Prevent mountpoint collision between ephemeral drive and shared drives (

#1589) Disable ephemeral mount if mountpoint clashes with shared file system mountpoint Set up kitchen tests for mountpoint collision for ephemeral and shared drives Signed-off-by: Jacopo De Amicis <[email protected]>
aws · Nov 10, 2022 · fe7d847 · fe7d847
1 parent 6e5c94c
commit fe7d847
Show file tree

Hide file tree

Showing 2 changed files with 40 additions and 17 deletions.
diff --git a/cookbooks/aws-parallelcluster-config/recipes/base.rb b/cookbooks/aws-parallelcluster-config/recipes/base.rb
@@ -27,16 +27,23 @@
 
 include_recipe 'aws-parallelcluster-config::nfs' unless virtualized?
 
-service "setup-ephemeral" do
-  supports restart: false
-  action :enable
-end
+# Mount the ephemeral drive unless there is a mountpoint collision with shared drives
+shared_dir_array = node['cluster']['ebs_shared_dirs'].split(',') + \
+                   node['cluster']['efs_shared_dirs'].split(',') + \
+                   node['cluster']['fsx_shared_dirs'].split(',') + \
+                   [ node['cluster']['raid_shared_dir'] ]
+unless shared_dir_array.include? node['cluster']['ephemeral_dir']
+  service "setup-ephemeral" do
+    supports restart: false
+    action :enable
+  end
 
-# Execution timeout 3600 seconds
-unless virtualized?
-  execute "Setup of ephemeral drivers" do
-    user "root"
-    command "/usr/local/sbin/setup-ephemeral-drives.sh"
+  # Execution timeout 3600 seconds
+  unless virtualized?
+    execute "Setup of ephemeral drives" do
+      user "root"
+      command "/usr/local/sbin/setup-ephemeral-drives.sh"
+    end
   end
 end
 

diff --git a/cookbooks/aws-parallelcluster-test/recipes/tests.rb b/cookbooks/aws-parallelcluster-test/recipes/tests.rb
@@ -334,9 +334,25 @@ module load intelmpi && mpirun --help | grep '#{node['cluster']['intelmpi']['kit
 ###################
 # instance store
 ###################
-bash 'test instance store' do
-  cwd Chef::Config[:file_cache_path]
-  code <<-EPHEMERAL
+
+ebs_shared_dirs_array = node['cluster']['ebs_shared_dirs'].split(',')
+
+if ebs_shared_dirs_array.include? node['cluster']['ephemeral_dir']
+  # In this case the ephemeral storage should not be mounted because the mountpoint
+  # clashes with the mountpoint coming from t
+  bash 'test instance store mountpoint collision' do
+    cwd Chef::Config[:file_cache_path]
+    user node['cluster']['cluster_user']
+    code <<-COLLISION
+      systemctl show setup-ephemeral.service -p ActiveState | grep "=inactive"
+      systemctl show setup-ephemeral.service -p UnitFileState | grep "=disabled"
+    COLLISION
+  end
+else
+  bash 'test instance store' do
+    cwd Chef::Config[:file_cache_path]
+    user node['cluster']['cluster_user']
+    code <<-EPHEMERAL
       set -xe
       EPHEMERAL_DIR="#{node['cluster']['ephemeral_dir']}"
 
@@ -389,11 +405,11 @@ module load intelmpi && mpirun --help | grep '#{node['cluster']['intelmpi']['kit
         mkdir -p ${EPHEMERAL_DIR}/test_dir
         touch ${EPHEMERAL_DIR}/test_dir/test_file
       fi
-  EPHEMERAL
-  user node['cluster']['cluster_user']
-end
-execute 'check setup-ephemeral service is enabled' do
-  command "systemctl is-enabled setup-ephemeral"
+    EPHEMERAL
+  end
+  execute 'check setup-ephemeral service is enabled' do
+    command "systemctl is-enabled setup-ephemeral"
+  end
 end
 
 ###################