Fix the way Pyxis and Enroot are configured.

  1. Pyxis is disabled by default. In particular, the SPANK config file and the Pyxis config file required to enable it are stored in `/opt/parallelcluster/examples` folder so that they are ineffective but can be used by the user to enable Pyxisby simply moving them to the expected location.

  2. Move Pyxis and Enroot configuration to build time (there was no reason to configure Pyxis and Enroot at runtime)

  3. Changed Pyxis runtime path to `/run/pyxis`. As per [documentation](https://github.com/NVIDIA/pyxis/wiki/Setup#slurm-plugstack-configuration) a tmpfs should be used. Asd a consequence, we needed to define a tmpfiles config to make sure that the dedicated folder is not deleted at boot time.

  4. Changed Enroot paths, following the suggestion in the [documentation](https://github.com/NVIDIA/pyxis/wiki/Setup#enroot-configuration-example)
    1. Using tmpfs storage for `ENROOT_RUNTIME_PATH` and `ENROOT_DATA_PATH`
    2. Using a persistent local storage for `ENROOT_CACHE_PATH` and `ENROOT_CONFIG_PATH`.

  5. *Minor*: Moved Pyxis attributes from platform cookbook to slurm cookbook because Pyxis is a SLURM plugin so it would be conceptually wrong to have its attributes defined in platform cookbook.

Signed-off-by: Giacomo Marciani <[email protected]>

Author: gmarciani

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

@@ -9,9 +9,10 @@
 # ArmPL
 default['conditions']['arm_pl_supported'] = arm_instance?
 
-# Enroot + Pyxis
+# Enroot
 default['cluster']['enroot']['version'] = '3.4.1'
-default['cluster']['pyxis']['version'] = '0.20.0'
+default['cluster']['enroot']['temporary_dir'] = '/run/enroot'
+default['cluster']['enroot']['persistent_dir'] = '/var/enroot'
 
 # NVidia
 default['cluster']['nvidia']['enabled'] = 'no'

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

@@ -26,6 +26,3 @@
 include_recipe 'aws-parallelcluster-platform::supervisord_config'
 fetch_config 'Fetch and load cluster configs'
 include_recipe 'aws-parallelcluster-platform::config_login' if node['cluster']['node_type'] == 'LoginNode'
-enroot 'Configure Enroot' do
-  action :configure
-end

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

@@ -21,6 +21,7 @@
 directory node['cluster']['license_dir']
 directory node['cluster']['configs_dir']
 directory node['cluster']['shared_dir']
+directory node['cluster']['examples_dir']
 directory node['cluster']['shared_dir_login_nodes']
 
 # Create ParallelCluster log folder

cookbooks/aws-parallelcluster-platform/resources/enroot/partial/_enroot_common.rb

@@ -1,6 +1,6 @@
 # frozen_string_literal: true
 #
-# Copyright:: 2013-2023 Amazon.com, Inc. or its affiliates. All Rights Reserved.
+# Copyright:: 2024 Amazon.com, Inc. or its affiliates. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License").
 # You may not use this file except in compliance with the License.
@@ -18,50 +18,37 @@
 action :setup do
   return if on_docker?
   action_install_package
-end
-
-action :configure do
-  return if on_docker?
-  return unless enroot_installed
 
-  cookbook_file "/tmp/enroot.template.conf" do
-    source 'enroot/enroot.template.conf'
-    cookbook 'aws-parallelcluster-platform'
+  template "/etc/enroot/enroot.conf" do
+    source 'enroot/enroot.conf.erb'
     owner 'root'
     group 'root'
-    mode '0755'
+    mode '0644'
     action :create_if_missing
   end
 
