Breaking Systems into Manageable Pieces - The LLD Mindset — Thinking in Components | OneNoughtOne

Low-Level DesignThe LLD Mindset

The LLD Mindset — Thinking in Components

LevelBeginner

Duration60 mins

TopicThe LLD Mindset

1 / 4

Breaking Systems into Manageable Pieces

The Essence of Component Thinking

Every complex software system that scales, survives, and evolves shares a common trait: it is built from well-defined, manageable pieces. This isn't an accident or a stylistic choice—it's a fundamental engineering principle that separates professional-grade software from fragile, monolithic codebases that collapse under their own weight.

At the heart of Low-Level Design lies a deceptively simple idea: decomposition. The ability to take a complex problem, break it into smaller sub-problems, solve each in isolation, and compose them back into a coherent whole. Yet, while the concept sounds straightforward, mastering it requires developing an intuition that only comes from understanding why decomposition matters and how to do it correctly.

What You Will Learn

By the end of this page, you will understand the principles of system decomposition, recognize the cognitive and engineering benefits of working with small pieces, learn systematic approaches for breaking down problems, and develop the intuition to identify natural component boundaries in any software system.

Why Decomposition Is Non-Negotiable

Before diving into techniques, we need to understand why decomposition isn't just a good practice but an absolute necessity for any non-trivial software project. The reasons are rooted in both human cognitive limits and fundamental engineering constraints.

The Cognitive Constraint: Miller's Law

Cognitive psychologist George Miller famously established that humans can hold approximately 7 ± 2 items in working memory simultaneously. This isn't a suggestion for software design—it's a hard constraint on what any individual developer can reason about at once.

Consider a codebase with 100,000 lines of code intertwined as a single, monolithic unit. No human can understand its complete state at any moment. Now consider the same functionality decomposed into 200 modules of 500 lines each, where each module has clear inputs, outputs, and responsibilities. A developer can fully understand any single module, trace its connections, and modify it confidently.

The 7 ± 2 Rule in Practice

When designing a class or module, aim for no more than 7 public methods. When designing a system, aim for no more than 7 major components interacting directly. This isn't a strict rule but a heuristic—if you're exceeding these numbers, consider whether you're asking too much of human comprehension.

The Engineering Constraint: Complexity Growth

As systems grow, the number of potential interactions between components grows quadratically. If you have n components that can all interact with each other, you have up to n(n-1)/2 possible connections.

Components	Potential Connections
5	10
10	45
20	190
50	1,225
100	4,950

This explosive growth in complexity is why poorly decomposed systems become unmanageable. Every change potentially affects thousands of interactions. Bugs become impossible to trace. Testing becomes infeasible.

The Complexity Explosion

A system with 100 tightly-coupled components has nearly 5,000 potential interactions to consider. No team can manage this. Decomposition with clear boundaries limits a component's interactions to just its immediate neighbors—reducing complexity from O(n²) to O(n).

The Team Constraint: Parallel Work

Software is rarely built by individuals. Teams work concurrently on different parts of a system. Without clear boundaries between components:

Two developers editing the same file create merge conflicts
Changes ripple unexpectedly across team boundaries
Integration becomes a nightmare of coordination
Ownership is unclear, leading to 'tragedy of the commons' where nobody takes responsibility

Proper decomposition enables parallel development where teams can work independently on their components, confident that if they maintain their interfaces, integration will succeed.

The Anatomy of a Well-Defined Component

A component, whether it's a class, a module, a service, or a subsystem, isn't just a random grouping of code. A well-defined component has specific properties that make it manageable, understandable, and maintainable.

The Five Essential Properties:

Properties of a Well-Defined Component

•Single, Clear Responsibility — The component does one thing and does it well. You can describe its purpose in a single sentence without using 'and' or 'or'.
•Well-Defined Interface — There's a clear contract specifying how to interact with the component. Inputs, outputs, and behaviors are documented or obvious from the interface.
•Hidden Implementation — The internal workings are invisible to users of the component. Changes inside don't break code outside.
•Minimal Dependencies — The component needs as few other components as possible. Each dependency is explicit and justified.
•Self-Containment — The component makes sense on its own. You can understand, test, and deploy it without understanding the entire system.

