Dockerfile.bench

# Benchmark image for ks-xlsx-parser.
#
# Builds once, then on each run downloads SpreadsheetBench (if not cached),
# parses the corpus, embeds chunks with a small sentence-transformer, and
# emits a recall@k report + failure-bucket triage. The output lands in
# tests/benchmarks/reports/ — mount that path as a volume to persist results.
#
# Usage:
#   docker build -f Dockerfile.bench -t ks-xlsx-parser-bench .
#   docker run --rm \
#       -v "$PWD/tests/benchmarks/reports:/app/tests/benchmarks/reports" \
#       -v "$PWD/data:/app/data" \
#       ks-xlsx-parser-bench
#
#   # Quick sanity run on 20 instances:
#   docker run --rm -e BENCH_SAMPLE=20 ks-xlsx-parser-bench

FROM python:3.12-slim

ENV PYTHONDONTWRITEBYTECODE=1 \
    PYTHONUNBUFFERED=1 \
    PIP_NO_CACHE_DIR=1 \
    PIP_DISABLE_PIP_VERSION_CHECK=1

WORKDIR /app

RUN apt-get update && apt-get install -y --no-install-recommends \
        curl unzip ca-certificates git \
    && rm -rf /var/lib/apt/lists/*

# Install deps first to keep layers cacheable across code edits.
COPY pyproject.toml README.md ./
COPY src ./src
RUN pip install -e ".[dev,bench]"

# Pre-warm the embedding model so the first ``docker run`` doesn't pay the
# ~80 MB download. Same model name eval_retrieval.py defaults to.
RUN python -c "from sentence_transformers import SentenceTransformer; \
SentenceTransformer('BAAI/bge-small-en-v1.5')"

COPY scripts ./scripts
COPY tests ./tests
COPY Makefile ./

ENV BENCH_SAMPLE=0 \
    BENCH_PARSERS=ks \
    BENCH_TIMEOUT=120

ENTRYPOINT ["bash", "scripts/run_bench.sh"]