# ============================================================
# Arabic Nativity Classifier — Multi-stage Docker Build
# ============================================================
# Stage 1: Install Python deps (cached unless requirements change)
# Stage 2: Copy app code + models → slim runtime image
# ============================================================

FROM python:3.11-slim AS base

# Avoid interactive prompts, set locale
ENV DEBIAN_FRONTEND=noninteractive \
    PYTHONUNBUFFERED=1 \
    PYTHONDONTWRITEBYTECODE=1 \
    PIP_NO_CACHE_DIR=1 \
    PIP_DISABLE_PIP_VERSION_CHECK=1

# System deps for librosa (libsndfile), soundfile, and general audio
RUN apt-get update && apt-get install -y --no-install-recommends \
    libsndfile1 \
    ffmpeg \
    libgomp1 \
    curl \
    && rm -rf /var/lib/apt/lists/*

WORKDIR /app

# ---- Stage: deps (cached layer) ----
FROM base AS deps

COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt && \
    pip install --no-cache-dir gunicorn

# Pre-download the HuggingFace Wav2Vec2 model so it's baked into the image
# (avoids ~1.2GB download on every container start)
# MUST match extract_embeddings.py model used for training embeddings
RUN python -c "\
from transformers import Wav2Vec2Model, AutoFeatureExtractor; \
AutoFeatureExtractor.from_pretrained('facebook/wav2vec2-lv-60-espeak-cv-ft'); \
Wav2Vec2Model.from_pretrained('facebook/wav2vec2-lv-60-espeak-cv-ft'); \
print('Wav2Vec2 lv-60-espeak-cv-ft cached.')"

# ---- Stage: runtime ----
FROM deps AS runtime

WORKDIR /app

# Copy application code
COPY scripts/ ./scripts/
COPY webapp/ ./webapp/
COPY manifests/embedding_dim.txt ./manifests/embedding_dim.txt
COPY manifests/test_data.csv ./manifests/test_data.csv

# Copy trained models (these must exist locally — they're gitignored)
COPY models/nativity_model.pth ./models/nativity_model.pth
COPY models/best_threshold.txt ./models/best_threshold.txt
COPY models/lgbm_nativity.txt ./models/lgbm_nativity.txt
COPY models/pca_transform.pkl ./models/pca_transform.pkl

# Create uploads dir for audio temp files
RUN mkdir -p /app/webapp/uploads

# Expose Flask port (7860 for Hugging Face Spaces)
EXPOSE 7860

# Health check — hit the /api/health endpoint
HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
    CMD curl -f http://localhost:7860/api/health || exit 1

# Run with gunicorn for production (2 workers, 600s timeout for long audio)
CMD ["gunicorn", \
     "--bind", "0.0.0.0:7860", \
     "--workers", "2", \
     "--threads", "4", \
     "--timeout", "600", \
     "--access-logfile", "-", \
     "--error-logfile", "-", \
     "webapp.app:app"]