Optimize Docker build (Smaller base image)

[dewey]

Right now the images are very large. We should optimize to make them smaller and maybe not require a base image if possible.

[fuchenshi]

Have you checked Docker's multi-stage builds? Maybe it can help:

https://docs.docker.com/build/building/multi-stage/

[dewey]

Yep, I have tried that before, but while the final image is smaller the build was still very long as Dockerfile.base has to install a bunch of dependencies. That's why now I just split it into Dockerfile.base, which I only re-build very rarely, and the regular Dockerfile which just uses the latest version of the image to run the app with.

I think these Python ML libraries are just very large, maybe there's not much I can optimize in general.

[fuchenshi]

Maybe we can keep Dockerfile.base as is and refactor Dockerfile into a multi-stage build:

Stage 1: Start from Dockerfile.base and install the project. This doesn't require Dockerfile.base to be rebuilt, so it saves time.

Stage 2: Start from a clean Python base, install only runtime dependencies, and copy the necessary artifacts from Stage 1. This leaves the build tools behind, so the final image is smaller.

Below is a draft suggested by AI. I've verified the logic but haven't tested it yet. It should be a good starting point. 😀

# syntax=docker/dockerfile:1

# -----------------------------------------------------------------------------
# Stage 1: Builder
# Use your pre-built base image to leverage cached ML dependencies
# -----------------------------------------------------------------------------
FROM ghcr.io/dewey/birdhomie-base:latest AS builder

WORKDIR /app

# Copy project files
COPY pyproject.toml uv.lock ./
COPY src/ ./src/
COPY migrations/ ./migrations/
COPY README.md ./

# Sync project dependencies
# This updates the pre-existing .venv from the base image with your project code
RUN --mount=type=cache,target=/root/.cache/uv \
    uv sync --no-dev

# -----------------------------------------------------------------------------
# Stage 2: Runner
# Start with a fresh, empty python-slim image to minimize size
# -----------------------------------------------------------------------------
FROM python:3.13-slim-bookworm

WORKDIR /app

# Install only the necessary runtime system libraries
# We re-install ffmpeg here to keep this image layer clean
RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
    apt-get update && apt-get install -y --no-install-recommends \
        libglib2.0-0 \
        ffmpeg \
        && rm -rf /var/lib/apt/lists/*

# Copy the fully built virtual environment from the builder stage
COPY --from=builder /app/.venv /app/.venv

# Copy application source code
COPY src/ ./src/
COPY migrations/ ./migrations/
COPY README.md ./

# create data directory
RUN mkdir -p /app/data

# Add the virtual environment to the PATH
# This allows us to run 'python' and 'gunicorn' directly without 'uv'
ENV PATH="/app/.venv/bin:$PATH"
ENV PORT=5000
ENV GUNICORN_WORKERS=2

EXPOSE 5000

# Healthcheck using standard python (since 'uv' is not installed in this stage)
HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
    CMD python -c "import urllib.request, os; urllib.request.urlopen(f'http://localhost:{os.environ.get(\"PORT\", \"5000\")}/')" || exit 1

# Run gunicorn directly
CMD ["gunicorn", "--config", "src/birdhomie/gunicorn.conf.py", "birdhomie.app:app"]