mirror of
https://github.com/tiennm99/DocsGPT.git
synced 2026-10-13 22:15:10 +00:00
build(docker): slim backend image, static frontend image, -docling variant
Backend (arc53/docsgpt): 4.5 GB compressed -> 0.9 GB with both embedding models and tiktoken baked in. - torch/transformers gone from the default install (docling extra only). - Ubuntu 24.04 ships python3.12: no deadsnakes PPA, no software-properties- common; every pin is a wheel, so no gcc/g++/rust in the builder. - COPY --chown and a prefetch that runs as the process user replace the trailing chown -R, which duplicated the 600 MB model layer. - .dockerignore keeps __pycache__, .coverage, local indexes and .env out. - EXTRAS build arg (INSTALL_DOCLING kept as an alias); the docling variant also bakes docling's layout/table/RapidOCR models (DOCLING_ARTIFACTS_PATH) and tesseract, and drops only the discovery documents of Google APIs the app never builds. - FLASK_DEBUG env removed (unused); OCI labels added. Frontend (arc53/docsgpt-fe): 302 MB Vite dev server -> 25 MB static build behind nginx. VITE_* variables are injected at container start into /config.js and read through src/env.ts, so the image no longer needs a rebuild per deployment; docker-compose.yaml keeps hot reload via the dev target. Publishing: every release and develop build now pushes a slim tag and a -docling tag (docling engine + models + tesseract). docker-compose-hub.yaml takes DOCSGPT_IMAGE_TAG / DOCSGPT_IMAGE_VARIANT; docker-compose-standalone.yaml runs the stack from pre-built images without a checkout and is attached to each release. setup.sh selects the -docling variant for OCR instead of requiring a local build. A new workflow builds the image on PRs that touch it and runs verify_offline under --network none; lint checks the exported requirements match uv.lock.
This commit is contained in:
Alex committed
1 parent
4707c45b93
commit
aecb596e99
30 files changed
+706
-247
No files matched your search
@@ -0,0 +1,23 @@
|
||||
# Build context is application/. Keep local state and caches out of the image.
|
||||
__pycache__/
|
||||
*.py[cod]
