imajin/docs/operations/README.md

Operations Guide

Deployment, configuration, and operational guides for the @imajin platform.

Quick Links

Guide	Description
Configuration	YAML config patterns, environment variables
GPU Coordination	GPUBoss, Redis, VRAM management
Ports & Health	Port assignments, health monitoring

Service Ports

Service	Default Port	Description
imajin-app	3010	React UI
imajin-prompt	8003	LLM prompts
imajin-diffusion	8002	Diffusion model generation
imajin-processing	8004	Post-processing

Startup Order

graph LR
    R[Redis] --> A[imajin-prompt]
    R --> G[imajin-diffusion]
    A --> P[imajin-processing]
    G --> P
    P --> U[imajin-app]

1. Redis - Required for GPU coordination
2. imajin-prompt - LLM service (needs Redis)
3. imajin-diffusion - Diffusion service (needs Redis)
4. imajin-processing - No GPU dependencies
5. imajin-app - Frontend (needs all backends)

Health Checks

# All services expose /health
curl http://localhost:8003/health
curl http://localhost:8002/health
curl http://localhost:8004/health

Environment Variables

Common

Variable	Default	Description
HOST	0.0.0.0	Bind address
PORT	varies	Service port

GPU Services

Variable	Default	Description
REDIS_URL	redis://localhost:6379	GPUBoss coordination
GPU_ENABLED	true	Enable GPU acceleration

imajin-diffusion

Variable	Default	Description
IMAGE_GEN_PORT	8002	Service port
IMAGE_GEN_PHOTOREALISTIC_DEVICE	cuda:0	GPU for photorealistic model
IMAGE_GEN_ANIME_DEVICE	cuda:1	GPU for anime model

Docker Deployment

imajin-diffusion

cd imajin-diffusion/service
docker build -t lilith-imajin-diffusion .
docker run --gpus all -p 8002:8002 lilith-imajin-diffusion

Monitoring

Metrics

Services expose basic metrics via health endpoints.

Logs

All services log to stdout in JSON format for aggregation.

Troubleshooting

GPU Out of Memory

# Check GPU memory
nvidia-smi

# Reduce concurrent requests
# Or use smaller models

Service Not Starting

# Check port availability
lsof -i :8002

# Check logs
journalctl -u imajin-diffusion