Refactor nvidia symlink creation

Avoid creation of broken symlink when dealing with VGPU drivers.
ComputeCanada · Feb 23, 2021 · 54ea3ec · 54ea3ec
1 parent 8664e64
commit 54ea3ec
Show file tree

Hide file tree

Showing 2 changed files with 70 additions and 31 deletions.
diff --git a/site/profile/facts.d/nvidia_driver_vers.sh b/site/profile/facts.d/nvidia_driver_vers.sh
@@ -1,16 +1,5 @@
 #!/bin/sh
-PROCESSOR=$(uname -p)
-VERSION="$(source /etc/os-release; echo $VERSION_ID)"
-PACKAGE="cuda-drivers"
-PACKAGE_REGEX="${PACKAGE}-\([0-9.]\{1,\}\)[-0-9]*\.${PROCESSOR}"
-DRIVER_VERSION=$(test -f /usr/sbin/dkms && /usr/sbin/dkms status | grep -m 1 -Po 'nvidia, \K(\d+.\d+[\.]\d*)')
-# If that didn't work let's try nvidia-smi
-if [ -z $DRIVER_VERSION ]; then
-    DRIVER_VERSION=$(if [ -x "$(command -v nvidia-smi)" ]; then nvidia-smi --query-gpu=driver_version --format=csv,noheader; fi)
+if [ -e /proc/driver/nvidia ]; then
+    DRIVER_VERSION=$(grep -m 1 -Po 'NVRM version:.* \K(\d+\.\d+\.\d+)' /proc/driver/nvidia/version)
 fi
-if [ -z $DRIVER_VERSION ]; then
-    BASE_URL="http://developer.download.nvidia.com/compute/cuda/repos"
-    CUDA_REPO_GZ=$(curl -s ${BASE_URL}/rhel${VERSION}/${PROCESSOR}/repodata/repomd.xml | sed '2 s/xmlns=".*"//g' | xmllint --xpath 'string(/repomd/data[@type="primary"]/location/@href)' -)
-    DRIVER_VERSION=$(curl -s ${BASE_URL}/rhel${VERSION}/${PROCESSOR}/${CUDA_REPO_GZ} | gunzip | sed -n "s/^.*\"${PACKAGE_REGEX}\.rpm\".*$/\1/p" | sort -V | tail -n1)
-fi
-echo "{ 'nvidia_driver_version' : '${DRIVER_VERSION}' }"
+echo "{ 'nvidia_driver_version' : '${DRIVER_VERSION}' }"
diff --git a/site/profile/manifests/gpu.pp b/site/profile/manifests/gpu.pp
@@ -58,24 +58,32 @@
     ensure => directory
   }
 
