From db45ae9c3ddd08f28bd89f177c522007b172a840 Mon Sep 17 00:00:00 2001
From: xlliu-scitix <xlliu@scitix.ai>
Date: Sun, 21 Dec 2025 15:40:24 +0000
Subject: [PATCH] update dockerfile

---
 .github/workflows/pre-check.yml        |  2 +-
 docker/Dockerfile.cuda12.x.ubuntu20.04 | 57 ++++++++++++++++----------
 docker/Dockerfile.cuda13.x.ubuntu22.04 | 18 +++++++-
 3 files changed, 53 insertions(+), 24 deletions(-)

diff --git a/.github/workflows/pre-check.yml b/.github/workflows/pre-check.yml
index f8f75d0..0b8d414 100644
--- a/.github/workflows/pre-check.yml
+++ b/.github/workflows/pre-check.yml
@@ -21,7 +21,7 @@ jobs:
           docker buildx build \
             -f docker/Dockerfile.cuda13.x.ubuntu22.04 \
             --platform linux/amd64 \
-            --target build \
+            --target package \
             --output type=local,dest=dist \
             .
       - name: List artifacts
diff --git a/docker/Dockerfile.cuda12.x.ubuntu20.04 b/docker/Dockerfile.cuda12.x.ubuntu20.04
index 540eff0..b358aa5 100644
--- a/docker/Dockerfile.cuda12.x.ubuntu20.04
+++ b/docker/Dockerfile.cuda12.x.ubuntu20.04
@@ -43,30 +43,30 @@ WORKDIR /workspace
 # 1. Base build dependencies
 # -------------------------
 RUN mv /etc/apt/sources.list.d/cuda*.list /tmp/disabled-cuda.list || true && \
-    apt-get -o Acquire::http::No-Cache=true update && \
+    { apt-get -o Acquire::http::No-Cache=true update > build.log 2>&1 && \
     apt-get install -y --no-install-recommends \
         build-essential gcc g++ curl git wget ca-certificates \
         make automake autoconf libtool pkg-config \
-        python3 python3-pip gzip xz-utils makeself > /dev/null 2>&1 && \
-    rm -rf /var/lib/apt/lists/*
+        python3 python3-pip gzip xz-utils >> build.log 2>&1 && \
+    rm -rf /var/lib/apt/lists/* && rm -f build.log; } || (cat build.log && false)
 
 # -------------------------
 # 2. Install CUDA keyring and restore NVIDIA repository
 # -------------------------
-RUN wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-keyring_1.1-1_all.deb && \
-    dpkg -i cuda-keyring_1.1-1_all.deb && \
-    apt-get update > /dev/null 2>&1
+RUN { wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-keyring_1.1-1_all.deb > build.log 2>&1 && \
+    dpkg -i cuda-keyring_1.1-1_all.deb >> build.log 2>&1 && \
+    apt-get update >> build.log 2>&1 && rm -f build.log; } || (cat build.log && false)
 
 # -------------------------
 # 3. Install NCCL (pinned version)
 # -------------------------
 RUN apt-mark unhold libnccl2 libnccl-dev || true && \
-    apt-get install -y --no-install-recommends \
+    { apt-get install -y --no-install-recommends \
     libnccl2=${NCCL_PACKAGE_VERSION} \
-    libnccl-dev=${NCCL_PACKAGE_VERSION}  > /dev/null 2>&1 && \
-    apt-mark hold libnccl2 libnccl-dev && \
-    ldconfig && \
-    rm -rf /var/lib/apt/lists/*
+    libnccl-dev=${NCCL_PACKAGE_VERSION} > build.log 2>&1 && \
+    apt-mark hold libnccl2 libnccl-dev >> build.log 2>&1 && \
+    ldconfig >> build.log 2>&1 && \
+    rm -rf /var/lib/apt/lists/* && rm -f build.log; } || (cat build.log && false)
 
 # -------------------------
 # 4. Build OpenMPI from source
@@ -75,15 +75,15 @@ RUN wget https://download.open-mpi.org/release/open-mpi/v${MPI_SERIES}/openmpi-$
     tar zxvf openmpi-${MPI_VERSION}.tar.gz > /dev/null 2>&1 && \
     cd openmpi-${MPI_VERSION} && \
     ./configure --prefix=/usr/local/sihpc --with-cuda=/usr/local/cuda > /dev/null  2>&1 && \
-    make -j$(nproc) > /dev/null 2>&1 && make install && \
+    make -j$(nproc) > /dev/null 2>&1 && make install > /dev/null 2>&1 && \
     rm -rf /workspace/openmpi-${MPI_VERSION} /workspace/openmpi-${MPI_VERSION}.tar.gz
 
