Upgrade dependencies

Signed-off-by: Hanwen <[email protected]>

Author: hanwen-cluster

cookbooks/aws-parallelcluster-awsbatch/test/controls/awsbatch_virtualenv_spec.rb

@@ -13,7 +13,7 @@
 pyenv_dir = "#{base_dir}/pyenv"
 
 control 'tag:install_awsbatch_virtualenv_created' do
-  python_version = os_properties.alinux2? ? '3.9.20' : '3.12.8'
+  python_version = os_properties.alinux2? ? '3.9.20' : '3.12.11'
   title "awsbatch virtualenv should be created on #{python_version}"
   only_if { !os_properties.redhat? }
 

cookbooks/aws-parallelcluster-computefleet/kitchen.computefleet-config.yml

@@ -31,7 +31,7 @@ suites:
     attributes:
       cluster:
         custom_node_package: https://github.com/aws/aws-parallelcluster-node/archive/develop.tar.gz
-        python-version: 3.12.8
+        python-version: 3.12.11
         node_virtualenv_path: /opt/parallelcluster/pyenv/versions/node_virtualenv
   - name: fleet_status
     run_list:

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

@@ -70,8 +70,8 @@
 
 default['cluster']['head_node_private_ip'] = nil
 
-default['cluster']['efa']['version'] = '1.41.0'
-default['cluster']['efa']['sha256'] = '3506354cdfbe31ff552fe75f5d0d9bb7efd29cf79bd99457347d29c751c38f9f'
+default['cluster']['efa']['version'] = '1.42.0'
+default['cluster']['efa']['sha256'] = '4114fe612905ee05083ae5cb391a00a012510f3abfecc642d86c9a5ae4be9008'
 
 default['cluster']['efs']['version'] = '2.3.1'
 default['cluster']['efs']['sha256'] = 'ced12f82e76f9740476b63f30c49bd76cc00b6375e12a9f5f7ba852635c49e15'

cookbooks/aws-parallelcluster-environment/spec/unit/resources/efa_spec.rb

@@ -2,8 +2,8 @@
 
 # parallelcluster default source dir defined in attributes
 source_dir = '/opt/parallelcluster/sources'
-efa_version = '1.41.0'
-efa_checksum = '3506354cdfbe31ff552fe75f5d0d9bb7efd29cf79bd99457347d29c751c38f9f'
+efa_version = '1.42.0'
+efa_checksum = '4114fe612905ee05083ae5cb391a00a012510f3abfecc642d86c9a5ae4be9008'
 
 class ConvergeEfa
   def self.setup(chef_run, efa_version: nil, efa_checksum: nil)

cookbooks/aws-parallelcluster-environment/test/controls/cfn_bootstrap_spec.rb

@@ -13,7 +13,7 @@
 pyenv_dir = "#{base_dir}/pyenv"
 
 control 'tag:install_cfnbootstrap_virtualenv_created' do
-  cfn_python_version = os_properties.alinux2? ? '3.9.20' : '3.12.8'
+  cfn_python_version = os_properties.alinux2? ? '3.9.20' : '3.12.11'
   title "cfnbootstrap virtualenv should be created on #{cfn_python_version}"
   only_if { !os_properties.redhat_on_docker? }
 

cookbooks/aws-parallelcluster-platform/attributes/platform.rb

@@ -16,10 +16,11 @@
 
 # NVidia
 default['cluster']['nvidia']['enabled'] = 'no'
-default['cluster']['nvidia']['driver_version'] = '570.86.15'
-default['cluster']['nvidia']['dcgm_version'] = '3.3.6'
+default['cluster']['nvidia']['driver_version'] = '570.172.08'
+default['cluster']['nvidia']['dcgm_version'] = '4-cuda12-4.2.3-2.'
 if platform?('amazon') && node['platform_version'] == "2"
   default['cluster']['nvidia']['driver_version'] = '550.127.08'
+  default['cluster']['nvidia']['dcgm_version'] = '3.3.6-1-'
 end
 
 # DCV

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

@@ -20,9 +20,9 @@
 # Cuda installer from https://developer.nvidia.com/cuda-toolkit-archive
 # Cuda installer naming: cuda_11.8.0_520.61.05_linux
 cuda_version = '12.8'
-cuda_patch = '0'
+cuda_patch = '1'
 cuda_complete_version = "#{cuda_version}.#{cuda_patch}"
-cuda_version_suffix = '570.86.10'
+cuda_version_suffix = '570.124.06'
 cuda_samples_version = '12.8'
 if platform?('amazon') && node['platform_version'] == "2"
   cuda_version = '12.4'

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

@@ -16,7 +16,7 @@
 # limitations under the License.
 
 intelmpi_supported = !arm_instance?
-intelmpi_version = '2021.13'
+intelmpi_version = '2021.16'
 
 node.default['conditions']['intel_mpi_supported'] = intelmpi_supported
 node.default['cluster']['intelmpi']['version'] = intelmpi_version
@@ -25,9 +25,9 @@
 
 return unless intelmpi_supported
 
-intelmpi_full_version = "#{intelmpi_version}.1.769"
+intelmpi_full_version = "#{intelmpi_version}.0.443"
 intelmpi_installation_path = "/opt/intel/mpi/#{intelmpi_version}"
-intelmpi_installer = "l_mpi_oneapi_p_#{intelmpi_full_version}_offline.sh"
+intelmpi_installer = "intel-mpi-#{intelmpi_full_version}_offline.sh"
 intelmpi_installer_path = "#{node['cluster']['sources_dir']}/#{intelmpi_installer}"
 intelmpi_installer_url = "#{node['cluster']['artifacts_s3_url']}/impi/#{intelmpi_installer}"
 intelmpi_qt_version = '6.5.3'

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

@@ -34,7 +34,7 @@
 end
 
 def dcgm_url
-  "#{node['cluster']['artifacts_s3_url']}/dependencies/nvidia_dcgm/#{platform}/#{dcgm_package}-#{package_version}-1-#{arch_suffix}.rpm"
+  "#{node['cluster']['artifacts_s3_url']}/dependencies/nvidia_dcgm/#{platform}/#{dcgm_package}-#{package_version}#{arch_suffix}.rpm"
 end
 
 def dcgm_package

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

@@ -2,9 +2,9 @@
 
 describe 'aws-parallelcluster-platform::cuda' do
   cached(:cuda_version) { '12.8' }
-  cached(:cuda_patch) { '0' }
+  cached(:cuda_patch) { '1' }
   cached(:cuda_complete_version) { "#{cuda_version}.#{cuda_patch}" }
-  cached(:cuda_version_suffix) { '570.86.10' }
+  cached(:cuda_version_suffix) { '570.124.06' }
 
   context 'when nvidia not enabled' do
     cached(:chef_run) do