Fix fabric manager install to install from a local package

cookbooks/aws-parallelcluster-awsbatch/recipes/awsbatch_virtualenv.rb

@@ -25,6 +25,27 @@
 activate_virtual_env virtualenv_name do
   pyenv_path virtualenv_path
   python_version python_version
+  not_if { ::File.exist?("#{virtualenv_path}/bin/activate") }
+end
+
+remote_file "#{node['cluster']['base_dir']}/awsbatch-dependencies.tgz" do
+  source "#{node['cluster']['artifacts_s3_url']}/dependencies/PyPi/#{node['kernel']['machine']}/awsbatch-dependencies.tgz"
+  mode '0644'
+  retries 3
+  retry_delay 5
+  action :create_if_missing
+end
+
+bash 'pip install' do
+  user 'root'
+  group 'root'
+  cwd "#{node['cluster']['base_dir']}"
+  code <<-REQ
+    set -e
+    tar xzf awsbatch-dependencies.tgz
+    cd awsbatch
+    #{virtualenv_path}/bin/pip install * -f ./ --no-index
+    REQ
 end
 
 node.default['cluster']['awsbatch_virtualenv_path'] = virtualenv_path

cookbooks/aws-parallelcluster-awsbatch/recipes/install.rb

@@ -44,32 +44,12 @@
       tar -xzf aws-parallelcluster.tgz --directory aws-parallelcluster-awsbatch-cli
       cd aws-parallelcluster-awsbatch-cli/*aws-parallelcluster-*
 
-      aws s3 cp #{node['cluster']['artifacts_build_url']}/PyPi/#{node['kernel']['machine']}/awsbatch-dependencies.tgz awsbatch-dependencies.tgz --region #{node['cluster']['region']}
-      tar xzf awsbatch-dependencies.tgz
-      cd awsbatch
-      #{node['cluster']['awsbatch_virtualenv_path']}/bin/pip install * -f ./ --no-index
-      cd ..
-
       #{node['cluster']['awsbatch_virtualenv_path']}/bin/pip install awsbatch-cli/
     CLI
   end
 else
   # Install aws-parallelcluster-awsbatch-cli package
-  bash "install aws-parallelcluster-awsbatch-cli" do
-    cwd Chef::Config[:file_cache_path]
-    code <<-CLI
-      set -e
-      package_url=#{node['cluster']['artifacts_build_url']}/awsbatch/aws-parallelcluster.tgz
-      aws s3 cp ${package_url} aws-parallelcluster.tgz --region #{node['cluster']['region']}
-      mkdir aws-parallelcluster-awsbatch-cli
-      tar -xzf aws-parallelcluster.tgz --directory aws-parallelcluster-awsbatch-cli
-      aws s3 cp #{node['cluster']['artifacts_build_url']}/PyPi/#{node['kernel']['machine']}/awsbatch-dependencies.tgz awsbatch-dependencies.tgz --region #{node['cluster']['region']}
-      tar xzf awsbatch-dependencies.tgz
-      cd awsbatch
-      #{node['cluster']['awsbatch_virtualenv_path']}/bin/pip install * -f ./ --no-index
-      cd ..
-      cd aws-parallelcluster-awsbatch-cli/*aws-parallelcluster-*
-      #{node['cluster']['awsbatch_virtualenv_path']}/bin/pip install awsbatch-cli/
-    CLI
+  execute "pip_install_parallelcluster_awsbatch_cli" do
+    command "#{node['cluster']['awsbatch_virtualenv_path']}/bin/pip install aws-parallelcluster-awsbatch-cli==#{node['cluster']['parallelcluster-awsbatch-cli-version']}"
   end
 end

cookbooks/aws-parallelcluster-computefleet/recipes/install/custom_parallelcluster_node.rb

@@ -20,6 +20,14 @@
 # TODO: once the pyenv Chef resource supports installing packages from a path (e.g. `pip install .`), convert the
 # bash block to a recipe that uses the pyenv resource.
 