-  bash "Configure enroot" do
-    user 'root'
-    code <<-ENROOT_CONFIGURE
-      set -e
-      ENROOT_CONFIG_RELEASE=pyxis
-      SHARED_DIR=#{node['cluster']['shared_dir']}
-      NONROOT_USER=#{node['cluster']['cluster_user']}
-      mkdir -p ${SHARED_DIR}/enroot
-      chown ${NONROOT_USER} ${SHARED_DIR}/enroot
-      ENROOT_CACHE_PATH=${SHARED_DIR}/enroot envsubst < /tmp/enroot.template.conf > /tmp/enroot.conf
-      mv /tmp/enroot.conf /etc/enroot/enroot.conf
-      chmod 0644 /etc/enroot/enroot.conf
-
-      mkdir -p /tmp/enroot
-      chmod 1777 /tmp/enroot
-      mkdir -p /tmp/enroot/data
-      chmod 1777 /tmp/enroot/data
-
-      chmod 1777 ${SHARED_DIR}/enroot
+  directory node['cluster']['enroot']['persistent_dir'] do
+    owner 'root'
+    group 'root'
+    mode '1777'
+    recursive true
+  end
 
-      mkdir -p ${SHARED_DIR}/pyxis/
-      chown ${NONROOT_USER} ${SHARED_DIR}/pyxis/
-      sed -i '${s/$/ runtime_path=${SHARED_DIR}\\/pyxis/}' /opt/slurm/etc/plugstack.conf.d/pyxis.conf
-      SHARED_DIR=${SHARED_DIR} envsubst < /opt/slurm/etc/plugstack.conf.d/pyxis.conf > /opt/slurm/etc/plugstack.conf.d/pyxis.tmp.conf
-      mv /opt/slurm/etc/plugstack.conf.d/pyxis.tmp.conf /opt/slurm/etc/plugstack.conf.d/pyxis.conf
+  directory node['cluster']['enroot']['temporary_dir'] do
+    owner 'root'
+    group 'root'
+    mode '1777'
+    recursive true
+  end
 
-    ENROOT_CONFIGURE
-    retries 3
-    retry_delay 5
+  # We assume the Enroot temporary dir to be a temporary folder in /run.
+  # Folders in /run must be defined in /usr/lib/tmpfiles.d, otherwise they get
+  # deleted on node boot.
+  template "/usr/lib/tmpfiles.d/enroot.conf" do
+    source 'enroot/tmpfiles/enroot.conf.erb'
+    owner 'root'
+    group 'root'
+    mode '0644'
   end
 end
 

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

@@ -36,6 +36,10 @@
         is_expected.to create_directory(node['cluster']['shared_dir'])
       end
 
+      it 'creates examples directory' do
+        is_expected.to create_directory(node['cluster']['examples_dir'])
+      end
+
       it 'creates log directory' do
         is_expected.to create_directory(node['cluster']['log_base_dir']).with(
           owner: 'root',

cookbooks/aws-parallelcluster-platform/spec/unit/resources/enroot_spec.rb

@@ -9,17 +9,9 @@ def self.setup(chef_run)
       end
     end
   end
-
-  def self.configure(chef_run)
-    chef_run.converge_dsl('aws-parallelcluster-platform') do
-      enroot 'configure' do
-        action :configure
-      end
-    end
-  end
 end
 
-describe 'enroot:package_version' do
+describe 'aws-parallelcluster-platform::enroot:package_version' do
   for_all_oses do |platform, version|
     context "on #{platform}#{version}" do
       cached(:chef_run) do
@@ -39,7 +31,7 @@ def self.configure(chef_run)
   end
 end
 
-describe 'enroot:arch_suffix' do
+describe 'aws-parallelcluster-platform::enroot:arch_suffix' do
   for_all_oses do |platform, version|
     context "on #{platform}#{version} - arm" do
       cached(:chef_run) do
@@ -81,15 +73,73 @@ def self.configure(chef_run)
   end
 end
 