 # -------------------------
 # 5. Build nccl-tests
 # -------------------------
-RUN git clone https://github.com/scitix/nccl-tests.git -b sicl && \
+RUN { git clone --depth 1 --single-branch -b sicl https://github.com/scitix/nccl-tests.git > build.log 2>&1 && \
     cd nccl-tests && \
-    make MPI=1 MPI_HOME=/usr/local/sihpc && \
+    make MPI=1 MPI_HOME=/usr/local/sihpc > build.log 2>&1 && rm -f build.log || (cat build.log && false); } && \
     mkdir -p /usr/local/sihpc/libexec/nccl-tests && \
     cp -rf build/*_perf /usr/local/sihpc/libexec/nccl-tests/ && \
     mkdir -p /usr/local/sihpc/bin && \
@@ -97,10 +97,11 @@ RUN git clone https://github.com/scitix/nccl-tests.git -b sicl && \
 # -------------------------
 # 6. Collect runtime libraries (strict selection)
 # -------------------------
-RUN set -eux && \
-    mkdir -p /usr/local/sihpc/lib && \
-    cp /usr/local/cuda/lib64/libcudart* /usr/local/sihpc/lib/ && \
-    cp /usr/lib/x86_64-linux-gnu/libnccl.so* /usr/local/sihpc/lib/
+RUN { set -e && \
+    mkdir -p /usr/local/sihpc/lib > build.log 2>&1 && \
+    cp /usr/local/cuda/lib64/libcudart* /usr/local/sihpc/lib/ >> build.log 2>&1 && \
+    cp /usr/lib/x86_64-linux-gnu/libnccl.so* /usr/local/sihpc/lib/ >> build.log 2>&1 && \
+    rm -f build.log; } || (cat build.log && false)
     # cp /lib/x86_64-linux-gnu/libltdl.so.7.3.1 /usr/local/sihpc/lib/ && \
     # cp /usr/lib/x86_64-linux-gnu/libhwloc.so* /usr/local/sihpc/lib/ && \
     # cp /usr/lib/x86_64-linux-gnu/libevent_core* /usr/local/sihpc/lib/ && \
@@ -125,15 +126,29 @@ RUN cd /usr/local/sihpc/lib && \
     # ln -sf libltdl.so.7.3.1 libltdl.so.7 && \
     # ln -sf libltdl.so.7 libltdl.so
 
+###########################
+# Package Stage
+###########################
+FROM ubuntu:20.04 AS package
+
+# Re-declare args for this stage (values are inherited)
+ARG UBUNTU_VERSION
+ARG NCCL_PACKAGE_VERSION
+ARG MPI_VERSION
+ARG BUILD_DATE
+
 # Expose versions/date as environment variables for runtime shell expansion
 ENV NCCL_PACKAGE_VERSION=${NCCL_PACKAGE_VERSION} \
     MPI_VERSION=${MPI_VERSION} \
     BUILD_DATE=${BUILD_DATE}
 
+COPY --from=build /usr/local/sihpc /usr/local/sihpc
+
 WORKDIR /
-RUN SAFE_NCCL_PKG=$(printf '%s\n' "${NCCL_PACKAGE_VERSION}" | tr '+' '-') && \
+RUN apt-get update && apt-get install -y --no-install-recommends makeself && \
+    SAFE_NCCL_PKG=$(printf '%s\n' "${NCCL_PACKAGE_VERSION}" | tr '+' '-') && \
     PACKAGE_FILENAME="sicl-nccl${SAFE_NCCL_PKG}-ompi${MPI_VERSION}-ubuntu${UBUNTU_VERSION}-${BUILD_DATE}.run" && \
-    makeself --gzip /usr/local/sihpc \
+    { makeself --gzip /usr/local/sihpc \
     "${PACKAGE_FILENAME}" \
     "SiHPC MPI + NCCL + NCCL-tests Portable Installer" \
-    ./bin/install_sihpc > /dev/null 2>&1
\ No newline at end of file
+      ./bin/install_sihpc > build.log 2>&1 && rm -f build.log; } || (cat build.log && false)
\ No newline at end of file
diff --git a/docker/Dockerfile.cuda13.x.ubuntu22.04 b/docker/Dockerfile.cuda13.x.ubuntu22.04
index ae389ea..82cb17d 100644
--- a/docker/Dockerfile.cuda13.x.ubuntu22.04
+++ b/docker/Dockerfile.cuda13.x.ubuntu22.04
@@ -46,7 +46,7 @@ RUN { apt-get -o Acquire::http::No-Cache=true update > build.log 2>&1 && \
     apt-get install -y --no-install-recommends \
         build-essential gcc g++ curl git wget ca-certificates \
         make automake autoconf libtool pkg-config \
-        python3 python3-pip gzip xz-utils makeself >> build.log 2>&1 && \
+        python3 python3-pip gzip xz-utils >> build.log 2>&1 && \
     rm -rf /var/lib/apt/lists/* && rm -f build.log; } || (cat build.log && false)
 
 # -------------------------
@@ -123,13 +123,27 @@ RUN cd /usr/local/sihpc/lib && \
 #    ln -sf libltdl.so.7.3.1 libltdl.so.7 && \
 #    ln -sf libltdl.so.7 libltdl.so
 
+###########################
+# Package Stage
+###########################
+FROM ubuntu:20.04 AS package
+
+# Re-declare args for this stage (values are inherited)
+ARG UBUNTU_VERSION
+ARG NCCL_PACKAGE_VERSION
+ARG MPI_VERSION
+ARG BUILD_DATE
+
 # Expose versions/date as environment variables for runtime shell expansion
 ENV NCCL_PACKAGE_VERSION=${NCCL_PACKAGE_VERSION} \
     MPI_VERSION=${MPI_VERSION} \
     BUILD_DATE=${BUILD_DATE}
 
+COPY --from=build /usr/local/sihpc /usr/local/sihpc
+
 WORKDIR /
-RUN SAFE_NCCL_PKG=$(printf '%s\n' "${NCCL_PACKAGE_VERSION}" | tr '+' '-') && \
+RUN apt-get update && apt-get install -y --no-install-recommends makeself && \
+    SAFE_NCCL_PKG=$(printf '%s\n' "${NCCL_PACKAGE_VERSION}" | tr '+' '-') && \
     PACKAGE_FILENAME="sicl-nccl${SAFE_NCCL_PKG}-ompi${MPI_VERSION}-ubuntu${UBUNTU_VERSION}-${BUILD_DATE}.run" && \
     { makeself --gzip /usr/local/sihpc \
       "${PACKAGE_FILENAME}" \