A Practical Example: Email Sender

Consider a component responsible for sending emails in an e-commerce system. Here's how the five properties apply:

EmailSender Component Analysis
Property	Good Design	Poor Design
Single Responsibility	Sends emails via configured SMTP/API	Sends emails AND logs analytics AND manages templates AND handles bounce tracking
Well-Defined Interface	`send(to, subject, body, options)` with clear types	Dozens of methods with unclear relationships
Hidden Implementation	Could switch from SMTP to SendGrid without external changes	External code depends on SMTP connection details
Minimal Dependencies	Only needs config and maybe a template engine	Depends on user database, order system, logging framework, analytics...
Self-Containment	Can be unit tested with a mock SMTP server	Requires full system running to test

The Litmus Test

Can you explain what this component does to a colleague in one breath? Can you write a unit test for it without mocking half the system? If the answers are no, the component isn't well-defined—it probably needs further decomposition.

Strategies for Breaking Down Systems

Decomposition isn't random. There are established strategies for identifying natural boundaries in a system. Understanding these strategies and knowing when to apply each is a core LLD skill.

Strategy 1: Functional Decomposition

The most intuitive approach: break the system into pieces based on what they do. Each component handles a distinct function.

Example: E-commerce Order Processing

OrderValidator — Validates order data, checks inventory
PaymentProcessor — Handles payment authorization and capture
InventoryReserver — Reserves items, manages allocation
NotificationManager — Sends confirmation emails/SMS
ShippingInitiator — Creates shipping labels, schedules pickup

Strategy 2: Data-Centric Decomposition

Group components around the data they manage. This is particularly powerful when certain data naturally clusters together and has specific access patterns.

Example: Social Media Platform

UserProfileManager — All user profile data (bio, settings, preferences)
PostRepository — Posts, their metadata, edit history
RelationshipGraph — Follows, blocks, friends
MediaLibrary — Images, videos, thumbnails
NotificationStore — Notification history and preferences

Each component owns its data entirely—no other component directly accesses that data without going through the owning component.

Strategy 3: Layer-Based Decomposition

Organize components into horizontal layers where each layer has a specific abstraction level. Higher layers depend on lower layers, never the reverse.

Classic Three-Layer Architecture:

Presentation Layer — UI, API endpoints, user interaction
Business Logic Layer — Domain rules, workflows, validations
Data Access Layer — Database queries, external API calls, caching

Extended Modern Layers:

Interface Adapters — Controllers, presenters, gateways
Application Services — Use cases, orchestration
Domain Layer — Core business entities and rules
Infrastructure — Databases, message queues, external services

layer-dependencies.txt

Dependency Direction in Layered Architecture:
 
┌─────────────────────────────────┐
│       Presentation Layer        │  ─── Depends on ───┐
└─────────────────────────────────┘                     │
                                                        ▼
┌─────────────────────────────────┐
│      Business Logic Layer       │  ─── Depends on ───┐
└─────────────────────────────────┘                     │
                                                        ▼
┌─────────────────────────────────┐
│        Data Access Layer        │
└─────────────────────────────────┘
 
Key Rule: Dependencies flow DOWNWARD only.
The Presentation Layer NEVER depends on Data Access directly.
The Business Layer NEVER depends on Presentation.

Strategy 4: Feature/Vertical Slice Decomposition

Instead of horizontal layers, slice the system vertically by feature. Each slice contains all layers needed for that feature.

Example: E-commerce System Vertical Slices:

Checkout Slice — Checkout UI + checkout logic + checkout data
Catalog Slice — Browse UI + product logic + product data
Account Slice — Profile UI + user logic + user data

This approach excels when features are largely independent and teams own entire features.

Choosing a Strategy

No single strategy is universally best. Often, you'll combine strategies—layer-based decomposition at the macro level, functional decomposition within each layer, and data-centric boundaries around core entities. The key is consistency within a system and clear documentation of the approach.

The Art of Right-Sizing Components

One of the most challenging aspects of decomposition is determining the right size for components. Too large, and you haven't solved the complexity problem. Too small, and you've created a different complexity problem—too many moving pieces with too much coordination overhead.

