Upgrade EFA to version 1.14.1

Starting from EFA 1.14.0, GDR support is enabled by default. Therefore, this commit also removes the logic to reinstall EFA if GDR is enabled.

This commit is cherry picked from 82154c9 and 9ebd58e

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

Author: hanwen-cluster

CHANGELOG.md

@@ -17,8 +17,6 @@ This file is used to list changes made in each version of the AWS ParallelCluste
 - Do not strip `-` from compute resource name when configuring Slurm nodes.
 - Upgrade Slurm to version 21.08.4.
 
-**BUG FIXES**
-- Fix issue that is preventing cluster names to start with `parallelcluster-` prefix.
 
 3.0.2
 ------
@@ -34,6 +32,9 @@ This file is used to list changes made in each version of the AWS ParallelCluste
   - Libfabric: ``libfabric-1.13.2``
   - Open MPI: ``openmpi40-aws-4.1.1-2``
 
+**BUG FIXES**
+- Fix issue that is preventing cluster names to start with `parallelcluster-` prefix.
+
 3.0.1
 ------
 

attributes/default.rb

@@ -187,9 +187,8 @@
 )
 
 # EFA
-default['cluster']['efa']['installer_version'] = '1.13.0'
+default['cluster']['efa']['installer_version'] = '1.14.1'
 default['cluster']['efa']['installer_url'] = "https://efa-installer.amazonaws.com/aws-efa-installer-#{node['cluster']['efa']['installer_version']}.tar.gz"
-default['cluster']['enable_efa_gdr'] = "no"
 default['cluster']['efa']['unsupported_aarch64_oses'] = %w(centos7)
 
 # NICE DCV

cookbooks/aws-parallelcluster-config/recipes/efa.rb

@@ -15,9 +15,6 @@
 # OR CONDITIONS OF ANY KIND, express or implied. See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Installation recipe must be re-executed at runtime to enable GDR
-include_recipe "aws-parallelcluster-install::efa"
-
 if platform?('ubuntu') && node['cluster']['enable_efa'] == 'compute' && node['cluster']['node_type'] == 'ComputeFleet'
   # Disabling ptrace protection is needed for EFA in order to use SHA transfer for intra-node communication.
   sysctl 'kernel.yama.ptrace_scope' do

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

@@ -19,7 +19,7 @@
 efa_installed = efa_installed?
 
 if efa_installed && !::File.exist?(efa_tarball)
-  Chef::Log.warn("Existing EFA version differs from the one shipped with ParallelCluster. Skipping ParallelCluster EFA installation and configuration. enable_gdr option will be ignored.")
+  Chef::Log.warn("Existing EFA version differs from the one shipped with ParallelCluster. Skipping ParallelCluster EFA installation and configuration.")
   return
 end
 
@@ -50,8 +50,6 @@
 installer_options = "-y"
 # skip efa-kmod installation on not supported platforms
 installer_options += " -k" unless node['conditions']['efa_supported']
-# enable gpudirect support
-installer_options += " -g" if efa_gdr_enabled?
 
 bash "install efa" do
   cwd node['cluster']['sources_dir']
@@ -62,5 +60,5 @@
     ./efa_installer.sh #{installer_options}
     rm -rf #{node['cluster']['sources_dir']}/aws-efa-installer
   EFAINSTALL
-  not_if { efa_installed && !efa_gdr_enabled? }
+  not_if { efa_installed }
 end

cookbooks/aws-parallelcluster-test/recipes/tests.rb

@@ -222,12 +222,34 @@ module load intelmpi && mpirun --help | grep '#{node['cluster']['intelmpi']['kit
 end
 
 ###################
-# EFA - GDR (GPUDirect RDMA)
+# EFA
 ###################
-if node['conditions']['efa_supported'] && efa_gdr_enabled?
-  execute 'check efa gdr installed' do
-    command "modinfo efa | grep 'gdr:\ *Y'"
-    user node['cluster']['cluster_user']
+if node['conditions']['efa_supported']
+  if node['cluster']['os'].end_with?("-custom")
+    # only check EFA is installed because when found in the base AMI we skip installation
+    bash 'check efa installed' do
+      cwd Chef::Config[:file_cache_path]
+      code <<-EFA
+        set -ex
+        modinfo efa
+        cat /opt/amazon/efa_installed_packages
+      EFA
+    end
+  else
+    # check EFA is installed and the version is expected
+    bash 'check correct version of efa installed' do
+      cwd Chef::Config[:file_cache_path]
+      code <<-EFA
+        set -ex
+        modinfo efa
+        grep "EFA installer version: #{node['cluster']['efa']['installer_version']}" /opt/amazon/efa_installed_packages
+      EFA
+    end
+    # GDR (GPUDirect RDMA)
+    execute 'check efa gdr installed' do
+      command "modinfo efa | grep 'gdr:\ *Y'"
+      user node['cluster']['cluster_user']
+    end
   end
 end
 

libraries/helpers.rb

@@ -396,17 +396,6 @@ def get_nvswitches
   nvswitch_check.stdout.strip.to_i
 end
 
-# Check if EFA GDR is enabled (and supported) on this instance
-def efa_gdr_enabled?
-  config_value = node['cluster']['enable_efa_gdr']
-  enabling_value = if node['cluster']['node_type'] == "ComputeFleet"
-                     "compute"
-                   else
-                     "head_node"
-                   end
-  (config_value == enabling_value || config_value == "cluster") && graphic_instance?
-end
-
 # Alinux OSs currently not correctly supported by NFS cookbook
 # Overwriting templates for node['nfs']['config']['server_template'] used by NFS cookbook for these OSs
 # When running, NFS cookbook will use nfs.conf.erb templates provided in this cookbook to generate server_template