GridLink Architecture

GridLink is an EV Fleet Charging Management Platform built for reliability and scale.

Platform Overview

Design Principles

API / Worker Separation

The platform separates synchronous API requests from asynchronous background work:

Component	Responsibility	Scaling Strategy
API Service	Handle client requests, validate input, return responses	Horizontal scaling behind load balancer
Worker Processes	Long-running tasks, optimization, batch operations	Scale based on queue depth

This separation ensures API response times remain fast regardless of background workload.

Dedicated WebSocket Infrastructure

Charger connections require special handling:

• Persistent connections - Each charger maintains a long-lived WebSocket
• Auto-scaling - WebSocket instances scale independently based on connection count
• Sticky sessions - Load balancer routes charger to same instance for connection lifetime
• Graceful handoff - During deploys, connections drain to new instances without message loss

Message Queue for Resilience

Asynchronous communication via message queue provides:

• Decoupling - Services operate independently
• Buffering - Spikes in load are absorbed by the queue
• Retry logic - Failed operations are retried automatically
• Guaranteed delivery - Messages persist until acknowledged

Key workflows using the queue:

• Charging schedule optimization requests
• Bulk charger commands
• Webhook delivery
• Report generation

Caching Strategy

Multi-layer caching reduces database load:

Cache Type	Use Case	TTL
Session cache	Active charging sessions	Real-time
Configuration cache	Charger settings, rate structures	Minutes
Query cache	Dashboard aggregations	Seconds

Reliability Patterns

Health Checks & Circuit Breakers

• All services expose health endpoints
• Load balancers remove unhealthy instances automatically
• External service calls use circuit breakers to prevent cascade failures

Data Consistency

• Database transactions for critical operations
• Idempotency keys for API mutations
• Event sourcing for charging session state

Offline Tolerance

When connectivity is lost:

• Chargers queue messages locally, replay on reconnect
• Platform handles out-of-order message delivery
• Transactions are reconstructed from queued events

Deployment Model

Layer	Infrastructure
Edge	CDN for static assets, DDoS protection
Load Balancing	Application LB for API, Network LB for WebSockets
Compute	Container orchestration with auto-scaling
Database	Managed relational database with read replicas
Cache	Managed in-memory store with clustering
Queue	Managed message broker with persistence

All components run in private subnets. Only load balancers are internet-facing.

High Availability

• Multi-AZ deployment for all stateful components
• Automated failover for database
• Zero-downtime deployments via rolling updates
• Connection draining for WebSocket instances