Signs a Component Is Too Large:

Component Too Large

•You struggle to describe its purpose concisely
•The class/module has many unrelated methods
•Different parts change for different reasons
•Testing requires complex setup with many dependencies
•New team members take weeks to understand it
•Changes frequently cause unexpected side effects
•The file is thousands of lines long

Component Too Small

•Almost all classes are just data holders with no logic
•Many one-method classes that just delegate
•Reading code requires jumping through dozens of files
•Tracing a single operation touches 20+ components
•Interface count vastly exceeds implementation count
•Most 'abstractions' have exactly one implementation
•The overhead of wiring exceeds the logic itself

The Goldilocks Zone

A well-sized component typically has:

5-15 public methods for a class
100-500 lines of code for a single file
3-7 direct dependencies for a module
A name that implies a single domain concept

These aren't rigid rules—they're starting points. A utility class might have 50 methods. A complex algorithm might fill 1,000 lines. But if you consistently exceed these ranges, examine why.

The Over-Engineering Trap

Excessive decomposition creates its own complexity. If you need a 'factory for factories' or 'managers managing managers,' you've gone too far. The goal is clarity, not abstraction for its own sake. Always ask: 'Is this decomposition making the code easier or harder to understand?'

Practical Heuristic: The 'Fitting in Your Head' Test

Ask yourself: Can a single developer, with moderate domain knowledge, fully understand this component in 15-30 minutes? This includes:

All its public methods and their contracts
Its internal state and invariants
How it fulfills its responsibilities
Its error conditions and edge cases

If the answer is no, the component is too complex and likely needs decomposition.

Identifying Natural Boundaries

Not all decomposition points are equal. Some are arbitrary—you could draw the line here or there without much consequence. Others are natural boundaries where the system almost demands a separation. Learning to recognize natural boundaries is a hallmark of experienced designers.

Signals of Natural Boundaries:

Natural Boundary Indicators

•Rate of Change — Parts of the system that change together belong together. Parts that change at different rates should be separate. UI changes weekly; database schema changes monthly.
•Reason for Change — Components should change for one reason. If a module changes when business rules change AND when the database changes AND when the UI changes, it's crossing too many boundaries.
•Team Ownership — Natural boundaries often align with team structures. If different teams work on different parts, those parts should be separate components with clear interfaces.
•Technology Differences — When you cross technologies (web to mobile, Python to Java, SQL to NoSQL), place a boundary with a clean interface.
•Domain Language Shifts — When the vocabulary changes ('user' becomes 'patient' becomes 'claimant'), you've likely crossed domain boundaries.
•Security/Trust Levels — When data crosses trust boundaries (authenticated vs public, internal vs external), explicit boundaries are mandatory.

The Domain-Driven Design Perspective

Domain-Driven Design (DDD) provides powerful tools for boundary identification:

Bounded Contexts — Parts of the system where a particular domain model applies. Within a bounded context, terms have precise meanings. Across boundaries, translation is needed.

Example: 'Customer' in Sales vs. Support

Sales Context: Customer has leads, opportunities, purchase history
Support Context: Customer has tickets, satisfaction scores, entitlements

These are different models that happen to share a name. They belong in separate components with an explicit integration layer.

bounded-context-example.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
// WRONG: Mixing contexts creates confusion
interface Customer {
    // Sales properties
    leadScore: number;
    opportunities: Opportunity[];
    lastPurchase: Date;
    
    // Support properties
    openTickets: Ticket[];
    satisfactionScore: number;
    supportTier: string;
}
 
// RIGHT: Separate bounded contexts with explicit mapping
// sales/customer.ts
interface SalesCustomer {
    id: string;
    leadScore: number;
    opportunities: Opportunity[];
    lastPurchase: Date;
}
 
// support/customer.ts
interface SupportCustomer {
    id: string;
    openTickets: Ticket[];
    satisfactionScore: number;
    supportTier: string;
}
 
// integration/customer-mapper.ts
function mapSalesToSupport(salesCustomer: SalesCustomer): SupportCustomer {
    return {
        id: salesCustomer.id,
        openTickets: [],
        satisfactionScore: inferFromPurchaseHistory(salesCustomer),
        supportTier: determineTierFromPurchases(salesCustomer),
    };
}

Natural Doesn't Mean Obvious

Natural boundaries emerge from the domain and the way the system evolves. They're 'natural' in that fighting them causes friction—but they're not always obvious at first. Experienced designers develop intuition through practice and refactoring when initial boundaries prove wrong.

