feat(ci): add package generation

stack-info: PR: #592, branch: aws-nslick/stack/33

.docker/Dockerfile.dnf

@@ -0,0 +1,44 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+ARG FAMILY=fedora
+ARG VERSION=rawhide
+ARG VARIANT=cuda
+ARG CUDA_DISTRO
+ARG AWS_BUILD
+ARG ENABLE_POWERTOOLS
+
+# Install EFA-installer deps.
+FROM ${FAMILY}:${VERSION} AS builder
+ARG CUDA_DISTRO
+ARG ENABLE_POWERTOOLS
+ENV CUDA_DISTRO=${CUDA_DISTRO}
+ENV ENABLE_POWERTOOLS=${ENABLE_POWERTOOLS}
+# Add NVIDIA repo for CUDA builds.
+COPY --from=efainstaller / /
+RUN --mount=type=cache,target=/var/cache/yum,sharing=locked \
+    --mount=type=cache,target=/var/cache/dnf,sharing=locked \
+    bash -c "cd /aws-efa-installer && dnf install -y gcc rpmdevtools rpmlint dnf-plugins-core util-linux && ./efa_installer.sh -n -l -k -d -y && rm -rf /aws-efa-installer" && \
+    dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/${CUDA_DISTRO}/$(uname -m)/cuda-${CUDA_DISTRO}.repo && \
+    ( test "${ENABLE_POWERTOOLS}" = "1" && sed -i 's/enabled=0/enabled=1/' /etc/yum.repos.d/Rocky-PowerTools.repo || /bin/true ) && \
+    dnf -y update && dnf -y upgrade
+RUN rpmdev-setuptree
+
+FROM builder AS environment
+ARG VARIANT
+ARG AWS_BUILD
+ENV VARIANT=${VARIANT}
+ENV AWS_BUILD=${AWS_BUILD}
+COPY --from=srpm . .
+RUN yum search hwloc
+RUN echo "%with_${VARIANT} 1" >> ~/.rpmmacros
+RUN echo "%with_platform_aws ${AWS_BUILD}" >> ~/.rpmmacros
+RUN --mount=type=cache,target=/var/cache/yum,sharing=locked \
+    --mount=type=cache,target=/var/cache/dnf,sharing=locked \
+    dnf -y install cuda-cudart-devel-12-6 && dnf -y builddep *.src.rpm && rpmbuild --rebuild *.src.rpm
+
+FROM scratch
+COPY --from=environment /root/rpmbuild/RPMS/**/* /

.docker/Dockerfile.dpkg

@@ -0,0 +1,65 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+ARG FAMILY=ubuntu
+ARG VERSION=latest
+ARG CUDA_DISTRO
+ARG DEBIAN_FRONTEND=noninteractive
+ARG AWS_BUILD
+
+FROM ${FAMILY}:${VERSION} AS build
+ARG CUDA_DISTRO
+ENV CUDA_DISTRO=${CUDA_DISTRO}
+ARG AWS_BUILD=0
+ENV AWS_BUILD=${AWS_BUILD}
+
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    apt-get update -y && apt-get install wget -y
+
+RUN wget https://developer.download.nvidia.com/compute/cuda/repos/${CUDA_DISTRO}/$(uname -m)/cuda-keyring_1.1-1_all.deb
+
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    dpkg -i cuda-keyring_1.1-1_all.deb
+
+COPY --from=efainstaller / .
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    bash -c "apt-get update -y && cd /aws-efa-installer && ./efa_installer.sh /efa_installer.sh -n -l -k -d -y && apt-get install -y autoconf automake libtool gcc g++ git libhwloc-dev make && rm -rf /aws-efa-installer"
+
+COPY --from=makedist / .
+RUN tar xvf ./aws-ofi-nccl*.tar.gz -C .
+RUN cd aws-ofi-nccl* && \
+	./configure --$(test "$ACCELERATOR" = "cuda" && echo "with-cuda=/usr/local/cuda" || echo "enable-neuron=yes") \
+	--prefix=/opt/amazon/libnccl-net-ofi$(test "$AWS_BUILD" -eq 0 || echo -n "-aws") \
+	--with-libfabric=/opt/amazon/efa \
+	--disable-tests \
+        --$(test "$AWS_BUILD" -eq 0 && echo -n "disable" || echo -n "enable")-platform-aws \
+	--with-mpi=no && make -j && make install
+
+FROM ubuntu:latest AS packager
+ARG FAMILY
+ARG VERSION
+ARG AWS_BUILD=0
+ENV AWS_BUILD=${AWS_BUILD}
+ENV FAMILY=${FAMILY}
+ENV VERSION=${VERSION}
+COPY --from=build /opt/amazon/ /opt/amazon/
+RUN find /opt/amazon/ | grep -E \.la$ | xargs rm
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    apt-get update -y && apt-get install -y ruby tar squashfs-tools binutils && gem install fpm
+RUN fpm \
+	-s dir -t deb \
+	--license Apache2.0 \
+	-p /libnccl-net-ofi$(test "$AWS_BUILD" -eq 0 || echo -n "-aws")-${FAMILY}-${VERSION}.deb \
+	--name nccl-net-ofi$(test "$AWS_BUILD" -eq 0 || echo -n "-aws") \
+	/opt/amazon/libnccl-net-ofi$(test "$AWS_BUILD" -eq 0 || echo -n "-aws")/=/opt/amazon/libnccl-net-ofi$(test "$AWS_BUILD" -eq 0 || echo -n "-aws")
+
+FROM scratch
+COPY --from=packager /libnccl-net-ofi* /
+

