Avoid scaling if content of SLURM_RESUME_FILE is empty

Avoid scaling if content of SLURM_RESUME_FILE is empty, logging an error. Signed-off-by: Luca Carrogu <[email protected]>
aws · Oct 12, 2023 · d70da33 · d70da33
1 parent ee6a953
commit d70da33
Show file tree

Hide file tree

Showing 3 changed files with 2 additions and 398 deletions.
diff --git a/src/slurm_plugin/instance_manager.py b/src/slurm_plugin/instance_manager.py
@@ -555,15 +555,7 @@ def add_instances(
         else:
             logger.error(
                 "Not possible to perform job level scaling because Slurm resume file content is empty. "
-                "Falling back to node list scaling"
-            )
-            logger.info("The nodes_resume list from Slurm Resume Program is %s", print_with_count(node_list))
-            self._add_instances_for_nodes(
-                node_list=node_list,
-                launch_batch_size=launch_batch_size,
-                assign_node_batch_size=assign_node_batch_size,
-                update_node_address=update_node_address,
-                scaling_strategy=scaling_strategy,
+                "No scaling actions will be taken."
             )
 
         self._terminate_unassigned_launched_instances(terminate_batch_size)

diff --git a/tests/slurm_plugin/test_instance_manager.py b/tests/slurm_plugin/test_instance_manager.py
@@ -1363,13 +1363,7 @@ def test_add_instances(
                 scaling_strategy=scaling_strategy,
             )
         else:
-            instance_manager._add_instances_for_nodes.assert_called_once_with(
-                node_list=node_list,
-                launch_batch_size=launch_batch_size,
-                assign_node_batch_size=assign_node_batch_size,
-                update_node_address=update_node_address,
-                scaling_strategy=scaling_strategy,
-            )
+            instance_manager._add_instances_for_resume_file.assert_not_called()
 
 
 class TestJobLevelScalingInstanceManager:
@@ -1470,13 +1464,6 @@ def test_add_instances(
 
         if not slurm_resume:
             instance_manager._add_instances_for_resume_file.assert_not_called()
-            instance_manager._add_instances_for_nodes.assert_called_with(
-                node_list=node_list,
-                launch_batch_size=launch_batch_size,
-                assign_node_batch_size=assign_node_batch_size,
-                update_node_address=update_node_address,
-                scaling_strategy=scaling_strategy,
-            )
             assert_that(caplog.text).contains(
                 "Not possible to perform job level scaling " "because Slurm resume file content is empty."
             )