Docker Builds¶

Build and test Docker images for CPU and GPU environments.

Quick Start¶

# Build and test CPU image
./scripts/build-docker.sh cpu

# Build and test GPU image
./scripts/build-docker.sh gpu

# Build both
./scripts/build-docker.sh all

Images¶

CPU Image¶

docker build --build-arg BUILD_TYPE=cpu -t docs2synth:cpu .

Size: ~2.5 GB
Base: python:3.11-slim
Use: Development, testing, CPU-only deployment
Build time: ~10-15 min

GPU Image¶

docker build --build-arg BUILD_TYPE=gpu -t docs2synth:gpu .

Size: ~11 GB
Base: nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04
Use: GPU training, GPU inference, production with CUDA
Build time: ~15-25 min

Testing¶

CPU Tests¶

# Run all tests
docker run --rm docs2synth:cpu pytest tests/ -v

# Interactive shell
docker run -it docs2synth:cpu /bin/bash

GPU Tests¶

# Check CUDA
docker run --rm --gpus all docs2synth:gpu python -c "import torch; print(torch.cuda.is_available())"

# Run tests with GPU
docker run --rm --gpus all docs2synth:gpu pytest tests/ -v

# Interactive shell
docker run --gpus all -it docs2synth:gpu /bin/bash

Requirements for GPU: - Linux with NVIDIA GPU - NVIDIA drivers installed - NVIDIA Container Toolkit installed

Production¶

# Build and tag
docker build --build-arg BUILD_TYPE=cpu -t your-registry/docs2synth:v0.1.0-cpu .
docker build --build-arg BUILD_TYPE=gpu -t your-registry/docs2synth:v0.1.0-gpu .

# Push
docker push your-registry/docs2synth:v0.1.0-cpu
docker push your-registry/docs2synth:v0.1.0-gpu

Run in Production¶

# CPU
docker run -d \
  --name docs2synth \
  -v /data:/app/data \
  -v /logs:/app/logs \
  your-registry/docs2synth:v0.1.0-cpu \
  python -m docs2synth.cli process --config /app/config.yml

# GPU
docker run -d \
  --gpus all \
  --name docs2synth-gpu \
  -v /data:/app/data \
  -v /logs:/app/logs \
  your-registry/docs2synth:v0.1.0-gpu \
  python -m docs2synth.cli process --config /app/config.yml

Why No CI for Docker?¶

GitHub Actions runners have limited disk space (~14GB). GPU image needs ~16GB total: - ~11GB image - ~5GB build cache

Solution: Manual builds locally or on dedicated machines.

Troubleshooting¶

No Space Left¶

docker image prune -a
docker container prune
docker builder prune

GPU Image Fails¶

# Verify NVIDIA setup
nvidia-smi
docker run --rm --gpus all nvidia/cuda:11.8.0-base-ubuntu22.04 nvidia-smi

Slow Builds¶

# Use BuildKit
DOCKER_BUILDKIT=1 docker build --build-arg BUILD_TYPE=cpu -t docs2synth:cpu .