Decomposition Anti-Patterns

Understanding what not to do is as valuable as knowing best practices. Here are common decomposition anti-patterns that lead to systems worse than the monolith they were meant to replace.

Anti-Pattern 1: The God Class

A single class that knows everything, controls everything, and does everything. It accumulates responsibilities over time as developers add 'just one more method' for convenience.

Symptoms:

File is thousands of lines
Has 50+ public methods
Name ends in 'Manager,' 'Handler,' 'Utils,' or 'Service' without qualification
Virtually every other class depends on it

Anti-Pattern 2: Nanoservices

The opposite extreme—breaking every tiny piece into a separate component. The overhead of communication, deployment, and coordination exceeds the complexity of the actual logic.

Symptoms:

Dozens of classes with one method each
'Thin' interfaces that just pass data through
Every operation requires tracing through 10+ files
Integration is more complex than the features

Anti-Pattern 3: The Distributed Monolith

The code is split into many components or services, but they're all tightly coupled. You can't deploy one without deploying all. Changes ripple everywhere. You have the complexity of distribution without the benefits.

Symptoms:

Deploying a 'microservice' requires deploying three others
Schema changes in one service break multiple others
A single business transaction touches most services
There's no stable interface; internal data structures are shared

Anti-Pattern 4: Anemic Domain Model

Components that are just data holders (DTOs) with all logic in separate 'service' classes. This violates encapsulation and leads to logic scattered across the codebase.

Symptoms:

Domain objects have only getters/setters, no methods
All business rules live in 'XService' classes
Same validation logic duplicated in multiple services
No cohesion between related data and behavior

Recognize and Refactor

If you recognize these patterns in your codebase, don't despair—nearly every system has them somewhere. The key is recognizing them as technical debt and systematically refactoring toward better boundaries when the opportunity arises.

Putting It into Practice

Let's apply these principles to a concrete example. Consider designing a Library Management System. We'll walk through the decomposition thought process.

Requirements Overview:

Manage books (add, remove, update, search)
Manage members (registration, profile, borrowing limits)
Handle borrowing/returning books
Send notifications (overdue reminders, reservation available)
Track fines for late returns

Step 1: Identify Core Entities

What are the main 'things' the system deals with?

Book
Member
BorrowRecord (transaction linking book and member)
Reservation
Fine
Notification

Step 2: Group by Responsibility

Applying functional decomposition:

Component	Responsibility	Entities Managed
CatalogManager	Book inventory, availability, search	Book
MemberManager	Member registration, profiles, limits	Member
BorrowingService	Checkout, returns, renewals	BorrowRecord
ReservationService	Reservations, queue management	Reservation
FineCalculator	Overdue detection, fine computation	Fine
NotificationService	Message sending, scheduling	Notification

Step 3: Define Interfaces and Dependencies

Each component exposes a clean interface. Dependencies are explicit:

component-interfaces.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
// Catalog Interface
interface ICatalogManager {
    addBook(book: BookDetails): Book;
    removeBook(bookId: string): void;
    searchBooks(criteria: SearchCriteria): Book[];
    getAvailability(bookId: string): AvailabilityStatus;
}
 
// Borrowing Interface
interface IBorrowingService {
    checkout(memberId: string, bookId: string): BorrowRecord;
    returnBook(recordId: string): Fine | null;
    renew(recordId: string): BorrowRecord;
    getBorrowHistory(memberId: string): BorrowRecord[];
}
 
