Dockerfile.gpu

# =============================================================================
# RuVector Cloud Run GPU Dockerfile
# Optimized for NVIDIA L4 GPUs on Google Cloud Run
# =============================================================================

# -----------------------------------------------------------------------------
# Stage 1: Build Environment
# -----------------------------------------------------------------------------
FROM nvidia/cuda:12.3.1-devel-ubuntu22.04 AS builder

# Prevent interactive prompts
ENV DEBIAN_FRONTEND=noninteractive

# Install build dependencies
RUN apt-get update && apt-get install -y \
    curl \
    build-essential \
    pkg-config \
    libssl-dev \
    cmake \
    git \
    clang \
    llvm \
    && rm -rf /var/lib/apt/lists/*

# Install Rust
RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
ENV PATH="/root/.cargo/bin:${PATH}"

# Set CUDA paths
ENV CUDA_HOME=/usr/local/cuda
ENV LD_LIBRARY_PATH=${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}
ENV PATH=${CUDA_HOME}/bin:${PATH}

WORKDIR /build

# Copy workspace Cargo files for dependency caching
COPY Cargo.toml Cargo.lock ./

# Copy all crate manifests
COPY crates/ruvector-core/Cargo.toml crates/ruvector-core/
COPY crates/ruvector-bench/Cargo.toml crates/ruvector-bench/
COPY crates/ruvector-gnn/Cargo.toml crates/ruvector-gnn/
COPY crates/ruvector-attention/Cargo.toml crates/ruvector-attention/
COPY crates/ruvector-raft/Cargo.toml crates/ruvector-raft/
COPY crates/ruvector-replication/Cargo.toml crates/ruvector-replication/
COPY crates/ruvector-cluster/Cargo.toml crates/ruvector-cluster/
COPY crates/ruvector-server/Cargo.toml crates/ruvector-server/
COPY crates/ruvector-collections/Cargo.toml crates/ruvector-collections/
COPY crates/ruvector-filter/Cargo.toml crates/ruvector-filter/
COPY crates/ruvector-metrics/Cargo.toml crates/ruvector-metrics/
COPY crates/ruvector-snapshot/Cargo.toml crates/ruvector-snapshot/

# Copy example manifest
COPY examples/google-cloud/Cargo.toml examples/google-cloud/

# Create stub files for dependency resolution
RUN mkdir -p crates/ruvector-core/src && echo "pub fn stub() {}" > crates/ruvector-core/src/lib.rs && \
    mkdir -p crates/ruvector-bench/src && echo "pub fn stub() {}" > crates/ruvector-bench/src/lib.rs && \
    mkdir -p crates/ruvector-gnn/src && echo "pub fn stub() {}" > crates/ruvector-gnn/src/lib.rs && \
    mkdir -p crates/ruvector-attention/src && echo "pub fn stub() {}" > crates/ruvector-attention/src/lib.rs && \
    mkdir -p crates/ruvector-raft/src && echo "pub fn stub() {}" > crates/ruvector-raft/src/lib.rs && \
    mkdir -p crates/ruvector-replication/src && echo "pub fn stub() {}" > crates/ruvector-replication/src/lib.rs && \
    mkdir -p crates/ruvector-cluster/src && echo "pub fn stub() {}" > crates/ruvector-cluster/src/lib.rs && \
    mkdir -p crates/ruvector-server/src && echo "pub fn stub() {}" > crates/ruvector-server/src/lib.rs && \
    mkdir -p crates/ruvector-collections/src && echo "pub fn stub() {}" > crates/ruvector-collections/src/lib.rs && \
    mkdir -p crates/ruvector-filter/src && echo "pub fn stub() {}" > crates/ruvector-filter/src/lib.rs && \
    mkdir -p crates/ruvector-metrics/src && echo "pub fn stub() {}" > crates/ruvector-metrics/src/lib.rs && \
    mkdir -p crates/ruvector-snapshot/src && echo "pub fn stub() {}" > crates/ruvector-snapshot/src/lib.rs && \
    mkdir -p examples/google-cloud/src && echo "fn main() {}" > examples/google-cloud/src/main.rs

# Build dependencies (cached layer)
RUN cargo build --release -p ruvector-cloudrun-gpu 2>/dev/null || true

# Copy actual source code
COPY crates/ crates/
COPY examples/google-cloud/src/ examples/google-cloud/src/

# Build the benchmark binary
RUN cargo build --release -p ruvector-cloudrun-gpu

# -----------------------------------------------------------------------------
# Stage 2: Runtime Environment
# -----------------------------------------------------------------------------
FROM nvidia/cuda:12.3.1-runtime-ubuntu22.04

# Install runtime dependencies
RUN apt-get update && apt-get install -y \
    libssl3 \
    ca-certificates \
    curl \
    && rm -rf /var/lib/apt/lists/*

# Create non-root user
RUN useradd -m -u 1000 -s /bin/bash ruvector

# Create app directory
WORKDIR /app

# Copy binary from builder
COPY --from=builder /build/target/release/gpu-benchmark ./

# Set ownership
RUN chown -R ruvector:ruvector /app

# Switch to non-root user
USER ruvector

# Environment variables
ENV NVIDIA_VISIBLE_DEVICES=all
ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
ENV RUVECTOR_GPU_ENABLED=true
ENV RUST_LOG=info
ENV PORT=8080

# Health check
HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
    CMD curl -f http://localhost:${PORT}/health || exit 1

# Expose port
EXPOSE 8080

# Default command: start server
CMD ["./gpu-benchmark", "serve", "--port", "8080"]