+remote_file "#{Chef::Config[:file_cache_path]}/node-dependencies.tgz" do
+  source "#{node['cluster']['artifacts_s3_url']}/dependencies/PyPi/#{node['kernel']['machine']}/node-dependencies.tgz"
+  mode '0644'
+  retries 3
+  retry_delay 5
+  action :create_if_missing
+end
+
 bash "install custom aws-parallelcluster-node" do
   cwd Chef::Config[:file_cache_path]
   code <<-NODE
@@ -38,12 +46,6 @@
     mkdir aws-parallelcluster-custom-node
     tar -xzf aws-parallelcluster-node.tgz --directory aws-parallelcluster-custom-node
     cd aws-parallelcluster-custom-node/*aws-parallelcluster-node-*
-    aws s3 cp #{node['cluster']['artifacts_build_url']}/PyPi/#{node['kernel']['machine']}/node-dependencies.tgz node-dependencies.tgz --region #{node['cluster']['region']}
-    tar xzf node-dependencies.tgz
-    cd node
-    #{node_virtualenv_path}/bin/pip install * -f ./ --no-index
-    cd ..
-
     pip install .
     deactivate
   NODE

cookbooks/aws-parallelcluster-computefleet/recipes/install/parallelcluster_node.rb

@@ -36,27 +36,7 @@
 if is_custom_node?
   include_recipe 'aws-parallelcluster-computefleet::custom_parallelcluster_node'
 else
-  bash "install official aws-parallelcluster-node" do
-    cwd Chef::Config[:file_cache_path]
-    code <<-NODE
-    set -e
-    [[ ":$PATH:" != *":/usr/local/bin:"* ]] && PATH="/usr/local/bin:${PATH}"
-    echo "PATH is $PATH"
-    source #{node_virtualenv_path}/bin/activate
-    pip uninstall --yes aws-parallelcluster-node
-    node_url=#{node['cluster']['artifacts_build_url']}/node/aws-parallelcluster-node.tgz
-    aws s3 cp ${node_url} aws-parallelcluster-node.tgz --region #{node['cluster']['region']}
-    rm -fr aws-parallelcluster-node
-    mkdir aws-parallelcluster-node
-    tar -xzf aws-parallelcluster-node.tgz --directory aws-parallelcluster-node
-    aws s3 cp #{node['cluster']['artifacts_build_url']}/PyPi/#{node['kernel']['machine']}/node-dependencies.tgz node-dependencies.tgz --region #{node['cluster']['region']}
-    tar xzf node-dependencies.tgz
-    cd node
-    #{node_virtualenv_path}/bin/pip install * -f ./ --no-index
-    cd ..
-    cd aws-parallelcluster-node/*aws-parallelcluster-node-*
-    pip install .
-    deactivate
-  NODE
+  execute "install official aws-parallelcluster-node" do
+    command "#{virtualenv_path}/bin/pip install aws-parallelcluster-node==#{node['cluster']['parallelcluster-node-version']}"
   end
 end

cookbooks/aws-parallelcluster-computefleet/spec/unit/recipes/node_spec.rb

@@ -36,7 +36,7 @@
         end
 
         it 'installs official node package' do
-          is_expected.to run_bash('install official aws-parallelcluster-node')
+          is_expected.to run_execute('install official aws-parallelcluster-node')
         end
       end
 

cookbooks/aws-parallelcluster-platform/recipes/install/awscli.rb

@@ -35,5 +35,5 @@
 end
 
 bash 'install awscli' do
-  code "#{cookbook_virtualenv_path}/bin/python#{node['cluster']['python-major-minor-version']} #{file_cache_path}/awscli/awscli-bundle/install -i /usr/local/aws -b /usr/local/bin/aws"
+  code "#{cookbook_virtualenv_path}/bin/python #{file_cache_path}/awscli/awscli-bundle/install -i /usr/local/aws -b /usr/local/bin/aws"
 end

cookbooks/aws-parallelcluster-platform/recipes/install/cookbook_virtualenv.rb

@@ -24,11 +24,7 @@
 activate_virtual_env cookbook_virtualenv_name do
   pyenv_path cookbook_virtualenv_path
   python_version cookbook_python_version
-end
-
-cookbook_file "#{virtualenv_path}/requirements.txt" do
-  source "cookbook_virtualenv/requirements.txt"
-  mode '0755'
+  not_if { ::File.exist?("#{cookbook_virtualenv_path}/bin/activate") }
 end
 
 remote_file "#{node['cluster']['base_dir']}/cookbook-dependencies.tgz" do

cookbooks/aws-parallelcluster-platform/resources/arm_pl/partial/_arm_pl_common.rb

