[DFSM]Using cluster-config-version file to be used during Init and Update phase of cluster

cookbooks/aws-parallelcluster-environment/recipes/config/login_nodes_keys.rb → cookbooks/aws-parallelcluster-environment/recipes/init/login_nodes_keys.rb

@@ -21,7 +21,7 @@
 script_dir = "#{keys_dir}/scripts"
 script_path = "#{script_dir}/keys-manager.sh"
 
-sync_file_path = "#{keys_dir}/.login_nodes_keys_sync_file"
+sync_file_path = "#{keys_dir}/cluster-config-version"
 
 case node['cluster']['node_type']
 when 'ComputeFleet'
@@ -46,11 +46,11 @@
     user 'root'
   end
 
-  write_sync_file(sync_file_path)
+  write_config_version_file(sync_file_path)
 
 when 'LoginNode'
 
-  wait_sync_file(sync_file_path)
+  wait_cluster_config_file(sync_file_path)
 
   execute 'Import Login Nodes keys' do
     command "bash #{script_path} --import --folder-path #{keys_dir}"

cookbooks/aws-parallelcluster-environment/spec/unit/recipes/login_nodes_keys_spec.rb

@@ -2,7 +2,7 @@
 
 describe 'aws-parallelcluster-environment::login_nodes_keys' do
   SHARED_DIR_LOGIN_NODES = "/SHARED_DIR_LOGIN_NODES".freeze
-  SYNC_FILE = "#{SHARED_DIR_LOGIN_NODES}/.login_nodes_keys_sync_file".freeze
+  SYNC_FILE = "#{SHARED_DIR_LOGIN_NODES}/cluster-config-version".freeze
   CLUSTER_CONFIG_VERSION = "CLUSTER_CONFIG_VERSION".freeze
 
   for_all_oses do |platform, version|
@@ -94,10 +94,10 @@
         cached(:node) { chef_run.node }
 
         it "waits for cluster config version file" do
-          is_expected.to run_bash("Wait for synchronization file at #{SYNC_FILE} to be written for version #{CLUSTER_CONFIG_VERSION}").with(
+          is_expected.to run_bash("Wait for file at #{SYNC_FILE} to be updated by the head node").with(
             code: "[[ \"$(cat #{SYNC_FILE})\" == \"#{CLUSTER_CONFIG_VERSION}\" ]] || exit 1",
             retries: 30,
-            retry_delay: 10,
+            retry_delay: 15,
             timeout: 5
           )
         end

cookbooks/aws-parallelcluster-platform/resources/fetch_config.rb

@@ -162,27 +162,4 @@ def fetch_instance_type_data
       fetch_s3_object("copy_instance_type_data_from_s3", node['cluster']['instance_types_data_s3_key'], node['cluster']['instance_types_data_path'], instance_version_id)
     end
   end
-
-  def write_config_version_file(path)
-    # Write the cluster config version into the specified file.
-    # This file is used as a synchronization point between the head node and the other cluster nodes.
-    # In particular, the head node uses this file to signal to other cluster nodes that all files
-    # in the shared folder related to the cluster config have been updated with the current config version.
-    file path do
-      content node['cluster']['cluster_config_version']
-      mode '0644'
-      owner 'root'
-      group 'root'
-    end
-  end
-
-  def wait_cluster_config_file(path)
-    # Wait for the config version file to contain the current cluster config version.
-    bash "Wait cluster config files to be updated by the head node" do
-      code "[[ \"$(cat #{path})\" == \"#{node['cluster']['cluster_config_version']}\" ]] || exit 1"
-      retries 30
-      retry_delay 15
-      timeout 5
-    end
-  end
 end

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

@@ -59,7 +59,7 @@
     end
 
     it "does not wait for cluster config version file" do
-      is_expected.not_to run_execute("Wait cluster config files to be updated by the head node")
+      is_expected.not_to run_execute("Wait for file at /shared_dir_login_nodes/cluster-config-version to be updated by the head node")
     end
   end
 
@@ -90,7 +90,7 @@
       end
 
       it "does not wait for cluster config version file" do
-        is_expected.not_to run_execute("Wait cluster config files to be updated by the head node")
+        is_expected.not_to run_execute("Wait for file at #{cluster_config_path} to be updated by the head node")
       end
 
       it "reads config from shared folder" do
@@ -146,7 +146,7 @@
                               else
                                 raise "Unsupported node_type #{node_type}"
                               end
-        is_expected.to run_bash("Wait cluster config files to be updated by the head node").with(
+        is_expected.to run_bash("Wait for file at #{config_version_file} to be updated by the head node").with(
           code: "[[ \"$(cat #{config_version_file})\" == \"cluster_config_version\" ]] || exit 1",
           retries: 30,
           retry_delay: 15,

cookbooks/aws-parallelcluster-shared/libraries/helpers.rb

@@ -82,28 +82,28 @@ def is_custom_node?
   !custom_node_package.nil? && !custom_node_package.empty?
 end
 
-def write_sync_file(path)
-  # Write a synchronization file containing the current cluster config version.
-  # Synchronization files are used as a synchronization point between cluster nodes
-  # to signal that a group of actions have been completed.
+def write_config_version_file(path)
+  # Write the cluster config version into the specified file.
+  # This file is used as a synchronization point between the head node and the other cluster nodes.
+  # In particular, the head node uses this file to signal to other cluster nodes that all files
+  # in the shared folder related to the cluster config have been updated with the current config version.
   file path do
-    content node["cluster"]["cluster_config_version"]
-    mode "0644"
-    owner "root"
-    group "root"
+    content node['cluster']['cluster_config_version']
+    mode '0644'
+    owner 'root'
+    group 'root'
   end
 end
 
-def wait_sync_file(path)
+def wait_cluster_config_file(path)
   # Wait for a synchronization file to be written for the current cluster config version.
   # Synchronization files are used as a synchronization point between cluster nodes
   # to signal that a group of actions have been completed.
-  cluster_config_version = node["cluster"]["cluster_config_version"]
   # Wait for the config version file to contain the current cluster config version.
-  bash "Wait for synchronization file at #{path} to be written for version #{cluster_config_version}" do
-    code "[[ \"$(cat #{path})\" == \"#{cluster_config_version}\" ]] || exit 1"
+  bash "Wait for file at #{path} to be updated by the head node" do
+    code "[[ \"$(cat #{path})\" == \"#{node['cluster']['cluster_config_version']}\" ]] || exit 1"
     retries 30
-    retry_delay 10
+    retry_delay 15
     timeout 5
   end
 end