nvidia repo downloads

cookbooks/aws-parallelcluster-platform/resources/fabric_manager/partial/_fabric_manager_common.rb

@@ -27,15 +27,15 @@
   node_attributes "dump node attributes"
 
   # Add NVIDIA repo for fabric manager and datacenter-gpu-manager
-  nvidia_repo 'add nvidia repository' do
-    action :add
-  end
+  # nvidia_repo 'add nvidia repository' do
+  #   action :add
+  # end
 
   action_install_package
 
-  nvidia_repo 'remove nvidia repository' do
-    action :remove
-  end
+  # nvidia_repo 'remove nvidia repository' do
+  #   action :remove
+  # end
 end
 
 action :configure do

cookbooks/aws-parallelcluster-platform/resources/fabric_manager/partial/_fabric_manager_install_debian.rb

@@ -15,9 +15,27 @@
 action :install_package do
   # For ubuntu, CINC17 apt-package resources need full versions for `version`
   execute "install_fabricmanager_for_ubuntu" do
-    command "apt -y install #{fabric_manager_package}=#{fabric_manager_version} "\
+    bash "Install #{fabric_manager_package}" do
+      user 'root'
+      code <<-FABRIC_MANAGER
+      set -e
+      aws s3 cp #{fabric_manager_url} #{fabric_manager_package}-#{fabric_manager_version}.deb
+      FABRIC_MANAGER
+      retries 3
+      retry_delay 5
+    end
+
+    command "apt -y install #{fabric_manager_package}-#{fabric_manager_version}.deb "\
             "&& apt-mark hold #{fabric_manager_package}"
     retries 3
     retry_delay 5
   end
 end
+
+def arch_suffix
+  arm_instance? ? 'arm64' : 'amd64'
+end
+
+def fabric_manager_url
+  "s3://hgreebe-dependencies/archives/dependencies/nvidia_fabric/#{platform}/#{fabric_manager_package}_#{fabric_manager_version}-1_#{arch_suffix}.deb"
+end

cookbooks/aws-parallelcluster-platform/resources/fabric_manager/partial/_fabric_manager_install_rhel.rb

@@ -18,10 +18,19 @@
     user 'root'
     code <<-FABRIC_MANAGER_INSTALL
     set -e
-    yum install -y #{fabric_manager_package}-#{fabric_manager_version}
+    aws s3 cp #{fabric_manager_url} #{fabric_manager_package}-#{fabric_manager_version}.rpm
+    yum install -y #{fabric_manager_package}-#{fabric_manager_version}.rpm
     yum versionlock #{fabric_manager_package}
     FABRIC_MANAGER_INSTALL
     retries 3
     retry_delay 5
   end
 end
+
+def arch_suffix
+  arm_instance? ? 'aarch64' : 'x86_64'
+end
+
+def fabric_manager_url
+  "s3://hgreebe-dependencies/archives/dependencies/nvidia_fabric/#{platform}/#{fabric_manager_package}-#{fabric_manager_version}-1.#{arch_suffix}.rpm"
+end

cookbooks/aws-parallelcluster-platform/resources/nvidia_dcgm/nvidia_dcgm_amazon2.rb

@@ -15,6 +15,7 @@
 provides :nvidia_dcgm, platform: 'amazon', platform_version: '2'
 
 use 'partial/_nvidia_dcgm_common.rb'
+use 'partial/_nvidia_dcgm_rhel.rb'
 
 def _nvidia_dcgm_enabled
   !arm_instance? && _nvidia_enabled

cookbooks/aws-parallelcluster-platform/resources/nvidia_dcgm/nvidia_dcgm_centos7.rb

@@ -17,6 +17,7 @@
 end
 
 use 'partial/_nvidia_dcgm_common.rb'
+use 'partial/_nvidia_dcgm_rhel.rb'
 
 def _nvidia_dcgm_enabled
   !arm_instance? && _nvidia_enabled

cookbooks/aws-parallelcluster-platform/resources/nvidia_dcgm/nvidia_dcgm_redhat8.rb

@@ -17,6 +17,7 @@
 end
 
 use 'partial/_nvidia_dcgm_common.rb'
+use 'partial/_nvidia_dcgm_rhel.rb'
 
 def _nvidia_dcgm_enabled
   _nvidia_enabled

cookbooks/aws-parallelcluster-platform/resources/nvidia_dcgm/nvidia_dcgm_rocky8.rb

@@ -17,6 +17,7 @@
 end
 
 use 'partial/_nvidia_dcgm_common.rb'
+use 'partial/_nvidia_dcgm_rhel.rb'
 
 def _nvidia_dcgm_enabled
   _nvidia_enabled

cookbooks/aws-parallelcluster-platform/resources/nvidia_dcgm/nvidia_dcgm_ubuntu20+.rb