@@ -137,7 +137,7 @@
         tar -xf #{gcc_tarball}
         cd gcc-#{gcc_version}
         # Patch the download_prerequisites script to download over https and not ftp. This works better in China regions.
-        sed -i "s#ftp://gcc\.gnu\.org##{node['cluster']['artifacts_s3_url']}/dependencies/gcc/prerequisites#g" ./contrib/download_prerequisites
+        sed -i "s#ftp://gcc\.gnu\.org/pub/gcc/infrastructure##{node['cluster']['artifacts_s3_url']}/dependencies/gcc/prerequisites#g" ./contrib/download_prerequisites
         ./contrib/download_prerequisites
         mkdir build && cd build
         ../configure --prefix=/opt/arm/armpl/gcc/#{gcc_version} --disable-bootstrap --enable-checking=release --enable-languages=c,c++,fortran --disable-multilib

cookbooks/aws-parallelcluster-platform/resources/fabric_manager/fabric_manager_ubuntu20+.rb

@@ -24,7 +24,7 @@ def fabric_manager_package
 end
 
 def fabric_manager_version
-  "#{_nvidia_driver_version}*"
+  "#{_nvidia_driver_version}"
 end
 
 def platform

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

@@ -14,19 +14,21 @@
 
 action :install_package do
   # For ubuntu, CINC17 apt-package resources need full versions for `version`
-  execute "install_fabricmanager_for_ubuntu" do
-    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
+  remote_file "#{node['cluster']['sources_dir']}/#{fabric_manager_package}-#{fabric_manager_version}.deb" do
+    source "#{fabric_manager_url}"
+    mode '0644'
+    retries 3
+    retry_delay 5
+    action :create_if_missing
+  end
 
-    command "apt -y install #{fabric_manager_package}-#{fabric_manager_version}.deb "\
-            "&& apt-mark hold #{fabric_manager_package}"
+  bash "install_fabricmanager_for_ubuntu" do
+    user 'root'
+    cwd node['cluster']['sources_dir']
+    code <<-FABRIC_MANAGER
+    set -e
+    dpkg -i #{fabric_manager_package}-#{fabric_manager_version}.deb && apt-mark hold #{fabric_manager_package}
+    FABRIC_MANAGER
     retries 3
     retry_delay 5
   end
@@ -37,5 +39,5 @@ def arch_suffix
 end
 
 def fabric_manager_url
-  "#{node['cluster']['artifacts_build_url']}/nvidia_fabric/#{platform}/#{fabric_manager_package}_#{fabric_manager_version}-1_#{arch_suffix}.deb"
+  "#{node['cluster']['artifacts_s3_url']}/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

@@ -13,12 +13,20 @@
 # See the License for the specific language governing permissions and limitations under the License.
 
 action :install_package do
+  remote_file "#{node['cluster']['sources_dir']}/#{fabric_manager_package}-#{fabric_manager_version}.rpm" do
+    source "#{fabric_manager_url}"
+    mode '0644'
+    retries 3
+    retry_delay 5
+    action :create_if_missing
+  end
+
   package 'yum-plugin-versionlock'
   bash "Install #{fabric_manager_package}" do
     user 'root'
+    cwd node['cluster']['sources_dir']
     code <<-FABRIC_MANAGER_INSTALL
     set -e
-    aws s3 cp #{fabric_manager_url} #{fabric_manager_package}-#{fabric_manager_version}.rpm --region #{node['cluster']['region']}
     yum install -y #{fabric_manager_package}-#{fabric_manager_version}.rpm
     yum versionlock #{fabric_manager_package}
     FABRIC_MANAGER_INSTALL
@@ -32,5 +40,5 @@ def arch_suffix
 end
 
 def fabric_manager_url
-  "#{node['cluster']['artifacts_build_url']}/nvidia_fabric/#{platform}/#{fabric_manager_package}-#{fabric_manager_version}-1.#{arch_suffix}.rpm"
+  "#{node['cluster']['artifacts_s3_url']}/dependencies/nvidia_fabric/#{platform}/#{fabric_manager_package}-#{fabric_manager_version}-1.#{arch_suffix}.rpm"
 end

cookbooks/aws-parallelcluster-platform/resources/nvidia_dcgm/nvidia_dcgm_alinux2023.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/partial/_nvidia_dcgm_debian.rb