-describe 'enroot:setup' do
+describe 'aws-parallelcluster-platform::enroot:setup' do
   for_all_oses do |platform, version|
     context "on #{platform}#{version}" do
+      cached(:enroot_persistent_dir) { '/path/to/enroot/persistent/dir' }
+      cached(:enroot_temporary_dir) { '/path/to/enroot/temporary/dir' }
+
       let(:chef_run) do
         runner(platform: platform, version: version, step_into: ['enroot']) do |node|
           node.override['cluster']['enroot']['version'] = package_version
+          node.override['cluster']['enroot']['persistent_dir'] = enroot_persistent_dir
+          node.override['cluster']['enroot']['temporary_dir'] = enroot_temporary_dir
         end
       end
 
+      before do
+        ConvergeEnroot.setup(chef_run)
+      end
+
+      it 'creates the Enroot configuration' do
+        is_expected.to create_if_missing_template('/etc/enroot/enroot.conf').with(
+          source: 'enroot/enroot.conf.erb',
+          owner: 'root',
+          group: 'root',
+          mode: '0644'
+        )
+      end
+
+      # it 'the Enroot configuration has the correct content' do
+      #   is_expected.to render_file('/etc/enroot/enroot.conf')
+      #     .with_content("ENROOT_RUNTIME_PATH         #{enroot_temporary_dir}/runtime/user-$(id -u)")
+      #     .with_content("ENROOT_DATA_PATH            #{enroot_temporary_dir}/data/user-$(id -u)")
+      #     .with_content("ENROOT_CONFIG_PATH          #{enroot_persistent_dir}/config/user-$(id -u)")
+      #     .with_content("ENROOT_CACHE_PATH           #{enroot_persistent_dir}/cache/group-$(id -g)")
+      # end
+
+      it 'creates the Enroot persistent directory' do
+        is_expected.to create_directory(enroot_persistent_dir).with(
+          owner: 'root',
+          group: 'root',
+          mode: '1777',
+          recursive: true
+        )
+      end
+
+      it 'creates the Enroot temporary directory' do
+        is_expected.to create_directory(enroot_temporary_dir).with(
+          owner: 'root',
+          group: 'root',
+          mode: '1777',
+          recursive: true
+        )
+      end
+
+      it 'creates the Enroot tmpfiles.d configuration' do
+        is_expected.to create_template('/usr/lib/tmpfiles.d/enroot.conf').with(
+          source: 'enroot/tmpfiles/enroot.conf.erb',
+          owner: 'root',
+          group: 'root',
+          mode: '0644'
+        )
+      end
+
+      # it 'the Enroot tmpfile.d configuration has the correct content' do
+      #   is_expected.to render_file('/usr/lib/tmpfiles.d/enroot.conf')
+      #     .with_content("D #{enroot_temporary_dir} 0777 root root")
+      # end
+
       context 'when nvidia is enabled' do
         before do
           stubs_for_provider('enroot') do |resource|
@@ -128,44 +178,3 @@ def self.configure(chef_run)
     end
   end
 end
-
-describe 'enroot:configure' do
-  for_all_oses do |platform, version|
-    context "on #{platform}#{version}" do
-      let(:chef_run) do
-        runner(platform: platform, version: version, step_into: ['enroot'])
-      end
-
-      context 'when enroot is installed' do
-        before do
-          stubs_for_provider('enroot') do |resource|
-            allow(resource).to receive(:enroot_installed).and_return(true)
-          end
-          ConvergeEnroot.configure(chef_run)
-        end
-        it 'run configure enroot script' do
-          is_expected.to run_bash('Configure enroot')
-            .with(retries: 3)
-            .with(retry_delay: 5)
-            .with(user: 'root')
-        end
-      end
-
-      context 'when enroot is not installed' do
-        before do
-          stubs_for_provider('enroot') do |resource|
-            allow(resource).to receive(:enroot_installed).and_return(false)
-          end
-          ConvergeEnroot.configure(chef_run)
-        end
-
-        it 'does not run configure enroot script' do
-          is_expected.not_to run_bash('Configure enroot')
-            .with(retries: 3)
-            .with(retry_delay: 5)
-            .with(user: 'root')
-        end
-      end
-    end
-  end
-end

cookbooks/aws-parallelcluster-platform/files/enroot/enroot.template.conf renamed to cookbooks/aws-parallelcluster-platform/templates/enroot/enroot.conf.erb