.docker/Dockerfile.efa

@@ -0,0 +1,15 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+FROM alpine:latest AS efa_installer_extracted
+ARG EFA_INSTALLER_VERSION=latest
+ENV EFA_INSTALLER_VERSION=${EFA_INSTALLER_VERSION}
+RUN apk add tar curl
+RUN mkdir /libfabric
+RUN curl -s -L https://efa-installer.amazonaws.com/aws-efa-installer-${EFA_INSTALLER_VERSION}.tar.gz | tar -xvzf - -C /
+
+FROM scratch
+COPY --from=efa_installer_extracted /aws-efa-installer /aws-efa-installer

.docker/Dockerfile.makedist

@@ -0,0 +1,26 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+ARG ACCELERATOR
+ARG BASE_IMAGE=ubuntu:22.04
+FROM ${BASE_IMAGE} AS distbuilder
+ARG ACCELERATOR
+ENV ACCELERATOR=${ACCELERATOR}
+RUN mkdir /aws-efa-installer
+COPY --from=efainstaller /aws-efa-installer /aws-efa-installer
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    bash -c "apt-get update -y && cd /aws-efa-installer && ./efa_installer.sh /efa_installer.sh -n -l -k -d -y && apt-get install -y autoconf automake libtool gcc git libhwloc-dev make && rm -rf /aws-efa-installer"
+COPY ../ /proj
+WORKDIR /proj
+RUN autoreconf -ivf
+RUN ./configure --with-libfabric=/opt/amazon/efa --$(test "$ACCELERATOR" = "cuda" && echo "with-cuda=/usr/local/cuda" || echo "enable-neuron=yes") --with-libfabric=/opt/amazon/efa
+RUN make dist
+RUN ls -lart
+RUN pwd
+
+FROM scratch
+COPY --from=distbuilder /proj/aws-ofi-nccl*.tar.gz /

.docker/Dockerfile.srpm

@@ -0,0 +1,18 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+FROM fedora:rawhide AS packitimg
+RUN dnf install -y packit mock
+
+FROM packitimg AS srpm
+RUN mkdir /proj
+WORKDIR /proj
+COPY --from=src . .
+COPY --from=makedist . .
+RUN packit srpm
+
+FROM scratch
+COPY --from=srpm /proj/*.src.rpm /

