AZP/RELEASE: rm cuda-compat & separate gdrcopy

openucx · Feb 27, 2023 · cd8182b · cd8182b
1 parent f9e2f91
commit cd8182b
Show file tree

Hide file tree

Showing 6 changed files with 18 additions and 20 deletions.
diff --git a/buildlib/az-distro-release.yml b/buildlib/az-distro-release.yml
@@ -72,10 +72,11 @@ jobs:
           dpkg-buildpackage -us -uc -Pcuda
           cd ..                             # Move back to the working directory
           find . -name '*.deb'
-          VER="${POSTFIX#ucx-}"   # Remove 'ucx' prefix from the POSTFIX string
+          VER="${POSTFIX#ucx-}"   # Remove 'ucx-' prefix from the POSTFIX string
           # Rename DEB files
           find . -name "ucx*.deb" -exec bash -c 'mv "$1" "${1%%_*}-'"${VER}"'.deb"' _ {} \;
           dpkg-deb -I "ucx-${VER}.deb"
+          dpkg-deb -I "ucx-cuda-${VER}.deb"
           tar -cjf "${AZ_ARTIFACT_NAME}" *.deb  # Package all DEBs
           tar -tjf "${AZ_ARTIFACT_NAME}"
         displayName: Build DEB package

diff --git a/buildlib/azure-pipelines-release.yml b/buildlib/azure-pipelines-release.yml
@@ -26,7 +26,7 @@ resources:
     - container: ubuntu20_cuda11
       image: rdmz-harbor.rdmz.labs.mlnx/ucx/ubuntu20.04-mofed5-cuda11:3
     - container: ubuntu22_cuda11
-      image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/ubuntu22.04-mofed5-cuda11:3
+      image: rdmz-harbor.rdmz.labs.mlnx/ucx/ubuntu22.04-mofed5-cuda11:3
 
 stages:
   - stage: Prepare

diff --git a/buildlib/dockers/docker-compose.yml b/buildlib/dockers/docker-compose.yml
@@ -1,5 +1,8 @@
 version: "3.4"
 
+# CUDA runtime <-> driver versions compatibility per OS:
+# https://developer.nvidia.com/cuda-11-7-0-download-archive?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=22.04&target_type=runfile_local
+
 services:
   centos7-mofed5-cuda11:
     image: centos7-mofed5-cuda11:2
@@ -34,16 +37,6 @@ services:
         MOFED_OS: rhel8.2
         CUDA_VERSION: 11.4.0
         OS_VERSION: 8
-  ubuntu16.04-mofed5-cuda11:
-    image: ubuntu16.04-mofed5-cuda11:3
-    build:
-      context: .
-      network: host
-      dockerfile: ubuntu-release.Dockerfile
-      args:
-        MOFED_VERSION: 5.0-1.0.0.0
-        UBUNTU_VERSION: 16.04
-        CUDA_VERSION: 11.2.0
   ubuntu18.04-mofed5-cuda11:
     image: ubuntu18.04-mofed5-cuda11:3
     build:
@@ -54,6 +47,7 @@ services:
         MOFED_VERSION: 5.0-1.0.0.0
         UBUNTU_VERSION: 18.04
         CUDA_VERSION: 11.4.0
+        DRIVER_VERSION: 470
   ubuntu20.04-mofed5-cuda11:
     image: ubuntu20.04-mofed5-cuda11:3
     build:
@@ -64,6 +58,7 @@ services:
         MOFED_VERSION: 5.0-1.0.0.0
         UBUNTU_VERSION: 20.04
         CUDA_VERSION: 11.4.0
+        DRIVER_VERSION: 470
   ubuntu22.04-mofed5-cuda11:
     image: ubuntu22.04-mofed5-cuda11:3
     build:
@@ -74,3 +69,4 @@ services:
         MOFED_VERSION: 5.4-3.6.8.1
         UBUNTU_VERSION: 22.04
         CUDA_VERSION: 11.7.0
+        DRIVER_VERSION: 515
diff --git a/buildlib/dockers/push-release-images.sh b/buildlib/dockers/push-release-images.sh
@@ -1,11 +1,11 @@
-#!/bin/bash -eE
+#!/bin/bash -eEx
 
 # shellcheck disable=SC2086
 basedir=$(cd "$(dirname $0)" && pwd)
 
 registry=harbor.mellanox.com/ucx
 
-images=$(awk '/image:/ {print $2}' "${basedir}/docker-compose.yml")
+images=$(awk '!/#/ && /image:/ {print $2}' "${basedir}/docker-compose.yml")
 for img in $images; do
     target_name="${registry}/${img}"
     docker tag ${img} ${target_name}

diff --git a/buildlib/dockers/ubuntu-release.Dockerfile b/buildlib/dockers/ubuntu-release.Dockerfile
@@ -1,7 +1,8 @@
-ARG CUDA_VERSION=10.1
-ARG UBUNTU_VERSION=16.04
+ARG CUDA_VERSION
+ARG UBUNTU_VERSION
 FROM nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VERSION}
 
+ARG DRIVER_VERSION
 RUN apt-get update && \
     DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends tzdata && \
     apt-get install -y \
@@ -14,13 +15,15 @@ RUN apt-get update && \
         libcap2 \
         libnuma-dev \
         libtool \
+        libnvidia-compute-${DRIVER_VERSION} \
         make \
         maven \
         udev \
         wget \
         environment-modules \
         pkg-config \
-    && apt-get remove -y openjdk-11-* || apt-get autoremove -y \
+        sudo \
+    && apt-get remove -y openjdk-11-* cuda-compat* || apt-get autoremove -y \
     && apt-get clean && rm -rf /var/lib/apt/lists/*
 
 # MOFED
@@ -46,5 +49,3 @@ ENV CPATH /usr/local/cuda/include:${CPATH}
 ENV LD_LIBRARY_PATH /usr/local/cuda/lib64:/usr/local/cuda/compat:${LD_LIBRARY_PATH}
 ENV LIBRARY_PATH /usr/local/cuda/lib64:/usr/local/cuda/compat:${LIBRARY_PATH}
 ENV PATH /usr/local/cuda/compat:${PATH}
-
-RUN ml_stub=$(find /usr -name libnvidia-ml.so) && ln -s $ml_stub /usr/lib/x86_64-linux-gnu/libnvidia-ml.so.1
diff --git a/debian/control.in b/debian/control.in
@@ -41,7 +41,7 @@ Package: ucx-gdrcopy
 Section: libs
 Depends: ${misc:Depends}, ${shlibs:Depends}
 Architecture: any
-Build-Profiles: <cuda>
+Build-Profiles: <gdrcopy>
 Description: Unified Communication X - gdrcopy support
  UCX is a communication library implementing high-performance messaging.
  .