@@ -17,11 +17,12 @@
 end
 
 use 'partial/_nvidia_dcgm_common.rb'
+use 'partial/_nvidia_dcgm_debian.rb'
 
 def _nvidia_dcgm_enabled
   _nvidia_enabled
 end
 
-def package_version
-  "1:#{node['cluster']['nvidia']['dcgm_version']}" # The single digit "1" is epoch version. Without the "1", package install fails because version does not exist. See details here: https://askubuntu.com/questions/441879/why-do-some-packages-have-extra-numbers-before-a-colon-on-the-front-of-their-ver
-end
+# def package_version
+#   "1:#{node['cluster']['nvidia']['dcgm_version']}" # The single digit "1" is epoch version. Without the "1", package install fails because version does not exist. See details here: https://askubuntu.com/questions/441879/why-do-some-packages-have-extra-numbers-before-a-colon-on-the-front-of-their-ver
+# end

cookbooks/aws-parallelcluster-platform/resources/nvidia_dcgm/partial/_nvidia_dcgm_common.rb

@@ -21,25 +21,23 @@
   return unless _nvidia_dcgm_enabled
 
   # Add NVIDIA repo for fabric manager and datacenter-gpu-manager
-  nvidia_repo 'add nvidia repository' do
-    action :add
-  end
-
-  package 'datacenter-gpu-manager' do
-    retries 3
-    retry_delay 5
-    version package_version
-  end
-
-  nvidia_repo 'remove nvidia repository' do
-    action :remove
-  end
+  # nvidia_repo 'add nvidia repository' do
+  #   action :add
+  # end
+
+  # package 'datacenter-gpu-manager' do
+  #   retries 3
+  #   retry_delay 5
+  #   version package_version
+  # end
+
+  action_install_package
+
+  # nvidia_repo 'remove nvidia repository' do
+  #   action :remove
+  # end
 end
 
 def _nvidia_enabled
   nvidia_enabled.nil? ? ['yes', true].include?(node['cluster']['nvidia']['enabled']) : nvidia_enabled
-end
-
-def package_version
-  node['cluster']['nvidia']['dcgm_version']
-end
+end

cookbooks/aws-parallelcluster-platform/resources/nvidia_dcgm/partial/_nvidia_dcgm_debian.rb

@@ -0,0 +1,47 @@
+# frozen_string_literal: true
+#
+# Copyright:: 2013-2023 Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License").
+# You may not use this file except in compliance with the License.
+# A copy of the License is located at
+#
+# http://aws.amazon.com/apache2.0/
+#
+# or in the "LICENSE.txt" file accompanying this file.
+# This file is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+action :install_package do
+
+  bash "Install #{dcgm_package}" do
+    user 'root'
+    code <<-DCGM_INSTALL
+    set -e
+    aws s3 cp #{dcgm_url} #{dcgm_package}-#{package_version}.deb
+    DCGM_INSTALL
+    retries 3
+    retry_delay 5
+  end
+
+  command "apt -y install #{dcgm_package}-#{package_version}.deb "
+  retries 3
+  retry_delay 5
+
+end
+
+def dcgm_url
+  "s3://hgreebe-dependencies/archives/dependencies/nvidia_dcgm/#{platform}/#{dcgm_package}_#{package_version}_#{arch_suffix}.deb"
+end
+
+def dcgm_package
+  'datacenter-gpu-manager'
+end
+
+def arch_suffix
+  arm_instance? ? 'arm64' : 'amd64'
+end
+
+def package_version
+  node['cluster']['nvidia']['dcgm_version']
+end

cookbooks/aws-parallelcluster-platform/resources/nvidia_dcgm/partial/_nvidia_dcgm_rhel.rb

@@ -0,0 +1,44 @@
+# frozen_string_literal: true
+#
+# Copyright:: 2013-2023 Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License").
+# You may not use this file except in compliance with the License.
+# A copy of the License is located at
+#
+# http://aws.amazon.com/apache2.0/
+#
+# or in the "LICENSE.txt" file accompanying this file.
+# This file is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+action :install_package do
+
+  bash "Install #{dcgm_package}" do
+    user 'root'
+    code <<-DCGM_INSTALL
+    set -e
+    aws s3 cp #{dcgm_url} #{dcgm_package}-#{package_version}.rpm
+    yum install -y #{dcgm_package}-#{package_version}.rpm
+    DCGM_INSTALL
+    retries 3
+    retry_delay 5
+  end
+
+end
+
+def dcgm_url
+  "s3://hgreebe-dependencies/archives/dependencies/nvidia_dcgm/#{platform}/#{dcgm_package}-#{package_version}-1-#{arch_suffix}.rpm"
+end
+
+def dcgm_package
+  'datacenter-gpu-manager'
+end
+
+def arch_suffix
+  arm_instance? ? 'aarch64' : 'x86_64'
+end
+
+def package_version
+  node['cluster']['nvidia']['dcgm_version']
+end