Modular monolith

Modular Monolith

A Modular Monolith is a single deployable unit (Monolith) where the code is structured into independent modules (like Microservices) with strict boundaries. In 2025, this is the recommended default architecture for most startups and medium-scale apps.

When to Use

• Starting a new project (Greenfield).
• Domain boundaries are not yet fully clear.
• Wanting the speed of simple deployment (one CI pipeline, one DB instance) but preventing "spaghetti code".
• Precursor to Microservices.

Quick Start

/src
  /Modules
    /Catalog       <-- Public Interface defined here
      /Core        <-- Internals (Classes, Logic) hidden
      /API         <-- Public Contracts (DTOs)
    /Ordering
      /Core
    /Shipping
      /Core
  /Shared          <-- Infrastructure, Event Bus

// Communication via In-Process Interfaces or Events
public class OrderService {
    private readonly ICatalogModule _catalog; // In-memory reference, but strict contract

    public async Task Checkout(string productId) {
        var product = await _catalog.GetProduct(productId); // Fast 0ms call
        // ...
    }
}

Core Concepts

Module Boundaries

Code in Module A cannot access internal classes of Module B. It can only use Module B's "Public API" (Interfaces/DTOs). Enforced by compiler tools (ArchUnit, NetArchTest).

Single Deployment

Modules are compiled together into one binary/container and deployed to one server/cluster. Simplifies Ops.

Data Isolation (Virtual)

Ideally, each module has its own DB schema (or at least different tables). Cross-module Joins are forbidden.

Common Patterns

In-Memory Events

Using a mediator (like MediatR in .NET or Spring Events) to decouple modules. Module A publishes OrderCreated, Module B listens. No Kafka needed (yet).

Internal APIs

Defining strict interfaces that act as "Gateways" between modules. Changing the internals of Module A doesn't break Module B as long as the interface holds.

Best Practices

Do:

• Force Architecture Tests (e.g., "Classes in Ordering cannot depend on Shipping").
• Separate Data Schemas logically (different schemas in Postgres).
• Treat it as "Microservices ready to be extracted".

Don't:

• Don't bypass boundaries "just this once".
• Don't share Domain Entities across modules (use integration DTOs).

Advantages over Microservices

• Zero Latency communication.
• Transactional Consistency (ACID) is easier (though ideally, avoid cross-module transactions).
• Refactoring is cheap (IDE "Rename" works globally).

References

• Modular Monoliths (Kamil Grzybek)
• The Majestic Monolith