.docker/Dockerfile.yum

@@ -0,0 +1,42 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+ARG FAMILY=amazonlinux
+ARG VERSION=2
+ARG VARIANT=cuda
+ARG CUDA_DISTRO
+ARG AWS_BUILD
+
+# Install EFA-installer deps.
+FROM ${FAMILY}:${VERSION} AS builder
+ARG CUDA_DISTRO
+ENV CUDA_DISTRO=${CUDA_DISTRO}
+# Add NVIDIA repo for CUDA builds.
+COPY --from=efainstaller / /
+RUN --mount=type=cache,target=/var/cache/yum,sharing=locked \
+    --mount=type=cache,target=/var/cache/dnf,sharing=locked \
+    bash -c "cd /aws-efa-installer && yum install -y gcc rpmdevtools rpmlint yum-utils util-linux && ./efa_installer.sh -n -l -k -d -y && rm -rf /aws-efa-installer" && \
+    yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/${CUDA_DISTRO}/$(uname -m)/cuda-${CUDA_DISTRO}.repo && \
+    yum update -y
+RUN rpmdev-setuptree
+
+FROM builder AS environment
+ARG VARIANT
+ARG AWS_BUILD
+ARG TOOLKIT_VERSION=12-6
+ENV VARIANT=${VARIANT}
+ENV AWS_BUILD=${AWS_BUILD}
+ENV TOOLKIT_VERSION=${TOOLKIT_VERSION}
+COPY --from=srpm . .
+RUN echo "%with_${VARIANT} 1" >> ~/.rpmmacros
+RUN echo "%with_platform_aws ${AWS_BUILD}" >> ~/.rpmmacros
+RUN echo "%_cuda_toolkit_version ${TOOLKIT_VERSION}" >> ~/.rpmmacros
+RUN --mount=type=cache,target=/var/cache/yum,sharing=locked \
+    --mount=type=cache,target=/var/cache/dnf,sharing=locked \
+    yum install -y cuda-cudart-devel-${TOOLKIT_VERSION} && yum-builddep -y *.src.rpm && rpmbuild --rebuild *.src.rpm
+
+FROM scratch
+COPY --from=environment /root/rpmbuild/RPMS/**/* /

.github/workflows/packages.yaml

@@ -0,0 +1,93 @@
+name: Package Generation
+on:
+  push:
+    branches:
+      - master
+      - main
+      - v*
+  pull_request:
+
+jobs:
+  dist:
+    name: Call make dist
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+        with:
+          platforms: linux/amd64,linux/arm64
+      - name: docker buildx bake makedist
+        uses: docker/bake-action@v5
+        with:
+          set: |
+            *.cache-from=type=gha
+            *.cache-to=type=gha,mode=max
+          push: true
+          targets: makedist
+  srpm:
+    name: Generate a universal SRPM
+    needs: [ dist ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+        with:
+          platforms: linux/amd64,linux/arm64
+      - name: docker buildx bake srpm
+        uses: docker/bake-action@v5
+        with:
+          set: |
+            *.cache-from=type=gha
+            *.cache-to=type=gha,mode=max
+          push: true
+          targets: srpm
+  debs:
+    name: Generate Debian-like Packages
+    needs: [ dist ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+        with:
+          platforms: linux/amd64,linux/arm64
+      - name: docker buildx bake debs
+        uses: docker/bake-action@v5
+        with:
+          set: |
+            *.cache-from=type=gha
+            *.cache-to=type=gha,mode=max
+          push: ${{ github.event_name != 'pull_request' }}
+          targets: debs
+  rpms:
+    name: Generate RPM-like Packages
+    needs: [ srpm ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+        with:
+          platforms: linux/amd64,linux/arm64
+      - name: docker buildx bake rpms
+        uses: docker/bake-action@v5
+        with:
+          set: |
+            *.cache-from=type=gha
+            *.cache-to=type=gha,mode=max
+          push: ${{ github.event_name != 'pull_request' }}
+          targets: rpms

.gitignore

@@ -75,3 +75,5 @@ m4/lt~obsolete.m4
 .idea/
 .devenv/
 .direnv
+*.src.rpm
+dockerbld

.packit.yml

@@ -0,0 +1,20 @@
+---
+# vi:ts=2 sw=2 et:
+#
+# Docs: https://packit.dev/docs/
+
+
+srpm_build_deps:
+  - git
+
+actions:
+  get-current-version:
+    - bash -c "cat .version"
+  create-archive:
+    - bash -c "echo ./aws-ofi-nccl-${PACKIT_PROJECT_VERSION}.tar.gz"
+
+specfile_path: .packit/libnccl-net-ofi.spec
+upstream_package_name: libnccl-net-ofi
+downstream_package_name: libnccl-net-ofi
+release_suffix: "{PACKIT_PROJECT_BRANCH}"
+update_release: false