Avoid /opt/intel when backing up, restoring, and mounting fs mount po…

…ints on arm instances (#2718)
aws · May 2, 2024 · 8268835 · 8268835
1 parent caff1c3
commit 8268835
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 6 deletions.
diff --git a/cookbooks/aws-parallelcluster-computefleet/files/clusterstatusmgtd/clusterstatusmgtd.py b/cookbooks/aws-parallelcluster-computefleet/files/clusterstatusmgtd/clusterstatusmgtd.py
@@ -156,11 +156,11 @@ def __str__(self):  # noqa: D105
     @staticmethod
     def _transform_compute_fleet_data(compute_fleet_data):
         try:
-            compute_fleet_data[
-                ComputeFleetStatusManager.COMPUTE_FLEET_STATUS_ATTRIBUTE
-            ] = ComputeFleetStatus.EVENT_HANDLER_STATUS_MAPPING.value.get(
-                compute_fleet_data.get(ComputeFleetStatusManager.COMPUTE_FLEET_STATUS_ATTRIBUTE),
-                str(ComputeFleetStatus.UNKNOWN),
+            compute_fleet_data[ComputeFleetStatusManager.COMPUTE_FLEET_STATUS_ATTRIBUTE] = (
+                ComputeFleetStatus.EVENT_HANDLER_STATUS_MAPPING.value.get(
+                    compute_fleet_data.get(ComputeFleetStatusManager.COMPUTE_FLEET_STATUS_ATTRIBUTE),
+                    str(ComputeFleetStatus.UNKNOWN),
+                )
             )
             return compute_fleet_data
         except AttributeError as e:

diff --git a/cookbooks/aws-parallelcluster-environment/attributes/environment.rb b/cookbooks/aws-parallelcluster-environment/attributes/environment.rb
@@ -58,7 +58,12 @@
 # Since this is a shared directory, it needs to be defined here first instead of in the dependent cookbook for slurm
 default['cluster']['slurm']['install_dir'] = '/opt/slurm'
 
-default['cluster']['internal_shared_dirs'] = [node['cluster']['shared_dir'], node['cluster']['shared_dir_login_nodes'], node['cluster']['slurm']['install_dir'], "/opt/intel"]
+default['cluster']['internal_shared_dirs'] = if x86_instance?
+                                               [node['cluster']['shared_dir'], node['cluster']['shared_dir_login_nodes'], node['cluster']['slurm']['install_dir'], "/opt/intel"]
+                                             else
+                                               [node['cluster']['shared_dir'], node['cluster']['shared_dir_login_nodes'], node['cluster']['slurm']['install_dir']]
+                                             end
+
 default['cluster']['internal_initial_shared_dir'] = "#{node['cluster']['base_dir']}/init_shared"
 
 default['cluster']['head_node_private_ip'] = nil

diff --git a/cookbooks/aws-parallelcluster-environment/recipes/init/backup_internal_use_shared_data.rb b/cookbooks/aws-parallelcluster-environment/recipes/init/backup_internal_use_shared_data.rb
@@ -19,6 +19,7 @@
   # This is necessary to preserve any data in these directories that was
   # generated during the build of ParallelCluster AMIs after converting to
   # shared storage
+  Chef::Log.info("Backup internal dirs #{node['cluster']['internal_shared_dirs']}")
   node['cluster']['internal_shared_dirs'].each do |dir|
     bash "Backup #{dir}" do
       user 'root'