// Dependencies are injected, making them explicit
class BorrowingService implements IBorrowingService {
    constructor(
        private catalog: ICatalogManager,
        private members: IMemberManager,
        private fineCalculator: IFineCalculator,
        private notifications: INotificationService
    ) {}
    
    checkout(memberId: string, bookId: string): BorrowRecord {
        // Uses catalog to check availability
        // Uses members to verify borrowing limits
        // Could trigger a notification on success
    }
    
    returnBook(recordId: string): Fine | null {
        // Uses fineCalculator to compute any fine
        // Updates catalog availability
    }
}

Step 4: Validate the Decomposition

Ask the critical questions:

✅ Single Responsibility? Each component has one clear purpose
✅ Testable in Isolation? BorrowingService can be tested with mock dependencies
✅ Right Size? Each component has 5-10 methods, not 50
✅ Minimal Dependencies? BorrowingService needs 4 dependencies, all justified
✅ Change for One Reason? Fine calculation changes don't affect book search

This decomposition passes our checks. In implementation, you might refine further—but the core structure is sound.

Practice Makes Permanent

Decomposition is a skill that improves with practice. For every new system you design or codebase you join, ask: 'How is this decomposed? Why? What are the boundaries?' Train yourself to think in components, and it becomes instinctive.

Summary: Breaking Systems into Pieces

We've covered the foundational skill of system decomposition. Let's consolidate the essential lessons:

Key Takeaways

•Decomposition is non-negotiable — Human cognition and complexity growth make it essential for any non-trivial system.
•Well-defined components have five properties — Single responsibility, well-defined interface, hidden implementation, minimal dependencies, and self-containment.
•Multiple decomposition strategies exist — Functional, data-centric, layer-based, and vertical slices. Choose based on context and combine as needed.
•Size matters — Components too large are incomprehensible; too small creates coordination overhead. Aim for the Goldilocks zone.
•Natural boundaries exist — Look for rate of change, reason for change, team ownership, technology shifts, and domain language shifts.
•Avoid anti-patterns — God classes, nanoservices, distributed monoliths, and anemic domain models all indicate decomposition failures.

What's Next:

Breaking a system into pieces is only the first step. Next, we need to ensure those pieces have clear responsibilities and boundaries—that each component owns exactly what it should, and no more. That's what we'll explore on the next page.

Page Complete

You now understand the principles and practices of breaking complex systems into manageable components. This is the foundation of the LLD mindset—think in pieces, design interfaces, respect boundaries. Next, we'll dive deeper into how to identify and assign responsibilities within those boundaries.

1 / 4

Loading learning content...

Low-Level DesignThe LLD Mindset

The LLD Mindset — Thinking in Components

LevelBeginner

Duration60 mins

TopicThe LLD Mindset

1 / 4

Breaking Systems into Manageable Pieces

The Essence of Component Thinking

What You Will Learn

Why Decomposition Is Non-Negotiable

The Cognitive Constraint: Miller's Law

The 7 ± 2 Rule in Practice

The Engineering Constraint: Complexity Growth

Components	Potential Connections
5	10
10	45
20	190
50	1,225
100	4,950

The Complexity Explosion

The Team Constraint: Parallel Work

Software is rarely built by individuals. Teams work concurrently on different parts of a system. Without clear boundaries between components:

Two developers editing the same file create merge conflicts
Changes ripple unexpectedly across team boundaries
Integration becomes a nightmare of coordination
Ownership is unclear, leading to 'tragedy of the commons' where nobody takes responsibility

Proper decomposition enables parallel development where teams can work independently on their components, confident that if they maintain their interfaces, integration will succeed.

The Anatomy of a Well-Defined Component

The Five Essential Properties:

Properties of a Well-Defined Component

•Single, Clear Responsibility — The component does one thing and does it well. You can describe its purpose in a single sentence without using 'and' or 'or'.
•Well-Defined Interface — There's a clear contract specifying how to interact with the component. Inputs, outputs, and behaviors are documented or obvious from the interface.
•Hidden Implementation — The internal workings are invisible to users of the component. Changes inside don't break code outside.
•Minimal Dependencies — The component needs as few other components as possible. Each dependency is explicit and justified.
•Self-Containment — The component makes sense on its own. You can understand, test, and deploy it without understanding the entire system.