-  $driver_ver = $::facts['nvidia_driver_version']
   $nvidia_libs = [
-    "libnvidia-ml.so.${driver_ver}", 'libnvidia-ml.so.1', 'libnvidia-fbc.so.1',
-    "libnvidia-fbc.so.${driver_ver}", 'libnvidia-ifr.so.1', "libnvidia-ifr.so.${driver_ver}",
-    'libcuda.so', 'libcuda.so.1', "libcuda.so.${driver_ver}", "libnvcuvid.so.${driver_ver}",
-    'libnvcuvid.so.1', "libnvidia-compiler.so.${driver_ver}", 'libnvidia-encode.so.1',
-    "libnvidia-encode.so.${driver_ver}", "libnvidia-fatbinaryloader.so.${driver_ver}",
-    'libnvidia-opencl.so.1', "libnvidia-opencl.so.${driver_ver}", 'libnvidia-opticalflow.so.1',
-    "libnvidia-opticalflow.so.${driver_ver}", 'libnvidia-ptxjitcompiler.so.1', "libnvidia-ptxjitcompiler.so.${driver_ver}",
-    'libnvcuvid.so', 'libnvidia-cfg.so', 'libnvidia-encode.so',
-    'libnvidia-fbc.so', 'libnvidia-ifr.so', 'libnvidia-ml.so',
-    'libnvidia-ptxjitcompiler.so', 'libEGL_nvidia.so.0', "libEGL_nvidia.so.${driver_ver}",
-    'libGLESv1_CM_nvidia.so.1', "libGLESv1_CM_nvidia.so.${driver_ver}", 'libGLESv2_nvidia.so.2',
-    "libGLESv2_nvidia.so.${driver_ver}", 'libGLX_indirect.so.0', 'libGLX_nvidia.so.0',
-    "libGLX_nvidia.so.${driver_ver}", "libnvidia-cbl.so.${driver_ver}", 'libnvidia-cfg.so.1',
-    "libnvidia-cfg.so.${driver_ver}", "libnvidia-eglcore.so.${driver_ver}", "libnvidia-glcore.so.${driver_ver}",
-    "libnvidia-glsi.so.${driver_ver}", "libnvidia-glvkspirv.so.${driver_ver}", "libnvidia-rtcore.so.${driver_ver}",
-    "libnvidia-tls.so.${driver_ver}", 'libnvoptix.so.1', "libnvoptix.so.${driver_ver}"]
+    'libcuda.so.1',
+    'libcuda.so',
+    'libEGL_nvidia.so.0',
+    'libGLESv1_CM_nvidia.so.1',
+    'libGLESv2_nvidia.so.2',
+    'libGLX_indirect.so.0',
+    'libGLX_nvidia.so.0',
+    'libnvcuvid.so.1',
+    'libnvcuvid.so',
+    'libnvidia-cfg.so.1',
+    'libnvidia-cfg.so',
+    'libnvidia-encode.so.1',
+    'libnvidia-encode.so',
+    'libnvidia-fbc.so.1',
+    'libnvidia-fbc.so',
+    'libnvidia-ifr.so.1',
+    'libnvidia-ifr.so',
+    'libnvidia-ml.so.1',
+    'libnvidia-ml.so',
+    'libnvidia-opencl.so.1',
+    'libnvidia-opticalflow.so.1',
+    'libnvidia-ptxjitcompiler.so.1',
+    'libnvidia-ptxjitcompiler.so',
+    'libnvoptix.so.1',
+  ]
 
   $nvidia_libs.each |String $lib| {
     file { "/usr/lib64/nvidia/${lib}":
@@ -84,6 +92,48 @@
       seltype => 'lib_t'
     }
   }
+
+  # WARNING : since the fact is computed before Puppet agent run,
+  # on a clean host, the  symbolic links to the NVIDIA libraries
+  # that include the version number will be created on the
+  # second Puppet run only.
+  $driver_vers = $::facts['nvidia_driver_version']
+  if $driver_vers != '' {
+    $nvidia_libs_vers = [
+      "libcuda.so.${driver_vers}",
+      "libEGL_nvidia.so.${driver_vers}",
+      "libGLESv1_CM_nvidia.so.${driver_vers}",
+      "libGLESv2_nvidia.so.${driver_vers}",
+      "libGLX_nvidia.so.${driver_vers}",
+      "libnvcuvid.so.${driver_vers}",
+      "libnvidia-cbl.so.${driver_vers}",
+      "libnvidia-cfg.so.${driver_vers}",
+      "libnvidia-compiler.so.${driver_vers}",
+      "libnvidia-eglcore.so.${driver_vers}",
+      "libnvidia-encode.so.${driver_vers}",
+      "libnvidia-fatbinaryloader.so.${driver_vers}",
+      "libnvidia-fbc.so.${driver_vers}",
+      "libnvidia-glcore.so.${driver_vers}",
+      "libnvidia-glsi.so.${driver_vers}",
+      "libnvidia-glvkspirv.so.${driver_vers}",
+      "libnvidia-ifr.so.${driver_vers}",
+      "libnvidia-ml.so.${driver_vers}",
+      "libnvidia-opencl.so.${driver_vers}",
+      "libnvidia-opticalflow.so.${driver_vers}",
+      "libnvidia-ptxjitcompiler.so.${driver_vers}",
+      "libnvidia-rtcore.so.${driver_vers}",
+      "libnvidia-tls.so.${driver_vers}",
+      "libnvoptix.so.${driver_vers}"
+    ]
+
+    $nvidia_libs_vers.each |String $lib| {
+      file { "/usr/lib64/nvidia/${lib}":
+        ensure  => link,
+        target  => "/usr/lib64/${lib}",
+        seltype => 'lib_t'
+      }
+    }
+  }
 }
 
 class profile::gpu::install::passthrough(Array[String] $packages) {