@@ -13,20 +13,28 @@
 # See the License for the specific language governing permissions and limitations under the License.
 
 action :install_package do
+  remote_file "#{node['cluster']['sources_dir']}/#{dcgm_package}-#{package_version}.deb" do
+    source "#{dcgm_url}"
+    mode '0644'
+    retries 3
+    retry_delay 5
+    action :create_if_missing
+  end
+
   bash "Install #{dcgm_package}" do
     user 'root'
+    cwd node['cluster']['sources_dir']
     code <<-DCGM_INSTALL
     set -e
-    aws s3 cp #{dcgm_url} #{dcgm_package}-#{package_version}.deb --region #{node['cluster']['region']}
-    apt -y install #{dcgm_package}-#{package_version}.deb
+    dpkg -i #{dcgm_package}-#{package_version}.deb
     DCGM_INSTALL
     retries 3
     retry_delay 5
   end
 end
 
 def dcgm_url
-  "#{node['cluster']['artifacts_build_url']}/nvidia_dcgm/#{platform}/#{dcgm_package}_#{package_version}_#{arch_suffix}.deb"
+  "#{node['cluster']['artifacts_s3_url']}/dependencies/nvidia_dcgm/#{platform}/#{dcgm_package}_#{package_version}_#{arch_suffix}.deb"
 end
 
 def dcgm_package

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

@@ -13,11 +13,19 @@
 # See the License for the specific language governing permissions and limitations under the License.
 
 action :install_package do
+  remote_file "#{node['cluster']['sources_dir']}/#{dcgm_package}-#{package_version}.rpm" do
+    source "#{dcgm_url}"
+    mode '0644'
+    retries 3
+    retry_delay 5
+    action :create_if_missing
+  end
+
   bash "Install #{dcgm_package}" do
     user 'root'
+    cwd node['cluster']['sources_dir']
     code <<-DCGM_INSTALL
     set -e
-    aws s3 cp #{dcgm_url} #{dcgm_package}-#{package_version}.rpm --region #{node['cluster']['region']}
     yum install -y #{dcgm_package}-#{package_version}.rpm
     DCGM_INSTALL
     retries 3
@@ -26,7 +34,7 @@
 end
 
 def dcgm_url
-  "#{node['cluster']['artifacts_build_url']}/nvidia_dcgm/#{platform}/#{dcgm_package}-#{package_version}-1-#{arch_suffix}.rpm"
+  "#{node['cluster']['artifacts_s3_url']}/dependencies/nvidia_dcgm/#{platform}/#{dcgm_package}-#{package_version}-1-#{arch_suffix}.rpm"
 end
 
 def dcgm_package

cookbooks/aws-parallelcluster-platform/resources/nvidia_driver/partial/_nvidia_driver_common.rb

@@ -126,4 +126,4 @@ def nvidia_kernel_module
   else
     "kernel-open"
   end
-end
+end

cookbooks/aws-parallelcluster-platform/spec/unit/recipes/awscli_spec.rb

@@ -31,7 +31,7 @@
 
         it 'installs awscli into cookbook virtualev path' do
           is_expected.to run_bash('install awscli')
-            .with_code "#{cookbook_virtualenv_path}/bin/python#{node['cluster']['python-major-minor-version']} #{file_cache_path}/awscli/awscli-bundle/install -i /usr/local/aws -b /usr/local/bin/aws"
+            .with_code "#{cookbook_virtualenv_path}/bin/python #{file_cache_path}/awscli/awscli-bundle/install -i /usr/local/aws -b /usr/local/bin/aws"
         end
       end
 

cookbooks/aws-parallelcluster-platform/spec/unit/recipes/cookbook_virtualenv_spec.rb