A Practical Example: Email Sender

Consider a component responsible for sending emails in an e-commerce system. Here's how the five properties apply:

EmailSender Component Analysis
Property	Good Design	Poor Design
Single Responsibility	Sends emails via configured SMTP/API	Sends emails AND logs analytics AND manages templates AND handles bounce tracking
Well-Defined Interface	`send(to, subject, body, options)` with clear types	Dozens of methods with unclear relationships
Hidden Implementation	Could switch from SMTP to SendGrid without external changes	External code depends on SMTP connection details
Minimal Dependencies	Only needs config and maybe a template engine	Depends on user database, order system, logging framework, analytics...
Self-Containment	Can be unit tested with a mock SMTP server	Requires full system running to test

The Litmus Test

Strategies for Breaking Down Systems

Decomposition isn't random. There are established strategies for identifying natural boundaries in a system. Understanding these strategies and knowing when to apply each is a core LLD skill.

Strategy 1: Functional Decomposition

The most intuitive approach: break the system into pieces based on what they do. Each component handles a distinct function.

Example: E-commerce Order Processing

OrderValidator — Validates order data, checks inventory
PaymentProcessor — Handles payment authorization and capture
InventoryReserver — Reserves items, manages allocation
NotificationManager — Sends confirmation emails/SMS
ShippingInitiator — Creates shipping labels, schedules pickup

Strategy 2: Data-Centric Decomposition

Group components around the data they manage. This is particularly powerful when certain data naturally clusters together and has specific access patterns.

Example: Social Media Platform

UserProfileManager — All user profile data (bio, settings, preferences)
PostRepository — Posts, their metadata, edit history
RelationshipGraph — Follows, blocks, friends
MediaLibrary — Images, videos, thumbnails
NotificationStore — Notification history and preferences

Each component owns its data entirely—no other component directly accesses that data without going through the owning component.

Strategy 3: Layer-Based Decomposition

Organize components into horizontal layers where each layer has a specific abstraction level. Higher layers depend on lower layers, never the reverse.

Classic Three-Layer Architecture:

Presentation Layer — UI, API endpoints, user interaction
Business Logic Layer — Domain rules, workflows, validations
Data Access Layer — Database queries, external API calls, caching

Extended Modern Layers:

Interface Adapters — Controllers, presenters, gateways
Application Services — Use cases, orchestration
Domain Layer — Core business entities and rules
Infrastructure — Databases, message queues, external services

layer-dependencies.txt

Dependency Direction in Layered Architecture:
 
┌─────────────────────────────────┐
│       Presentation Layer        │  ─── Depends on ───┐
└─────────────────────────────────┘                     │
                                                        ▼
┌─────────────────────────────────┐
│      Business Logic Layer       │  ─── Depends on ───┐
└─────────────────────────────────┘                     │
                                                        ▼
┌─────────────────────────────────┐
│        Data Access Layer        │
└─────────────────────────────────┘
 
Key Rule: Dependencies flow DOWNWARD only.
The Presentation Layer NEVER depends on Data Access directly.
The Business Layer NEVER depends on Presentation.

Strategy 4: Feature/Vertical Slice Decomposition

Instead of horizontal layers, slice the system vertically by feature. Each slice contains all layers needed for that feature.

Example: E-commerce System Vertical Slices:

Checkout Slice — Checkout UI + checkout logic + checkout data
Catalog Slice — Browse UI + product logic + product data
Account Slice — Profile UI + user logic + user data

This approach excels when features are largely independent and teams own entire features.

Choosing a Strategy

The Art of Right-Sizing Components

Signs a Component Is Too Large:

Component Too Large

•You struggle to describe its purpose concisely
•The class/module has many unrelated methods
•Different parts change for different reasons
•Testing requires complex setup with many dependencies
•New team members take weeks to understand it
•Changes frequently cause unexpected side effects
•The file is thousands of lines long

Component Too Small

•Almost all classes are just data holders with no logic
•Many one-method classes that just delegate
•Reading code requires jumping through dozens of files
•Tracing a single operation touches 20+ components
•Interface count vastly exceeds implementation count
•Most 'abstractions' have exactly one implementation
•The overhead of wiring exceeds the logic itself

The Goldilocks Zone

A well-sized component typically has:

5-15 public methods for a class
100-500 lines of code for a single file
3-7 direct dependencies for a module
A name that implies a single domain concept

These aren't rigid rules—they're starting points. A utility class might have 50 methods. A complex algorithm might fill 1,000 lines. But if you consistently exceed these ranges, examine why.

