[SlurmTopo] We cleanup or generate Topology only if p6egb200_block_size exist

Author: Himani Anil Deshpande

cookbooks/aws-parallelcluster-slurm/files/default/head_node_slurm/slurm/pcluster_topology_generator.py

@@ -13,6 +13,7 @@
 import yaml
 import logging
 import traceback
+import os
 log = logging.getLogger()
 
 
@@ -118,22 +119,42 @@ def generate_topology_config_file(output_file: str, input_file: str, block_sizes
     log.info("Finished.")
 
 
+def cleanup_topology_config_file(file_path):
+    """Cleanup topology.conf file."""
+    try:
+        if os.path.exists(file_path):
+            log.info("Cleaning up %s", file_path)
+            os.remove(file_path)
+    except Exception as err:
+        log.warning("Unable to delete %s due to %s", file_path, err)
+
+
 def main():
     try:
         logging.basicConfig(
             level=logging.INFO, format="%(asctime)s - [%(name)s:%(funcName)s] - %(levelname)s - %(message)s"
         )
         log.info("Running ParallelCluster Topology Config Generator")
         parser = argparse.ArgumentParser(description="Take in Topology configuration generator related parameters")
+        cleanup_or_generate_exclusive_group = parser.add_mutually_exclusive_group(required=True)
         parser.add_argument("--output-file", help="The output file for generated topology.conf", required=True)
         parser.add_argument(
             "--input-file",
             help="Yaml file containing pcluster CLI configuration file with default values",
             required=True,
         )
-        parser.add_argument("--block-sizes", help="Block Size of topology.conf", required=True)
+        cleanup_or_generate_exclusive_group.add_argument("--block-sizes", help="Block Sizes of topology.conf")
+        cleanup_or_generate_exclusive_group.add_argument(
+            "--cleanup",
+            action="store_true",
+            help="Cleanup topology.conf",
+        )
         args = parser.parse_args()
-        generate_topology_config_file(args.output_file, args.input_file, args.block_sizes)
+        if args.cleanup:
+            cleanup_topology_config_file(args.output_file)
+        else:
+            generate_topology_config_file(args.output_file, args.input_file, args.block_sizes)
+        log.info("Completed Execution of ParallelCluster Topology Config Generator")
     except Exception as e:
         log.exception("Failed to generate Topology.conf, exception: %s", e)
         raise

cookbooks/aws-parallelcluster-slurm/recipes/update/update_head_node.rb

@@ -161,14 +161,22 @@ def update_nodes_in_queue(strategy, queues)
       mode '0644'
     end
 
+    if node['cluster']['p6egb200_block_sizes'].nil? && are_queues_updated? && ::File.exist?("#{node['cluster']['slurm']['install_dir']}/etc/topology.conf")
+      # If topology.conf exist and Capacity Block is removed, we cleanup
+      topology_generator_command_args = " --cleanup"
+    elsif node['cluster']['p6egb200_block_sizes'].nil? && !are_queues_updated?
+      # We do nothing if p6e-gb200 is not used and queues are not updated
+      topology_generator_command_args = nil
+    else
+      topology_generator_command_args = " --block-sizes #{node['cluster']['p6egb200_block_sizes']}"
+    end
     # Update Slurm topology.conf file
     execute "update or cleanup topology.conf" do
       command "#{cookbook_virtualenv_path}/bin/python #{node['cluster']['scripts_dir']}/slurm/pcluster_topology_generator.py"\
                 " --output-file #{node['cluster']['slurm']['install_dir']}/etc/topology.conf"\
-                " --block-sizes #{node['cluster']['p6egb200_block_sizes']}"\
-                " --input-file #{node['cluster']['cluster_config_path']}"
-      not_if { ::File.exist?(node['cluster']['previous_cluster_config_path']) && !are_queues_updated? && node['cluster']['p6egb200_block_sizes'].nil? }
-      #TODO: Need to remove topology.conf if CB is removed
+                " --input-file #{node['cluster']['cluster_config_path']}"\
+                "#{topology_generator_command_args}"
+      not_if { ::File.exist?(node['cluster']['previous_cluster_config_path']) && topology_generator_command_args.nil? }
     end
   end
   not_if { platform?('amazon') && node['platform_version'] == "2" }