|
||||
.pytest_cache/
|
||||
.ruff_cache/
|
||||
.coverage
|
||||
htmlcov/
|
||||
*.log
|
||||
|
||||
# Runtime data: bind-mounted or created at run time, never baked in.
|
||||
indexes/
|
||||
inputs/
|
||||
vectors/
|
||||
*.faiss
|
||||
*.pkl
|
||||
|
||||
# Secrets and local config.
|
||||
.env
|
||||
.env.*
|
||||
|
||||
# Not needed inside the image.
|
||||
Dockerfile
|
||||
.dockerignore
|
||||
+112
-96
@@ -1,74 +1,81 @@
|
||||
# Builder Stage
|
||||
FROM ubuntu:24.04 as builder
|
||||
# DocsGPT backend image.
|
||||
#
|
||||
# Build args:
|
||||
# EXTRAS comma-separated optional extras to bake in, matching the
|
||||
# pyproject extras / requirements-<extra>.txt files:
|
||||
# docling (layout-model parser + OCR backend), milvus.
|
||||
# INSTALL_DOCLING legacy alias for EXTRAS=docling (setup.sh writes it).
|
||||
# INSTALL_TESSERACT bake the tesseract binary for OCR_ENGINE=tesseract.
|
||||
# EMBEDDINGS_PREFETCH registry names of the embedding models to bake; empty
|
||||
# bakes both defaults (mpnet for upgrades, granite for
|
||||
# new installs).
|
||||
#
|
||||
# Everything the default configuration needs is inside the image: embedding
|
||||
# models, their tokenizers, tiktoken's encoding and, with the docling extra,
|
||||
# docling's layout/table/OCR models. `python -m application.scripts.verify_offline`
|
||||
# under `docker run --network none` proves it.
|
||||
|
||||
FROM ubuntu:24.04 AS builder
|
||||
|
||||
ENV DEBIAN_FRONTEND=noninteractive
|
||||
|
||||
# Ubuntu 24.04 ships Python 3.12 in its main archive: no PPA needed. Every pin
|
||||
# resolves to a wheel, so no compiler toolchain either.
|
||||
RUN apt-get update && \
|
||||
apt-get install -y --no-install-recommends python3.12 python3.12-venv ca-certificates && \
|
||||
rm -rf /var/lib/apt/lists/*
|
||||
|
||||
COPY requirements.txt requirements-docling.txt requirements-milvus.txt ./
|
||||
|
||||
RUN python3.12 -m venv /venv
|
||||
ENV PATH="/venv/bin:$PATH"
|
||||
|
||||
RUN pip install --no-cache-dir --upgrade pip && \
|
||||
pip install --no-cache-dir --only-binary=:all: -r requirements.txt
|
||||
|
||||
# Optional extras. Each requirements-<extra>.txt is exported from the same
|
||||
# lock as requirements.txt, so installing it on top only adds the extra's
|
||||
# packages. The docling file takes torch from the CPU-only PyTorch index.
|
||||
# Not wheels-only: docling's antlr4 runtime ships as a pure-Python sdist.
|
||||
ARG EXTRAS=""
|
||||
ARG INSTALL_DOCLING=false
|
||||
RUN set -e; \
|
||||
extras="$EXTRAS"; \
|
||||
if [ "$INSTALL_DOCLING" = "true" ]; then extras="$extras,docling"; fi; \
|
||||
for extra in $(echo "$extras" | tr ',' ' '); do \
|
||||
echo "Installing extra: $extra"; \
|
||||
pip install --no-cache-dir -r "requirements-$extra.txt"; \
|
||||
done
|
||||
|
||||
# google-api-python-client bundles discovery documents for ~600 Google APIs
|
||||
# (99 MB). The application builds one client, Drive v3; keep only its document.
|
||||
# Building another API's client needs its file back, or static_discovery=False.
|
||||
RUN find /venv/lib/python3.12/site-packages/googleapiclient/discovery_cache/documents \
|
||||
-type f ! -name 'drive.v3.json' -delete
|
||||
|
||||
|
||||
FROM ubuntu:24.04 AS final
|
||||
|
||||
ENV DEBIAN_FRONTEND=noninteractive
|
||||
|
||||
RUN apt-get update && \
|
||||
apt-get install -y software-properties-common && \
|
||||
add-apt-repository ppa:deadsnakes/ppa && \
|
||||
apt-get update && \
|
||||
apt-get install -y --no-install-recommends gcc g++ wget unzip libc6-dev python3.12 python3.12-venv python3.12-dev && \
|
||||
rm -rf /var/lib/apt/lists/*
|
||||
|
||||
# Verify Python installation and setup symlink
|
||||
RUN if [ -f /usr/bin/python3.12 ]; then \
|
||||
ln -s /usr/bin/python3.12 /usr/bin/python; \
|
||||
else \
|
||||
echo "Python 3.12 not found"; exit 1; \
|
||||
fi
|
||||
|
||||
# Install Rust
|
||||
RUN wget -q -O - https://sh.rustup.rs | sh -s -- -y
|
||||
|
||||
# Clean up to reduce container size
|
||||
RUN apt-get remove --purge -y wget unzip && apt-get autoremove -y && rm -rf /var/lib/apt/lists/*
|
||||
|
||||
# Copy requirements manifests
|
||||
COPY requirements.txt requirements-docling.txt ./
|
||||
|
||||
# Setup Python virtual environment
|
||||
RUN python3.12 -m venv /venv
|
||||
|
||||
# Activate virtual environment and install Python packages
|
||||
ENV PATH="/venv/bin:$PATH"
