[Docker][KVConnector] Install mooncake from official wheels instead of a custom build (#51067)

Signed-off-by: Zhewen Li <zhewenli@inferact.ai>
Co-authored-by: Zhewen Li <zhewenli@inferact.ai>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Zhewen Li
2026-08-05 11:58:44 -07:00
committed by GitHub
co-authored by Zhewen Li Claude Opus 5
parent 613411a90c
commit 14e57ad47d
3 changed files with 26 additions and 47 deletions
+17 -23
View File
@@ -975,32 +975,26 @@ RUN --mount=type=cache,target=/opt/uv/cache \
# clean up -dev packages, keep runtime libraries
rm -rf /var/lib/apt/lists/* \
); \
# Force-reinstall the matching CUDA wheel so the correct nixl_ep_cpp.so is installed.
uv pip install --system --force-reinstall --no-deps nixl-cu${CUDA_MAJOR}; \
fi
# Optional override: install mooncake-transfer-engine from a URL instead of the
# PyPI release pulled in above. Use this for wheels built with non-default CMake
# flags (e.g. `STORE_USE_ETCD=ON` for master HA). The URL's manylinux glibc
# floor must be <= the FINAL_BASE_IMAGE's glibc.
ARG MOONCAKE_WHEEL_AARCH64
ARG MOONCAKE_WHEEL_X86_64
RUN if [ "$INSTALL_KV_CONNECTORS" = "true" ]; then \
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
WHEEL="${MOONCAKE_WHEEL_AARCH64}"; \
else \
WHEEL="${MOONCAKE_WHEEL_X86_64}"; \
fi && \
if [ -n "${WHEEL}" ]; then \
uv pip install --system "${WHEEL}" && \
CUDA_MAJOR="${CUDA_VERSION%%.*}" && \
if [ ! -f /usr/local/cuda/lib64/libcudart.so ] && \
[ -f "/usr/local/cuda/lib64/libcudart.so.${CUDA_MAJOR}" ]; then \
ln -s "libcudart.so.${CUDA_MAJOR}" /usr/local/cuda/lib64/libcudart.so; \
fi; \
# Force-reinstall the matching CUDA wheel so the correct nixl_ep_cpp.so is
# installed. Chained with && through the mooncake swap below: this RUN has no
# `set -e`, so its status is that of the last command, and a trailing no-op
# would otherwise mask a failure here.
uv pip install --system --force-reinstall --no-deps nixl-cu${CUDA_MAJOR} && \
# The default mooncake wheel links libcudart.so.12, so on CUDA 13 swap in the
# cuda13 variant. Uninstall first; both ship the `mooncake` package.
# Mirrors .buildkite/scripts/install-kv-connectors.sh.
MOONCAKE_VERSION=$(python3 -c "import importlib.metadata as m; print(m.version('mooncake-transfer-engine'))" 2>/dev/null || true) && \
if [ "${CUDA_MAJOR}" = "13" ] && [ -n "${MOONCAKE_VERSION}" ]; then \
uv pip uninstall --system mooncake-transfer-engine 2>/dev/null || true; \
uv pip install --system "mooncake-transfer-engine-cuda13==${MOONCAKE_VERSION}"; \
fi; \
fi
# Mooncake registers GPU memory for RDMA via nvidia-peermem by default. Hosts
# without that module loaded (e.g. GB200) must pass WITH_NVIDIA_PEERMEM=0 at run
# time; see docs/features/mooncake_connector_usage.md. Deliberately not an ENV
# here so the image does not override mooncake's own default.
ENV VLLM_USAGE_SOURCE production-docker-image
ENV VLLM_BUILD_COMMIT=${VLLM_BUILD_COMMIT:-unknown} \
VLLM_BUILD_PIPELINE=${VLLM_BUILD_PIPELINE:-local} \