The Over-Engineering Trap

Practical Heuristic: The 'Fitting in Your Head' Test

Ask yourself: Can a single developer, with moderate domain knowledge, fully understand this component in 15-30 minutes? This includes:

All its public methods and their contracts
Its internal state and invariants
How it fulfills its responsibilities
Its error conditions and edge cases

If the answer is no, the component is too complex and likely needs decomposition.

Identifying Natural Boundaries

Signals of Natural Boundaries:

Natural Boundary Indicators

•Rate of Change — Parts of the system that change together belong together. Parts that change at different rates should be separate. UI changes weekly; database schema changes monthly.
•Reason for Change — Components should change for one reason. If a module changes when business rules change AND when the database changes AND when the UI changes, it's crossing too many boundaries.
•Team Ownership — Natural boundaries often align with team structures. If different teams work on different parts, those parts should be separate components with clear interfaces.
•Technology Differences — When you cross technologies (web to mobile, Python to Java, SQL to NoSQL), place a boundary with a clean interface.
•Domain Language Shifts — When the vocabulary changes ('user' becomes 'patient' becomes 'claimant'), you've likely crossed domain boundaries.
•Security/Trust Levels — When data crosses trust boundaries (authenticated vs public, internal vs external), explicit boundaries are mandatory.

The Domain-Driven Design Perspective

Domain-Driven Design (DDD) provides powerful tools for boundary identification:

Bounded Contexts — Parts of the system where a particular domain model applies. Within a bounded context, terms have precise meanings. Across boundaries, translation is needed.

Example: 'Customer' in Sales vs. Support

Sales Context: Customer has leads, opportunities, purchase history
Support Context: Customer has tickets, satisfaction scores, entitlements

These are different models that happen to share a name. They belong in separate components with an explicit integration layer.

bounded-context-example.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
// WRONG: Mixing contexts creates confusion
interface Customer {
    // Sales properties
    leadScore: number;
    opportunities: Opportunity[];
    lastPurchase: Date;
    
    // Support properties
    openTickets: Ticket[];
    satisfactionScore: number;
    supportTier: string;
}
 
// RIGHT: Separate bounded contexts with explicit mapping
// sales/customer.ts
interface SalesCustomer {
    id: string;
    leadScore: number;
    opportunities: Opportunity[];
    lastPurchase: Date;
}
 
// support/customer.ts
interface SupportCustomer {
    id: string;
    openTickets: Ticket[];
    satisfactionScore: number;
    supportTier: string;
}
 
// integration/customer-mapper.ts
function mapSalesToSupport(salesCustomer: SalesCustomer): SupportCustomer {
    return {
        id: salesCustomer.id,
        openTickets: [],
        satisfactionScore: inferFromPurchaseHistory(salesCustomer),
        supportTier: determineTierFromPurchases(salesCustomer),
    };
}

Natural Doesn't Mean Obvious

Decomposition Anti-Patterns

Understanding what not to do is as valuable as knowing best practices. Here are common decomposition anti-patterns that lead to systems worse than the monolith they were meant to replace.

Anti-Pattern 1: The God Class

A single class that knows everything, controls everything, and does everything. It accumulates responsibilities over time as developers add 'just one more method' for convenience.

Symptoms:

File is thousands of lines
Has 50+ public methods
Name ends in 'Manager,' 'Handler,' 'Utils,' or 'Service' without qualification
Virtually every other class depends on it

Anti-Pattern 2: Nanoservices

The opposite extreme—breaking every tiny piece into a separate component. The overhead of communication, deployment, and coordination exceeds the complexity of the actual logic.

