feat: Add Dockerfile for docling-serve

[edwinjosechittilappilly]

Add a new Dockerfile (Dockerfile.docling_serve) to run docling-serve in a slim Python 3.11 container. Installs system libraries required for OCR/OpenCV, installs uv as a fast package runner, and uses uv pip to install docling-serve (1.5.0) with UI and OCR extras plus onnxruntime, easyocr, docling-core (2.48.1) and opencv-python-headless. Exposes port 8000 and starts the server on 0.0.0.0:8000 with 2 workers.

[Copilot]

Pull request overview

Adds a dedicated Dockerfile to build/run docling-serve as a standalone service container, intended to support OpenRAG’s document parsing/OCR pipeline.

Changes:

• Introduces Dockerfile.docling_serve based on python:3.11-slim.
• Installs OS libraries needed for OCR/OpenCV usage.
• Installs uv and uses it to install docling-serve + related OCR/vision dependencies, then starts the server.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The Dockerfile runs docling-serve on port 8000, but the rest of the repo defaults to port 5001 (e.g., backend proxies to http://<host>:5001 unless DOCLING_SERVE_URL is set, and Helm values set global.docling.port: 5001). If this image is meant to be used by the existing OpenRAG stack without extra configuration, it should expose/bind 5001 (or document/update the required DOCLING_SERVE_URL/Helm values to match 8000).

Suggested change

	# Expose port
	EXPOSE 8000
	# Start server
	CMD ["docling-serve", "run", "--host", "0.0.0.0", "--port", "8000", "--workers", "2"]
	# Expose port (aligned with OpenRAG default docling port)
	EXPOSE 5001
	# Start server
	CMD ["docling-serve", "run", "--host", "0.0.0.0", "--port", "5001", "--workers", "2"]

[Copilot]

apt-get install here omits --no-install-recommends (used in production Dockerfiles like Dockerfile.backend) and includes libxrender-dev (a build-time -dev package) which unnecessarily increases image size for a runtime container. Consider adding --no-install-recommends and switching to the runtime library package where possible.

Suggested change

	RUN apt-get update && apt-get install -y \
	libglib2.0-0 \
	libgl1 \
	libsm6 \
	libxext6 \
	libxrender-dev \
	RUN apt-get update && apt-get install -y --no-install-recommends \
	libglib2.0-0 \
	libgl1 \
	libsm6 \
	libxext6 \
	libxrender1 \

[Copilot]

Only docling-serve and docling-core are pinned; the other Python deps (onnxruntime, easyocr, docling[...], opencv-python-headless, and even uv) are left unpinned, which makes container rebuilds non-reproducible and can introduce sudden breakages when upstream releases. Consider pinning these (or using a lockfile-based install approach) for predictable image builds.

Suggested change

	RUN pip install --no-cache-dir uv
	# Install docling-serve with all extras
	RUN uv pip install \
	"docling-serve[ui]==1.5.0" \
	onnxruntime \
	easyocr \
	"docling[easyocr,rapidocr,vlm]" \
	"docling-core==2.48.1" \
	opencv-python-headless
	RUN pip install --no-cache-dir "uv==0.4.18"
	# Install docling-serve with all extras
	RUN uv pip install \
	"docling-serve[ui]==1.5.0" \
	"onnxruntime==1.18.0" \
	"easyocr==1.7.1" \
	"docling[easyocr,rapidocr,vlm]==2.48.1" \
	"docling-core==2.48.1" \
	"opencv-python-headless==4.10.0.84"