@@ -1,9 +1,9 @@
 #ENROOT_LIBRARY_PATH        /usr/lib/enroot
 #ENROOT_SYSCONF_PATH        /etc/enroot
-ENROOT_RUNTIME_PATH        /tmp/enroot/user-$(id -u)
-ENROOT_CONFIG_PATH         ${ENROOT_CONFIG_PATH}
-ENROOT_CACHE_PATH          ${ENROOT_CACHE_PATH}
-ENROOT_DATA_PATH           /tmp/enroot/data/user-$(id -u)
+ENROOT_RUNTIME_PATH         <%= node['cluster']['enroot']['temporary_dir'] %>/runtime/user-$(id -u)
+ENROOT_DATA_PATH            <%= node['cluster']['enroot']['temporary_dir'] %>/data/user-$(id -u)
+ENROOT_CONFIG_PATH          <%= node['cluster']['enroot']['persistent_dir'] %>/config/user-$(id -u)
+ENROOT_CACHE_PATH           <%= node['cluster']['enroot']['persistent_dir'] %>/cache/group-$(id -g)
 #ENROOT_TEMP_PATH           ${TMPDIR:-/tmp}
 
 # Gzip program used to uncompress digest layers.
@@ -68,4 +68,4 @@ ENROOT_RESTRICT_DEV        no
 #all_proxy
 #no_proxy
 #http_proxy
-#https_proxy
+#https_proxy

cookbooks/aws-parallelcluster-platform/templates/enroot/tmpfiles/enroot.conf.erb

@@ -0,0 +1 @@
+D <%= node['cluster']['enroot']['temporary_dir'] %> 0777 root root

cookbooks/aws-parallelcluster-platform/test/controls/enroot_spec.rb

@@ -14,16 +14,36 @@
 
   expected_enroot_version = node['cluster']['enroot']['version']
 
-  describe "gdrcopy version is expected to be #{expected_enroot_version}" do
+  describe "enroot version is expected to be #{expected_enroot_version}" do
     subject { command('enroot version').stdout.strip() }
     it { should eq expected_enroot_version }
   end
+
+  persistent_dirs = %w(/etc/enroot /var/enroot)
+  persistent_dirs.each do |path|
+    describe directory(path) do
+      it { should exist }
+      its('owner') { should eq 'root' }
+      its('group') { should eq 'root' }
+      its('mode') { should cmp '01777' }
+    end
+  end
+
+  temporary_dirs = [ "/run/enroot" ]
+  temporary_dirs.each do |path|
+    describe directory(path) do
+      it { should exist }
+      its('owner') { should eq 'root' }
+      its('group') { should eq 'root' }
+      its('mode') { should cmp '01777' }
+    end
+  end
 end
 
 control 'tag:config_enroot_enabled_on_graphic_instances' do
   only_if { !os_properties.on_docker? && ['yes', true].include?(node['cluster']['nvidia']['enabled']) }
 
-  describe file("/opt/parallelcluster/shared/enroot") do
+  describe file("/var/enroot/cache-group-1000") do
     it { should exist }
     its('group') { should eq 'root' }
   end unless os_properties.redhat_on_docker?

cookbooks/aws-parallelcluster-shared/attributes/cluster.rb

@@ -4,6 +4,7 @@
 default['cluster']['license_dir'] = "#{node['cluster']['base_dir']}/licenses"
 default['cluster']['configs_dir'] = "#{node['cluster']['base_dir']}/configs"
 default['cluster']['shared_dir'] = "#{node['cluster']['base_dir']}/shared"
+default['cluster']['examples_dir'] = "#{node['cluster']['base_dir']}/examples"
 default['cluster']['shared_dir_login_nodes'] = "#{node['cluster']['base_dir']}/shared_login_nodes"
 default['cluster']['log_base_dir'] = '/var/log/parallelcluster'
 default['cluster']['etc_dir'] = '/etc/parallelcluster'