Symptoms:

Dozens of classes with one method each
'Thin' interfaces that just pass data through
Every operation requires tracing through 10+ files
Integration is more complex than the features

Anti-Pattern 3: The Distributed Monolith

Symptoms:

Deploying a 'microservice' requires deploying three others
Schema changes in one service break multiple others
A single business transaction touches most services
There's no stable interface; internal data structures are shared

Anti-Pattern 4: Anemic Domain Model

Components that are just data holders (DTOs) with all logic in separate 'service' classes. This violates encapsulation and leads to logic scattered across the codebase.

Symptoms:

Domain objects have only getters/setters, no methods
All business rules live in 'XService' classes
Same validation logic duplicated in multiple services
No cohesion between related data and behavior

Recognize and Refactor

Putting It into Practice

Let's apply these principles to a concrete example. Consider designing a Library Management System. We'll walk through the decomposition thought process.

Requirements Overview:

Manage books (add, remove, update, search)
Manage members (registration, profile, borrowing limits)
Handle borrowing/returning books
Send notifications (overdue reminders, reservation available)
Track fines for late returns

Step 1: Identify Core Entities

What are the main 'things' the system deals with?

Book
Member
BorrowRecord (transaction linking book and member)
Reservation
Fine
Notification

Step 2: Group by Responsibility

Applying functional decomposition:

Component	Responsibility	Entities Managed
CatalogManager	Book inventory, availability, search	Book
MemberManager	Member registration, profiles, limits	Member
BorrowingService	Checkout, returns, renewals	BorrowRecord
ReservationService	Reservations, queue management	Reservation
FineCalculator	Overdue detection, fine computation	Fine
NotificationService	Message sending, scheduling	Notification

Step 3: Define Interfaces and Dependencies

Each component exposes a clean interface. Dependencies are explicit:

component-interfaces.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
// Catalog Interface
interface ICatalogManager {
    addBook(book: BookDetails): Book;
    removeBook(bookId: string): void;
    searchBooks(criteria: SearchCriteria): Book[];
    getAvailability(bookId: string): AvailabilityStatus;
}
 
// Borrowing Interface
interface IBorrowingService {
    checkout(memberId: string, bookId: string): BorrowRecord;
    returnBook(recordId: string): Fine | null;
    renew(recordId: string): BorrowRecord;
    getBorrowHistory(memberId: string): BorrowRecord[];
}
 
// Dependencies are injected, making them explicit
class BorrowingService implements IBorrowingService {
    constructor(
        private catalog: ICatalogManager,
        private members: IMemberManager,
        private fineCalculator: IFineCalculator,
        private notifications: INotificationService
    ) {}
    
    checkout(memberId: string, bookId: string): BorrowRecord {
        // Uses catalog to check availability
        // Uses members to verify borrowing limits
        // Could trigger a notification on success
    }
    
    returnBook(recordId: string): Fine | null {
        // Uses fineCalculator to compute any fine
        // Updates catalog availability
    }
}

Step 4: Validate the Decomposition

Ask the critical questions:

✅ Single Responsibility? Each component has one clear purpose
✅ Testable in Isolation? BorrowingService can be tested with mock dependencies
✅ Right Size? Each component has 5-10 methods, not 50
✅ Minimal Dependencies? BorrowingService needs 4 dependencies, all justified
✅ Change for One Reason? Fine calculation changes don't affect book search

This decomposition passes our checks. In implementation, you might refine further—but the core structure is sound.

Practice Makes Permanent

Summary: Breaking Systems into Pieces

We've covered the foundational skill of system decomposition. Let's consolidate the essential lessons:

Key Takeaways

•Decomposition is non-negotiable — Human cognition and complexity growth make it essential for any non-trivial system.
•Well-defined components have five properties — Single responsibility, well-defined interface, hidden implementation, minimal dependencies, and self-containment.
•Multiple decomposition strategies exist — Functional, data-centric, layer-based, and vertical slices. Choose based on context and combine as needed.
•Size matters — Components too large are incomprehensible; too small creates coordination overhead. Aim for the Goldilocks zone.
•Natural boundaries exist — Look for rate of change, reason for change, team ownership, technology shifts, and domain language shifts.
•Avoid anti-patterns — God classes, nanoservices, distributed monoliths, and anemic domain models all indicate decomposition failures.

What's Next:

Page Complete

1 / 4