|
||||
|
||||
# Install Python packages
|
||||
RUN pip install --no-cache-dir --upgrade pip && \
|
||||
pip install --no-cache-dir tiktoken && \
|
||||
pip install --no-cache-dir -r requirements.txt
|
||||
|
||||
# Optional docling parser engine (DOC_PARSER_ENGINE=docling, the docling OCR
|
||||
# backend, read_document's structured output) — OFF by default: it pulls the
|
||||
# layout/OCR model stack and adds gigabytes to the image. anydoc (in
|
||||
# requirements.txt) is the default parser and needs none of it, and OCR runs
|
||||
# natively on tesseract (below, also opt-in) without it.
|
||||
ARG INSTALL_DOCLING=false
|
||||
RUN if [ "$INSTALL_DOCLING" = "true" ]; then \
|
||||
pip install --no-cache-dir -r requirements-docling.txt; \
|
||||
fi
|
||||
|
||||
# Final Stage
|
||||
FROM ubuntu:24.04 as final
|
||||
|
||||
RUN apt-get update && \
|
||||
apt-get install -y software-properties-common && \
|
||||
add-apt-repository ppa:deadsnakes/ppa && \
|
||||
apt-get update && apt-get install -y --no-install-recommends \
|
||||
python3.12 \
|
||||
libgl1 \
|
||||
libglib2.0-0 \
|
||||
poppler-utils \
|
||||
&& \
|
||||
apt-get install -y --no-install-recommends python3.12 poppler-utils ca-certificates && \
|
||||
ln -s /usr/bin/python3.12 /usr/bin/python && \
|
||||
rm -rf /var/lib/apt/lists/*
|
||||
|
||||
# opencv (rapidocr, part of the docling extra) needs libGL at import time.
|
||||
ARG EXTRAS=""
|
||||
ARG INSTALL_DOCLING=false
|
||||
RUN if [ "$INSTALL_DOCLING" = "true" ] || echo ",$EXTRAS," | grep -q ",docling,"; then \
|
||||
apt-get update && \
|
||||
apt-get install -y --no-install-recommends libgl1 libglib2.0-0 && \
|
||||
rm -rf /var/lib/apt/lists/*; \
|
||||
fi
|
||||
|
||||
# Optional tesseract OCR engine (OCR_ENABLED=true with OCR_ENGINE=tesseract,
|
||||
# the default engine) — OFF by default like every other OCR dependency; OCR
|
||||
# itself is off unless configured. Opt in with --build-arg
|
||||
# INSTALL_TESSERACT=true (setup.sh writes it to .env when OCR is enabled);
|
||||
# ~35 MB of system packages. Extra language packs are a deployment concern
|
||||
# (apt: tesseract-ocr-<lang>, then list them in OCR_LANGS). A DeepSeek-OCR
|
||||
# endpoint (OCR_ENGINE=deepseek) needs none of this.
|
||||
# the default engine); ~35 MB of system packages. Extra language packs are a
|
||||
# deployment concern (apt: tesseract-ocr-<lang>, then list them in OCR_LANGS).
|
||||
# A DeepSeek-OCR endpoint (OCR_ENGINE=deepseek) needs none of this.
|
||||
ARG INSTALL_TESSERACT=false
|
||||
RUN if [ "$INSTALL_TESSERACT" = "true" ]; then \
|
||||
apt-get update && \
|
||||
@@ -76,61 +83,70 @@ RUN if [ "$INSTALL_TESSERACT" = "true" ]; then \
|
||||
rm -rf /var/lib/apt/lists/*; \
|
||||
fi
|
||||
|
||||
# Set working directory
|
||||
LABEL org.opencontainers.image.source="https://github.com/arc53/DocsGPT" \
|
||||
org.opencontainers.image.title="DocsGPT" \
|
||||
org.opencontainers.image.description="DocsGPT backend: API and Celery worker" \
|
||||
org.opencontainers.image.licenses="MIT"
|
||||
|
||||
WORKDIR /app
|
||||
|
||||
# Create a non-root user: `appuser` (Feel free to choose a name)
|
||||
# The process user owns /app so the model prefetch below can run as it: an
|
||||
# unprivileged prefetch writes the model files with the right owner up front,
|
||||
# instead of a trailing chown -R that rewrites every model file into a second
|
||||
# layer.
|
||||
RUN groupadd -r appuser && \
|
||||
useradd -r -g appuser -d /app -s /sbin/nologin -c "Docker image user" appuser
|
||||
useradd -r -g appuser -d /app -s /sbin/nologin -c "Docker image user" appuser && \
