Add integration test for scontrol update nodelist sorting bug

CHANGELOG.md

@@ -1,5 +1,12 @@
 CHANGELOG
 =========
+
+3.4.1
+-----
+
+**BUG FIXES**
+- Fix an issue with the Slurm scheduler that might incorrectly apply updates to its internal registry of compute nodes. This might result in EC2 instances to become inaccessible or backed by an incorrect instance type.
+
 3.4.0
 -----
 

tests/integration-tests/configs/common/common.yaml

@@ -511,6 +511,12 @@ schedulers:
         instances: {{ common.INSTANCES_DEFAULT_X86 }}
         oss: ["ubuntu2004"]
         schedulers: ["slurm"]
+  test_slurm.py::test_scontrol_update_nodelist_sorting:
+    dimensions:
+      - regions: ["ca-central-2"]
+        instances: {{ common.INSTANCES_DEFAULT_X86 }}
+        oss: ["alinux2"]
+        schedulers: ["slurm"]
   test_slurm_accounting.py::test_slurm_accounting:
     dimensions:
       - regions: ["us-east-1", "ap-south-1"]

tests/integration-tests/tests/schedulers/test_slurm.py

@@ -560,6 +560,56 @@ def test_update_slurm_reconfigure_race_condition(
         )
 
 
+@pytest.mark.usefixtures("region", "os", "instance", "scheduler")
+def test_scontrol_update_nodelist_sorting(
+    pcluster_config_reader,
+    clusters_factory,
+    test_datadir,
+    scheduler_commands_factory,
+):
+    """
+    Test that scontrol update node follows the order of the nodelist provided by the user.
+
+    In Slurm 22.05 the scontrol update node logic was modified and a sorting routine was
+    introduced, which modified the order of the nodes in the nodelist.
+    If `scontrol update node nodename=nodelist nodeaddr=nodeaddrlist` is called, only the
+    nodelist was sorted (not the nodeaddrlist). This causes mismatches between the Slurm
+    nodenames and the assigned addresses.
+
+    See https://bugs.schedmd.com/show_bug.cgi?id=15731
+    """
+
+    max_count_cr1 = max_count_cr2 = 4
+
+    cluster_config = pcluster_config_reader(
+        config_file="pcluster.config.yaml",
+        output_file="pcluster.config.initial.yaml",
+        max_count_cr1=max_count_cr1,
+        max_count_cr2=max_count_cr2,
+    )
+    cluster = clusters_factory(cluster_config)
+    remote_command_executor = RemoteCommandExecutor(cluster)
+    slurm_commands = scheduler_commands_factory(remote_command_executor)
+
+    assert_compute_node_states(slurm_commands, compute_nodes=None, expected_states=["idle~"])
+
+    nodes_in_queue1 = slurm_commands.get_compute_nodes("queue1", all_nodes=True)
+    nodes_in_queue2 = slurm_commands.get_compute_nodes("queue2", all_nodes=True)
+
+    # Create an unsorted list of nodes to be updated (queue2 is alphabetically after queue1)``:s
+    nodelist = f"{nodes_in_queue2[0]},{nodes_in_queue1[0]}"
+
+    # Stop clustermgtd since it may fix the situation under the hood if it calls scontrol update
+    # with a sorted list of nodes
+    remote_command_executor.run_remote_command("sudo systemctl stop supervisord")
+
+    # Run scontrol update with unsorted list of nodes
+    remote_command_executor.run_remote_command(f"sudo -i scontrol update nodename={nodelist} nodeaddr={nodelist}")
+
+    assert_that(slurm_commands.get_node_attribute(nodes_in_queue1[0], "NodeAddr")).is_equal_to(nodes_in_queue1[0])
+    assert_that(slurm_commands.get_node_attribute(nodes_in_queue2[0], "NodeAddr")).is_equal_to(nodes_in_queue2[0])
+
+
 @pytest.mark.usefixtures("region", "os", "instance", "scheduler")
 def test_slurm_overrides(
     scheduler,

tests/integration-tests/tests/schedulers/test_slurm/test_scontrol_update_nodelist_sorting/pcluster.config.yaml

@@ -0,0 +1,31 @@
+Image:
+  Os: {{ os }}
+HeadNode:
+  InstanceType: {{ instance }}
+  Networking:
+    SubnetId: {{ public_subnet_id }}
+  Ssh:
+    KeyName: {{ key_name }}
+Scheduling:
+  Scheduler: slurm
+  SlurmQueues:
+    - Name: queue1
+      Networking:
+        SubnetIds:
+          - {{ private_subnet_id }}
+      ComputeResources:
+        - Name: resource1
+          Instances:
+            - InstanceType: {{ instance }}
+          MinCount: 0
+          MaxCount: {{ max_count_cr1 }}
+    - Name: queue2
+      Networking:
+        SubnetIds:
+          - {{ private_subnet_id }}
+      ComputeResources:
+        - Name: resource2
+          Instances:
+            - InstanceType: {{ instance }}
+          MinCount: 0
+          MaxCount: {{ max_count_cr2 }}