[Lustre] Fix device discovery for older versions of Lustre (<2.15.5) (#21901)

* Add test for non-yaml output of lctl dl

* Implement fallback to non-yaml version of lctl dl

* Add changelog

* Format

* Assert for exact match of devices in test

* Unify device list fixtures for readability

* Improve code based on review

* Remove redundant devices definition

* Format

* Minor syntax improvement

Author: dkirov-dd

lustre/changelog.d/21901.fixed

@@ -0,0 +1 @@
+Fix device discovery for older versions of Lustre (<2.15.5)

lustre/datadog_checks/lustre/check.py

@@ -15,6 +15,7 @@
 from .constants import (
     CURATED_PARAMS,
     DEFAULT_STATS,
+    DEVICE_ATTR_NAMES,
     EXTRA_STATS,
     FILESYSTEM_DISCOVERY_PARAM_MAPPING,
     IGNORED_LNET_GROUPS,
@@ -117,6 +118,7 @@ def __init__(self, name: str, init_config: Dict[str, Any], instances: List[Dict[
         self.filesystems: List[str] = self.instance.get('filesystems', [])
         # If filesystems were provided by the instance, do not update the filesystem list
         self.filesystem_discovery: bool = False if self.filesystems else True
+        self._use_yaml: bool = True  # Older versions of Lustre (<2.15.5) do not support yaml as an output
         self.node_type: str = self.instance.get('node_type', self._find_node_type())
 
         self.tags: List[str] = self.instance.get('tags', [])
@@ -172,9 +174,28 @@ def _update_devices(self) -> None:
         Find devices using the lctl dl command.
         '''
         self.log.debug('Updating device list...')
-        output = self._run_command('lctl', 'dl', '-y')
-        device_data = yaml.safe_load(output)
-        self.devices = device_data.get('devices', [])
+        devices = []
+        if self._use_yaml:
+            try:
+                output = self._run_command('lctl', 'dl', '-y')
+                device_data = yaml.safe_load(output)
+                devices = device_data.get('devices', [])
+            except AttributeError:
+                self.log.debug('Device update failed with yaml flag, retrying without it.')
+                self._use_yaml = False
+        if not self._use_yaml:
+            output = self._run_command('lctl', 'dl')
+            for device_line in output.splitlines():
+                device_attr = device_line.split()
+                if not len(device_attr) == len(DEVICE_ATTR_NAMES):
+                    self.log.error('Could not parse device info: %s', device_line)
+                    continue
+                devices.append(dict(zip(DEVICE_ATTR_NAMES, device_attr)))
+        if not devices:
+            self.log.error("No devices detected.")
+            return
+        self.devices = devices
+        self.log.debug('Devices successfully updated.')
 
     def _update_filesystems(self) -> None:
         '''

lustre/datadog_checks/lustre/constants.py

@@ -10,6 +10,8 @@
     'client': (r'llite.*.stats', r'(?<=llite\.).*(?=-[^-]*\.stats)'),
 }
 
+DEVICE_ATTR_NAMES = ['index', 'status', 'type', 'name', 'uuid', 'refcount']
+
 IGNORED_STATS = {
     'snapshot_time',
     'start_time',

lustre/tests/fixtures/client_dl.txt

@@ -0,0 +1,5 @@
+  0 UP mgc MGC172.31.16.218@tcp 7d3988a7-145f-444e-9953-58e3e6d97385 5
+  1 UP lov lustre-clilov-ffff8b904341d000 ac8e54e3-1334-4865-a3f5-4f61ce87bdd1 4
+  2 UP lmv lustre-clilmv-ffff8b904341d000 ac8e54e3-1334-4865-a3f5-4f61ce87bdd1 5
+  3 UP mdc lustre-MDT0000-mdc-ffff8b904341d000 ac8e54e3-1334-4865-a3f5-4f61ce87bdd1 5
+  4 UP osc lustre-OST0001-osc-ffff8b904341d000 ac8e54e3-1334-4865-a3f5-4f61ce87bdd1 5

lustre/tests/test_unit.py

@@ -653,3 +653,79 @@ def test_sanitize_command(bin_path, should_pass):
         # Should raise ValueError
         with pytest.raises(ValueError):
             _sanitize_command(bin_path)
+
+
+@pytest.mark.parametrize(
+    ['yaml_fixture', 'non_yaml_fixture'],
+    [
+        pytest.param('', '', id='no devices'),
+        pytest.param('client_dl_yaml.txt', '', id='devices from yaml'),
+        pytest.param('', 'client_dl.txt', id='devices without yaml'),
+    ],
+)
+def test_device_discovery(mock_lustre_commands, yaml_fixture, non_yaml_fixture):
+    """Devices should be discovered regardless of Lustre version"""
+    mapping = {
+        'lctl get_param -ny version': 'all_version.txt',
+        'lctl dl -y': yaml_fixture,
+        'lctl dl': non_yaml_fixture,
+        'lfs changelog': 'test_changelog',
+    }
+
+    with mock_lustre_commands(mapping):
+        check = LustreCheck('lustre', {}, [{}])
+
+    # Assert device contents
+    if not yaml_fixture and not non_yaml_fixture:
+        assert check.devices == []
+    else:
+        # Expected device structure - same for both YAML and non-YAML fixtures
+        expected_devices = [
+            {
+                'index': '0',
+                'status': 'UP',
+                'type': 'mgc',
+                'name': 'MGC172.31.16.218@tcp',
+                'uuid': '7d3988a7-145f-444e-9953-58e3e6d97385',
+                'refcount': '5',
+            },
+            {
+                'index': '1',
+                'status': 'UP',
+                'type': 'lov',
+                'name': 'lustre-clilov-ffff8b904341d000',
+                'uuid': 'ac8e54e3-1334-4865-a3f5-4f61ce87bdd1',
+                'refcount': '4',
+            },
+            {
+                'index': '2',
+                'status': 'UP',
+                'type': 'lmv',
+                'name': 'lustre-clilmv-ffff8b904341d000',
+                'uuid': 'ac8e54e3-1334-4865-a3f5-4f61ce87bdd1',
+                'refcount': '5',
+            },
+            {
+                'index': '3',
+                'status': 'UP',
+                'type': 'mdc',
+                'name': 'lustre-MDT0000-mdc-ffff8b904341d000',
+                'uuid': 'ac8e54e3-1334-4865-a3f5-4f61ce87bdd1',
+                'refcount': '5',
+            },
+            {
+                'index': '4',
+                'status': 'UP',
+                'type': 'osc',
+                'name': 'lustre-OST0001-osc-ffff8b904341d000',
+                'uuid': 'ac8e54e3-1334-4865-a3f5-4f61ce87bdd1',
+                'refcount': '5',
+            },
+        ]
+
+        # Convert YAML integer types to strings for consistency
+        actual_devices = [
+            {k: str(v) if isinstance(v, int) else v for k, v in device.items()} for device in check.devices
+        ]
+
+        assert actual_devices == expected_devices