Dockerfile.infer

# ==============================================================================
# Nikola Inference — Lightweight Docker Image
# v0.2.5: Inference-only runner (no autonomy/training/SIE stack)
#
# GPU build (default):
#   docker build -f Dockerfile.infer -t nikola-infer:latest .
#   docker run --gpus all -p 8080:8080 nikola-infer:latest --serve
#
# CPU-only build:
#   docker build -f Dockerfile.infer \
#     --build-arg BUILDER_IMAGE=ubuntu:24.04 \
#     --build-arg RUNTIME_IMAGE=ubuntu:24.04 \
#     --build-arg CUDA_ARCH="" \
#     -t nikola-infer:cpu .
#
# ~60% smaller than the full image (no ZMQ, no training, no security stack).
# ==============================================================================

ARG CUDA_VERSION=12.6.3
ARG UBUNTU_VERSION=24.04
ARG ORT_VERSION=1.21.1
ARG CMAKE_BUILD_TYPE=Release
ARG CUDA_ARCH=86
ARG BUILDER_IMAGE=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VERSION}
ARG RUNTIME_IMAGE=nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu${UBUNTU_VERSION}

# ==============================================================================
# Stage 1: Builder
# ==============================================================================

FROM ${BUILDER_IMAGE} AS builder

ARG CMAKE_BUILD_TYPE
ARG ORT_VERSION
ARG CUDA_ARCH
ARG TARGETARCH

ENV DEBIAN_FRONTEND=noninteractive

RUN apt-get update && apt-get install -y --no-install-recommends \
    build-essential \
    cmake \
    ninja-build \
    git \
    pkg-config \
    libeigen3-dev \
    libprotobuf-dev \
    protobuf-compiler \
    libzmq3-dev \
    libczmq-dev \
    libcurl4-openssl-dev \
    liblmdb-dev \
    libsodium-dev \
    libssl-dev \
    && rm -rf /var/lib/apt/lists/*

# cppzmq (required by nikola_core)
RUN git clone --depth 1 --branch v4.10.0 \
    https://github.com/zeromq/cppzmq.git /tmp/cppzmq && \
    cd /tmp/cppzmq && mkdir build && cd build && \
    cmake .. -DCPPZMQ_BUILD_TESTS=OFF -G Ninja && \
    ninja install && rm -rf /tmp/cppzmq

# Catch2 v3
RUN git clone --depth 1 --branch v3.4.0 \
    https://github.com/catchorg/Catch2.git /tmp/catch2 && \
    cd /tmp/catch2 && mkdir build && cd build && \
    cmake .. -DBUILD_TESTING=OFF -G Ninja && \
    ninja install && rm -rf /tmp/catch2

# ONNX Runtime
RUN mkdir -p /opt/onnxruntime && \
    apt-get update && apt-get install -y --no-install-recommends wget && \
    wget -q "https://github.com/microsoft/onnxruntime/releases/download/v${ORT_VERSION}/onnxruntime-linux-x64-gpu-${ORT_VERSION}.tgz" \
        -O /tmp/ort.tgz && \
    tar xzf /tmp/ort.tgz -C /opt/onnxruntime --strip-components=1 && \
    rm /tmp/ort.tgz && \
    rm -rf /var/lib/apt/lists/*

WORKDIR /build/nikola
COPY . .

# Build — only nikola-infer target (faster build, smaller binary)
RUN mkdir -p build && cd build && \
    CMAKE_ARGS="-G Ninja \
        -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE} \
        -DORT_ROOT=/opt/onnxruntime \
        -DNIKOLA_ORT_MODEL_PATH=/build/nikola/models/bert-tiny-onnx/model.onnx \
        -DNIKOLA_ORT_TOKENIZER_PATH=/build/nikola/models/bert-tiny-onnx \
        -DNIKOLA_BUILD_TESTS=OFF" && \
    if [ -n "${CUDA_ARCH}" ]; then \
        CMAKE_ARGS="${CMAKE_ARGS} -DCMAKE_CUDA_ARCHITECTURES=${CUDA_ARCH}"; \
    fi && \
    cmake .. ${CMAKE_ARGS} && \
    ninja nikola_infer -j$(nproc)

# ==============================================================================
# Stage 2: Runtime — inference only
# ==============================================================================

FROM ${RUNTIME_IMAGE} AS runtime

ENV DEBIAN_FRONTEND=noninteractive

# Minimal runtime deps (no ZMQ — inference server uses raw HTTP)
RUN apt-get update && apt-get install -y --no-install-recommends \
    libprotobuf-lite32t64 \
    libzmq5 \
    libcurl4t64 \
    liblmdb0 \
    libsodium23 \
    libssl3t64 \
    libgomp1 \
    && rm -rf /var/lib/apt/lists/*

# Non-root user
RUN groupadd -r nikola && useradd -r -g nikola -m -s /bin/bash nikola

# Directory structure
RUN mkdir -p /opt/nikola/bin \
             /opt/nikola/lib \
             /opt/nikola/models \
             /data/nikola/checkpoints && \
    chown -R nikola:nikola /data/nikola

# Copy only the inference binary
COPY --from=builder /build/nikola/build/nikola-infer /opt/nikola/bin/

# Copy ONNX Runtime shared libs
COPY --from=builder /opt/onnxruntime/lib/libonnxruntime*.so* /opt/nikola/lib/

# Copy BERT-tiny model
COPY --from=builder /build/nikola/models/ /opt/nikola/models/

# Environment
ENV LD_LIBRARY_PATH="/opt/nikola/lib:${LD_LIBRARY_PATH}"
ENV PATH="/opt/nikola/bin:${PATH}"
ENV NIKOLA_ORT_MODEL_PATH="/opt/nikola/models/bert-tiny-onnx/model.onnx"
ENV NIKOLA_ORT_TOKENIZER_PATH="/opt/nikola/models/bert-tiny-onnx"

# HTTP API port
EXPOSE 8080

# Health check via HTTP endpoint
HEALTHCHECK --interval=10s --timeout=3s --retries=3 --start-period=15s \
    CMD curl -sf http://localhost:8080/v1/health || exit 1

USER nikola
WORKDIR /data/nikola

ENTRYPOINT ["nikola-infer"]
CMD ["--serve", "--port", "8080"]