Skip to content

0001-modular-backend-architecture.md

Latest commit

 

History

History
44 lines (31 loc) · 2.17 KB

0001-modular-backend-architecture.md

File metadata and controls

44 lines (31 loc) · 2.17 KB

1. Modular Backend Architecture

Date: 2025-11-29

Status

Accepted

Context

The IndiaAI IDP Platform needs to run in diverse environments:

  1. Local Development: Developers need to run the full stack on their laptops without complex infrastructure (Redis, S3, PostgreSQL).
  2. Production: The system needs to scale using robust cloud-native services (Cloud Storage, Managed Queues, Managed DB).

We need an architecture that supports both without requiring code changes or conditional logic scattered throughout the codebase.

Decision

We will implement a Modular Service Architecture using the Strategy Pattern. Key infrastructure components will be defined as abstract base classes (interfaces) with multiple implementations.

Components

  1. Storage: StorageService interface.
    • LocalStorageService: Saves files to the local filesystem.
    • R2StorageService: Saves files to Cloudflare R2 (S3-compatible).
  2. Queue: QueueService interface.
    • MemoryQueueService: Uses Python's asyncio.Queue for in-process queuing.
    • RedisQueueService: Uses Redis for distributed queuing.
  3. OCR: OCRService interface.
    • PaddleOCRService: Uses PaddleOCR.
    • EasyOCRService: Uses EasyOCR (future/alternative).

Configuration

The active implementation will be selected at runtime based on environment variables (e.g., STORAGE_TYPE=local vs STORAGE_TYPE=s3) using a factory pattern in app/core/config.py or service getters.

Consequences

Positive

  • Zero-Setup Dev Environment: Developers can git clone and run without installing Redis or MinIO.
  • Testability: Easy to mock services for unit testing.
  • Vendor Lock-in Avoidance: Switching from R2 to AWS S3 or Azure Blob Storage only requires adding a new StorageService implementation.

Negative

  • Complexity: Adds a layer of abstraction. Simple direct calls (e.g., open()) are replaced by service calls (storage.upload()).
  • Feature Intersection: The "lowest common denominator" of features must be supported by all implementations, or specific features ( like S3 presigned URLs) might not be available in the Local implementation.