Schema

Entities are split by service ownership. For example, User Service owns User entity with attributes like user_id, name, email. Order Service owns Order entity with order_id, user_id (foreign key reference by ID only, not enforced by DB), product details, status. No shared tables. Relationships across services handled via events and IDs.

Scaling Discussion

Bottlenecks

Single database per service can become a bottleneck under heavy load

Cross-service data queries can be slow or inconsistent

Distributed transactions are complex and can reduce performance

Eventual consistency may cause stale data issues

Message queue can become a bottleneck if overloaded

Solutions

Scale databases vertically or horizontally (read replicas, sharding) per service

Use caching and CQRS pattern to optimize cross-service queries

Implement Saga pattern to manage distributed transactions reliably

Design services to tolerate eventual consistency and provide user feedback

Scale message queue cluster and partition topics to handle load

Interview Tips

Time: Spend 10 minutes understanding requirements and clarifying scope, 20 minutes designing the decomposition strategy and data flow, 10 minutes discussing scaling and trade-offs, 5 minutes summarizing.

Explain importance of bounded contexts and service ownership of data

Discuss pros and cons of database per service pattern

Highlight how asynchronous communication supports eventual consistency

Mention challenges of distributed transactions and solutions like Saga

Show awareness of scaling bottlenecks and mitigation strategies

Practice

1. Which of the following best describes vertical decomposition in database design for microservices?

easy

A. Dividing a database by rows to distribute data across multiple databases

B. Combining multiple databases into one large database

C. Separating databases based on geographic location

D. Splitting a database by grouping related tables or columns into separate databases

Solution

1. Step 1: Understand vertical decomposition

   Vertical decomposition means splitting a database by grouping related tables or columns, often by business capability or domain.

2. Step 2: Compare with other options

   Horizontal decomposition splits by rows, geographic is location-based, and combining is the opposite of decomposition.

3. Final Answer:

   Splitting a database by grouping related tables or columns into separate databases -> Option D

4. Quick Check:

   Vertical decomposition = splitting by columns/tables [OK]

Hint: Vertical = split by columns or tables, horizontal = split by rows [OK]

Common Mistakes:

• Confusing vertical with horizontal decomposition
• Thinking vertical means geographic split
• Assuming decomposition means combining databases

2. Which of the following is the correct description of horizontal decomposition in microservices database design?

easy

A. Dividing data by rows, such as by customer or region

B. Splitting data by columns or tables based on functionality

C. Merging multiple databases into one for simplicity

D. Separating databases by different database engines

Solution

1. Step 1: Define horizontal decomposition

   Horizontal decomposition splits data by rows, for example, dividing customers by region or user ID ranges.

2. Step 2: Eliminate incorrect options

   Splitting data by columns or tables based on functionality describes vertical decomposition, C is merging (not decomposition), and D is about engines, not decomposition strategy.

3. Final Answer:

   Dividing data by rows, such as by customer or region -> Option A

4. Quick Check:

   Horizontal decomposition = split by rows [OK]

Hint: Horizontal = split by rows, vertical = split by columns [OK]

Common Mistakes:

• Mixing horizontal with vertical decomposition
• Thinking horizontal means merging databases
• Confusing database engine separation with decomposition

3. Consider a microservices system where the user database is split by region using horizontal decomposition. If a query requests all users from Europe, which database(s) will be queried?

medium

A. Only the database shard containing European users

B. All database shards regardless of region

C. Only the database shard containing North American users

D. A combined database with all users merged

Solution

1. Step 1: Understand horizontal decomposition by region

   Horizontal decomposition splits data by rows, so each shard holds users from a specific region.

2. Step 2: Identify which shard to query

   Querying European users targets only the shard holding European data, not others.

3. Final Answer:

   Only the database shard containing European users -> Option A

4. Quick Check:

   Query targets relevant shard only [OK]

Hint: Query only the shard holding requested data region [OK]

Common Mistakes:

• Querying all shards unnecessarily
• Querying wrong region shard
• Assuming data is merged in one database

4. A microservices team decomposed their database vertically but notices frequent cross-service joins causing latency. What is the likely cause and fix?

medium

A. Cause: Using NoSQL instead of SQL; Fix: Switch to SQL databases

B. Cause: Horizontal decomposition; Fix: Merge databases into one

C. Cause: Poor vertical decomposition causing cross-service joins; Fix: Redesign to reduce cross-service dependencies

D. Cause: Too many database shards; Fix: Increase shards further

Solution

1. Step 1: Identify problem with vertical decomposition

   Vertical decomposition splits by tables/domains, but if services need to join data often, it causes latency.

2. Step 2: Recommend fix

   Redesign to reduce cross-service joins by better domain boundaries or data duplication to avoid latency.

3. Final Answer:

   Poor vertical decomposition causing cross-service joins; Fix: Redesign to reduce cross-service dependencies -> Option C

4. Quick Check:

   Cross-service joins cause latency; fix by better decomposition [OK]

Hint: Cross-service joins mean bad vertical split; redesign domains [OK]

Common Mistakes:

• Confusing horizontal with vertical decomposition issues
• Thinking merging databases fixes latency
• Blaming database type instead of design

5. A company wants to scale their microservices database by splitting user data by country (horizontal) and splitting user profile and orders into separate databases (vertical). What is the best approach to handle queries that need both profile and order data for users in a specific country?

hard

A. Perform cross-database joins directly on all shards for each country

B. Use API composition to aggregate data from profile and order services after querying country-specific shards

C. Merge profile and order data into a single database shard per country

D. Store all user data in one large database to avoid complexity

Solution

1. Step 1: Understand combined vertical and horizontal decomposition

   Data is split horizontally by country and vertically by data type (profile, orders), so data is in different shards and databases.

2. Step 2: Choose best query approach

   Cross-database joins are expensive and complex; merging data loses benefits. API composition aggregates data from services after querying relevant shards efficiently.

3. Final Answer:

   Use API composition to aggregate data from profile and order services after querying country-specific shards -> Option B

4. Quick Check:

   API composition handles multi-db queries efficiently [OK]

Hint: Use API composition to combine data from vertical and horizontal splits [OK]

Common Mistakes:

• Trying cross-database joins causing latency
• Merging databases losing scalability
• Ignoring decomposition benefits for simplicity