Async Inference Queue Design

Design an asynchronous inference queue for handling high-volume LLM requests.

## Volume Requirements
{{volume_requirements}}

## Latency SLAs
- P50: {{p50_latency}}s
- P99: {{p99_latency}}s

## Queue Technology
{{queue_technology}}

Design the system:

**Queue Architecture**
- Priority lanes
- Dead letter handling
- Retry policies

**Worker Design**
- Auto-scaling triggers
- Health checking
- Graceful shutdown

**Result Delivery**
- Webhook callbacks
- Polling endpoint
- WebSocket streaming

**Observability**
- Queue depth monitoring
- Latency tracking
- Error rate alerting

Provide:
- Architecture diagram
- Configuration schemas
- Implementation code
- Capacity planning formulas