|
||||
chown appuser:appuser /app && \
|
||||
install -d -o appuser -g appuser /app/models /app/application
|
||||
|
||||
# Copy the virtual environment and model from the builder stage
|
||||
COPY --from=builder /venv /venv
|
||||
|
||||
# Pre-fetch the embedding models into FastEmbed's cache so a fresh container
|
||||
# does not download on first ingest and an air-gapped install works at all.
|
||||
# Both defaults are baked: an upgraded deployment keeps using mpnet until it
|
||||
# runs the re-embed script, while a new one starts on granite.
|
||||
# The prefetch writes hub-layout snapshots (including tokenizer.json) here, so
|
||||
# HF_HUB_CACHE has to point at the same directory: chunking loads the tokenizer
|
||||
# through ``tokenizers``, which reads the hub cache and would otherwise fetch
|
||||
# over the network on first ingest -- and fall back to cl100k when offline.
|
||||
# Every cache the application reads at run time lives under /app/models and is
|
||||
# filled at build time:
|
||||
# EMBEDDINGS_CACHE_DIR / HF_HUB_CACHE FastEmbed models and their tokenizers
|
||||
# (chunking reads tokenizer.json from
|
||||
# the same hub-layout snapshot)
|
||||
# TIKTOKEN_CACHE_DIR cl100k_base for token accounting
|
||||
# DOCLING_ARTIFACTS_PATH docling's models (docling extra only)
|
||||
ENV EMBEDDINGS_CACHE_DIR=/app/models \
|
||||
HF_HUB_CACHE=/app/models
|
||||
HF_HUB_CACHE=/app/models \
|
||||
TIKTOKEN_CACHE_DIR=/app/models/tiktoken \
|
||||
DOCLING_ARTIFACTS_PATH=/app/models/docling \
|
||||
HF_HUB_DISABLE_TELEMETRY=1 \
|
||||
PATH="/venv/bin:$PATH"
|
||||
|
||||
# Only the modules the prefetch imports are copied first, so an unrelated
|
||||
# source edit does not invalidate the model layer.
|
||||
COPY --chown=appuser:appuser __init__.py /app/application/__init__.py
|
||||
COPY --chown=appuser:appuser scripts/__init__.py scripts/prefetch_models.py /app/application/scripts/
|
||||
COPY --chown=appuser:appuser vectorstore/__init__.py vectorstore/model_registry.py /app/application/vectorstore/
|
||||
|
||||
USER appuser
|
||||
|
||||
# Only the modules the prefetch imports are copied first. It reaches nothing
|
||||
# beyond model_registry, which is stdlib-only, so keeping the full source copy
|
||||
# below this layer stops an unrelated edit from re-downloading ~780 MB of model
|
||||
# artifacts on every build.
|
||||
COPY __init__.py /app/application/__init__.py
|
||||
COPY scripts/__init__.py scripts/prefetch_models.py /app/application/scripts/
|
||||
COPY vectorstore/__init__.py vectorstore/model_registry.py /app/application/vectorstore/
|
||||
ARG EMBEDDINGS_PREFETCH=""
|
||||
RUN PYTHONPATH=/app /venv/bin/python -m application.scripts.prefetch_models ${EMBEDDINGS_PREFETCH}
|
||||
RUN PYTHONPATH=/app python -m application.scripts.prefetch_models ${EMBEDDINGS_PREFETCH} && \
|
||||
rm -rf /app/models/.locks /app/.cache
|
||||
|
||||
# Copy your application code
|
||||
COPY . /app/application
|
||||
# docling downloads its layout, table-structure and OCR models on first parse;
|
||||
# bake them so the docling variant is as self-contained as the default image.
|
||||
RUN if python -c "import docling" 2>/dev/null; then \
|
||||
docling-tools models download --output-dir /app/models/docling layout tableformer rapidocr && \
|
||||
rm -rf /app/.cache; \
|
||||
fi
|
||||
|
||||
# Change the ownership of the /app directory to the appuser
|
||||
COPY --chown=appuser:appuser . /app/application
|
||||
|
||||
RUN mkdir -p /app/application/inputs/local
|
||||
RUN chown -R appuser:appuser /app
|
||||
|
||||
# Set environment variables
|
||||
ENV FLASK_APP=app.py \
|
||||
FLASK_DEBUG=true \
|
||||
PATH="/venv/bin:$PATH"
|
||||
ENV FLASK_APP=app.py
|
||||
|
||||
ENV MALLOC_ARENA_MAX=2 \
|
||||
OMP_NUM_THREADS=4 \
|
||||
MKL_NUM_THREADS=4 \
|
||||
OPENBLAS_NUM_THREADS=4
|
||||
|
||||
# Expose the port the app runs on
|
||||
EXPOSE 7091
|
||||
|
||||
# Switch to non-root user
|
||||
USER appuser
|
||||
|
||||
# BoundedDrainUvicornWorker makes max_requests recycles safe with held-open SSE
|
||||
# connections (see application/gunicorn_worker.py); with recycles now safe,
|
||||
# --max-requests is raised (kept for memory hygiene) to cut churn.
|
||||
|
||||
Reference in new issue
Block a user