aj-geddes/feature-flag-system

Feature Flag System

Table of Contents

• Overview
• When to Use
• Quick Start
• Reference Guides
• Best Practices

Overview

Implement feature flags to decouple deployment from release, enable gradual rollouts, A/B testing, and provide emergency kill switches.

When to Use

• Gradual feature rollouts
• A/B testing and experiments
• Canary deployments
• Beta features for specific users
• Emergency kill switches
• Trunk-based development
• Dark launching
• Operational flags (maintenance mode)
• User-specific features

Quick Start

Minimal working example:

interface FlagConfig {
  key: string;
  enabled: boolean;
  description: string;
  rules?: FlagRule[];
  variants?: FlagVariant[];
  createdAt: Date;
  updatedAt: Date;
}

interface FlagRule {
  type: "user" | "percentage" | "attribute" | "datetime";
  operator: "in" | "equals" | "contains" | "gt" | "lt" | "between";
  attribute?: string;
  values: any[];
}

interface FlagVariant {
  key: string;
  weight: number;
  value: any;
}

interface EvaluationContext {
  userId?: string;
// ... (see reference guides for full implementation)

Reference Guides

Detailed implementations in the references/ directory:

| Guide | Contents | |---|---| | Feature Flag Service (TypeScript) | Feature Flag Service (TypeScript) | | React Hook for Feature Flags | React Hook for Feature Flags | | Feature Flag with Analytics | Feature Flag with Analytics | | LaunchDarkly-Style SDK | LaunchDarkly-Style SDK | | Admin UI for Feature Flags | Admin UI for Feature Flags |

Best Practices

✅ DO

• Use descriptive flag names
• Document flag purpose and lifecycle
• Implement gradual rollouts
• Track flag evaluations
• Clean up old flags regularly
• Use feature flags for experiments
• Implement kill switches for critical features
• Test both enabled and disabled states
• Use consistent hashing for stable rollouts
• Provide admin UI for non-technical users

❌ DON'T

• Use flags for permanent configuration
• Accumulate technical debt with old flags
• Skip flag cleanup
• Make flags too granular
• Hard-code flag checks everywhere
• Skip analytics and monitoring