Merge branch 'develop' into develop

hgreebe · Nov 19, 2023 · 4418311 · 4418311
2 parents f65fba9 + ba70d62
commit 4418311
Show file tree

Hide file tree

Showing 11 changed files with 37 additions and 12 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -27,7 +27,7 @@ This file is used to list changes made in each version of the AWS ParallelCluste
 - Add support for Rocky Linux 8.
 - Add support for EC2 Capacity Blocks for ML.
 - Add support for `Scheduling/SlurmSettings/Database/DatabaseName` parameter to render `StorageLoc` in the slurmdbd configuration generated by ParallelCluster.
-- Add the option to use EFS storage instead of NFS exports from the head node root volume for intra-cluster shared ParallelCluster, Intel, Slurm, and login node data.
+- Add the option to use EFS storage instead of NFS exports from the head node root volume for intra-cluster shared ParallelCluster, Intel, Slurm, login node, and /home data.
 - Allow for mounting `home` as an EFS or FSx external shared storage via the `SharedStorage` section of the config file.
 
 **CHANGES**

diff --git a/cookbooks/aws-parallelcluster-environment/kitchen.environment-config.yml b/cookbooks/aws-parallelcluster-environment/kitchen.environment-config.yml
@@ -642,7 +642,7 @@ suites:
       cluster:
         node_type: 'HeadNode'
         scheduler: 'slurm'
-        internal_shared_storage_type: 'efs'
+        shared_storage_type: 'efs'
         efs_shared_dirs: '/opt/parallelcluster/init_shared'
         efs_fs_ids: 'fs-03ad31942a4205839' # Existing FS, needs to be set when running the test
         ebs_shared_dirs: ''
@@ -661,7 +661,7 @@ suites:
       cluster:
         node_type: 'ComputeFleet'
         scheduler: 'slurm'
-        internal_shared_storage_type: 'efs'
+        shared_storage_type: 'efs'
         efs_shared_dirs: '/opt/parallelcluster/init_shared'
         efs_fs_ids: 'fs-03ad31942a4205839' # Existing FS, needs to be set when running the test
         ebs_shared_dirs: ''
@@ -680,7 +680,7 @@ suites:
       cluster:
         node_type: 'LoginNode'
         scheduler: 'slurm'
-        internal_shared_storage_type: 'efs'
+        shared_storage_type: 'efs'
         efs_shared_dirs: '/opt/parallelcluster/init_shared'
         efs_fs_ids: 'fs-03ad31942a4205839' # Existing FS, needs to be set when running the test
         ebs_shared_dirs: ''

diff --git a/cookbooks/aws-parallelcluster-environment/recipes/config.rb b/cookbooks/aws-parallelcluster-environment/recipes/config.rb
@@ -23,7 +23,7 @@
 # Export the home dir from the head node when using ebs
 include_recipe 'aws-parallelcluster-environment::export_home'
 
-if node['cluster']['internal_shared_storage_type'] == 'ebs'
+if node['cluster']['shared_storage_type'] == 'ebs'
   # Export internal use dirs from the head node
   include_recipe 'aws-parallelcluster-environment::export_internal_use_ebs'
   # Mount intel on compute and login nodes

diff --git a/cookbooks/aws-parallelcluster-environment/recipes/config/export_home.rb b/cookbooks/aws-parallelcluster-environment/recipes/config/export_home.rb
@@ -17,7 +17,7 @@
 # Check if home is a shared filesystem and return if it is because there is nothing to export
 shared_storage = node['cluster']['efs_shared_dirs'].split(',') + node['cluster']['fsx_shared_dirs'].split(',') +
                  node['cluster']['ebs_shared_dirs'].split(',') + node['cluster']['raid_shared_dir'].split(',')
-return if shared_storage.include?('/home') || shared_storage.include?('home')
+return if shared_storage.include?('/home') || shared_storage.include?('home') || node['cluster']['shared_storage_type'] == 'efs'
 
 case node['cluster']['node_type']
 when 'HeadNode'

diff --git a/cookbooks/aws-parallelcluster-environment/recipes/config/mount_home.rb b/cookbooks/aws-parallelcluster-environment/recipes/config/mount_home.rb
@@ -22,6 +22,7 @@
 