@@ -33,13 +33,6 @@
           is_expected.to write_node_attributes('dump node attributes')
         end
 
-        it 'copies requirements file' do
-          is_expected.to create_cookbook_file("#{virtualenv_path}/requirements.txt").with(
-            source: "cookbook_virtualenv/requirements.txt",
-            mode: '0755'
-          )
-        end
-
         it 'installs python packages' do
           is_expected.to run_bash("pip install").with(
             user: 'root',

cookbooks/aws-parallelcluster-platform/spec/unit/resources/fabric_manager_spec.rb

@@ -169,7 +169,7 @@ def self.configure(chef_run)
   for_all_oses do |platform, version|
     context "on #{platform}#{version}" do
       cached(:fabric_manager_package) { platform == 'ubuntu' ? 'nvidia-fabricmanager-535' : 'nvidia-fabric-manager' }
-      cached(:fabric_manager_version) { platform == 'ubuntu' ? "#{nvidia_driver_version}*" : nvidia_driver_version }
+      cached(:fabric_manager_version) { platform == 'ubuntu' ? "#{nvidia_driver_version}" : nvidia_driver_version }
 
       context 'when fabric manager is to install' do
         cached(:chef_run) do
@@ -193,10 +193,10 @@ def self.configure(chef_run)
 
         if platform == 'ubuntu'
           it 'installs fabric manager for ubuntu' do
-            is_expected.to run_execute('install_fabricmanager_for_ubuntu')
+            is_expected.to run_bash('install_fabricmanager_for_ubuntu')
               .with_retries(3)
               .with_retry_delay(5)
-              .with_command("apt -y install #{fabric_manager_package}-#{fabric_manager_version}.deb && apt-mark hold #{fabric_manager_package}")
+              .with_code(/dpkg -i #{fabric_manager_package}-#{fabric_manager_version}.deb && apt-mark hold #{fabric_manager_package}/)
           end
         else
           it 'installs yum-plugin-versionlock' do
@@ -222,7 +222,7 @@ def self.configure(chef_run)
   for_all_oses do |platform, version|
     context "on #{platform}#{version}" do
       cached(:fabric_manager_package) { platform == 'ubuntu' ? 'nvidia-fabricmanager-535' : 'nvidia-fabric-manager' }
-      cached(:fabric_manager_version) { platform == 'ubuntu' ? "#{nvidia_driver_version}*" : nvidia_driver_version }
+      cached(:fabric_manager_version) { platform == 'ubuntu' ? "#{nvidia_driver_version}" : nvidia_driver_version }
 
       context('when nvswithes are > 1') do
         cached(:chef_run) do

cookbooks/aws-parallelcluster-shared/attributes/environment.rb

@@ -5,6 +5,5 @@
 
 # URL for ParallelCluster Artifacts stored in public S3 buckets
 # ['cluster']['region'] will need to be defined by image_dna.json during AMI build.
-default['cluster']['base_build_url'] = "s3://#{node['cluster']['region']}-aws-parallelcluster"
-default['cluster']['artifacts_build_url'] = "#{node['cluster']['base_build_url']}/archives/dependencies"
-default['cluster']['artifacts_s3_url'] = "https://#{node['cluster']['region']}-aws-parallelcluster.s3.#{node['cluster']['region']}.#{node['cluster']['aws_domain']}/archives"
+default['cluster']['artifacts_build_url'] = "s3://#{node['cluster']['region']}-aws-parallelcluster/archives/dependencies"
+default['cluster']['artifacts_s3_url'] = "https://#{node['cluster']['region']}-aws-parallelcluster.s3.#{node['cluster']['region']}.#{node['cluster']['aws_domain']}/archives"

cookbooks/aws-parallelcluster-shared/resources/install_pyenv.rb

@@ -30,7 +30,7 @@
   end
 
   remote_file "#{prefix}/Python-#{python_version}.tgz" do
-    source "#{python_url}"
+    source python_url
     mode '0644'
     retries 3
     retry_delay 5
@@ -52,5 +52,4 @@
     make install
     VENV
   end
-
-end
+end