 # Check if home is a shared filesystem
 shared_home = 'none'
+shared_home = 'internal' if node['cluster']['shared_storage_type'] == 'efs'
 shared_storage.each do |type, dirs|
   next unless dirs.include?('/home') || dirs.include?('home')
   shared_home = type
@@ -51,6 +52,19 @@
   include_recipe "aws-parallelcluster-environment::update_fs_mapping"
   include_recipe "aws-parallelcluster-environment::backup_home_shared_data"
   case shared_home
+  when 'internal'
+    shared_storage['efs'].each_with_index do |dir, index|
+      next unless dir == node['cluster']['internal_initial_shared_dir']
+      efs "mount internal shared efs home" do
+        shared_dir_array ['/home']
+        efs_fs_id_array [node['cluster']['efs_fs_ids'].split(',')[index]]
+        efs_encryption_in_transit_array [node['cluster']['efs_encryption_in_transits'].split(',')[index]]
+        efs_iam_authorization_array [node['cluster']['efs_iam_authorizations'].split(',')[index]]
+        efs_mount_point_array ['/home']
+        action :mount
+      end
+      break
+    end
   when 'efs'
     shared_storage['efs'].each_with_index do |dir, index|
       next unless dir == "/home" || dir == 'home'

diff --git a/cookbooks/aws-parallelcluster-environment/recipes/init.rb b/cookbooks/aws-parallelcluster-environment/recipes/init.rb
@@ -17,13 +17,13 @@
   action :configure
 end
 
-case node['cluster']['internal_shared_storage_type']
+case node['cluster']['shared_storage_type']
 when 'efs'
   include_recipe "aws-parallelcluster-environment::mount_internal_use_efs"
 when 'ebs'
   include_recipe "aws-parallelcluster-environment::mount_internal_use_ebs"
 else
-  raise "internal_shared_storage_type must be ebs or efs"
+  raise "shared_storage_type must be ebs or efs"
 end
 
 # Mount the home directory to all nodes if it is shared, otherwise mount the NFS share to compute and login nodes

diff --git a/cookbooks/aws-parallelcluster-environment/recipes/init/mount_internal_use_efs.rb b/cookbooks/aws-parallelcluster-environment/recipes/init/mount_internal_use_efs.rb
@@ -57,6 +57,17 @@
     end
   end unless initial_shared_dir_array.empty?
 
+  # Add the mount point for home.  If users decide to use a shared home via the managed EFS
+  # either on creation or update then this directory will be needed.  If users don't need it, it will stay empty
+  # on the managed FS and be invisible to users
+  directory "#{node['cluster']['internal_initial_shared_dir']}/home" do
+    user 'root'
+    group 'root'
+    mode '0755'
+    action :create
+    recursive true
+  end
+
   # Unmount the root of the EFS after creating the shared directories
   # TODO this doesn't seem to unmount the EFS
   efs "unmount internal efs" do

diff --git a/cookbooks/aws-parallelcluster-slurm/kitchen.slurm-config.yml b/cookbooks/aws-parallelcluster-slurm/kitchen.slurm-config.yml
@@ -176,7 +176,7 @@ suites:
       cluster:
         node_type: ComputeFleet
         head_node_private_ip: '127.0.0.1'
-        internal_shared_storage_type: ebs
+        shared_storage_type: ebs
   - name: config_head_node_munge
     run_list:
     - recipe[aws-parallelcluster-tests::setup]

diff --git a/cookbooks/aws-parallelcluster-slurm/recipes/config/config_head_node.rb b/cookbooks/aws-parallelcluster-slurm/recipes/config/config_head_node.rb
@@ -42,7 +42,7 @@
   network get_vpc_cidr_list
   writeable true
   options ['no_root_squash']
-  only_if { node['cluster']['internal_shared_storage_type'] == 'ebs' }
+  only_if { node['cluster']['shared_storage_type'] == 'ebs' }
 end unless on_docker?
 
 template "#{node['cluster']['slurm']['install_dir']}/etc/slurm.conf" do

diff --git a/cookbooks/aws-parallelcluster-slurm/recipes/config/mount_slurm_dir.rb b/cookbooks/aws-parallelcluster-slurm/recipes/config/mount_slurm_dir.rb
@@ -20,5 +20,5 @@
   action %i(mount enable)
   retries 10
   retry_delay 6
-  only_if { node['cluster']['internal_shared_storage_type'] == 'ebs' }
+  only_if { node['cluster']['shared_storage_type'] == 'ebs' }
 end
diff --git a/kitchen.validate-config.yml b/kitchen.validate-config.yml
@@ -26,7 +26,7 @@ _common_cluster_attributes: &_common_cluster_attributes
   enable_efa: 'efa'
   nvidia:
     enabled: <%= ENV['NVIDIA_ENABLED'] %>
-  internal_shared_storage_type: ebs
+  shared_storage_type: ebs
 
 _head_node_cluster_attributes: &_head_node_cluster_attributes
   << : *_common_cluster_attributes