Software RAID: Ignore missing component devices or holder disks

With the addition of devices with type 'MD' during
list_all_block_devices (for "empty" devices), partitions on top
of md devices are now returned as well. These don't need deletion,
so skip these during software RAID cleanup.

Story: #2006355
Task: #36121

Change-Id: Idff9b9b0cbe42b57aa315ae07c0b51c35c9ade1e

2 files changed, 31 insertions, 8 deletions

diff --git a/ironic_python_agent/hardware.py b/ironic_python_agent/hardware.py
index 9ed0c5b8..bbaf9ee9 100644
--- a/ironic_python_agent/hardware.py
+++ b/ironic_python_agent/hardware.py
@@ -127,15 +127,16 @@ def _get_component_devices(raid_device):
     if not raid_device:
         return []
 
-    component_devices = []
     try:
         out, _ = utils.execute('mdadm', '--detail', raid_device,
                                use_standard_locale=True)
     except processutils.ProcessExecutionError as e:
         msg = ('Could not get component devices of %(dev)s: %(err)s' %
                {'dev': raid_device, 'err': e})
-        raise errors.SoftwareRAIDError(msg)
+        LOG.warning(msg)
+        return []
 
+    component_devices = []
     lines = out.splitlines()
     # the first line contains the md device itself
     for line in lines[1:]:
@@ -156,16 +157,16 @@ def get_holder_disks(raid_device):
     if not raid_device:
         return []
 
-    holder_disks = []
-
     try:
         out, _ = utils.execute('mdadm', '--detail', raid_device,
                                use_standard_locale=True)
     except processutils.ProcessExecutionError as e:
         msg = ('Could not get holder disks of %(dev)s: %(err)s' %
                {'dev': raid_device, 'err': e})
-        raise errors.SoftwareRAIDError(msg)
+        LOG.warning(msg)
+        return []
 
+    holder_disks = []
     lines = out.splitlines()
     # the first line contains the md device itself
     for line in lines[1:]:
@@ -1549,12 +1550,21 @@ class GenericHardwareManager(HardwareManager):
         raid_devices = list_all_block_devices(block_type='raid',
                                               ignore_raid=False)
         for raid_device in raid_devices:
+            component_devices = _get_component_devices(raid_device.name)
+            if not component_devices:
+                # A "Software RAID device" without components is usually
+                # a partition on an md device (as, for instance, created
+                # by the conductor for the config drive). This will be
+                # cleaned with the hosting md device.
+                msg = ("Software RAID cleaning is skipping "
+                       "partition %s" % raid_device.name)
+                LOG.info(msg)
+                continue
+            holder_disks = get_holder_disks(raid_device.name)
+
             LOG.info("Deleting Software RAID device {}".format(
                      raid_device.name))
-
-            component_devices = _get_component_devices(raid_device.name)
             LOG.debug('Found component devices %s', component_devices)
-            holder_disks = get_holder_disks(raid_device.name)
             LOG.debug('Found holder disks %s', holder_disks)
 
             # Remove md devices.
diff --git a/ironic_python_agent/tests/unit/test_hardware.py b/ironic_python_agent/tests/unit/test_hardware.py
index f540f1f5..49cae296 100644
--- a/ironic_python_agent/tests/unit/test_hardware.py
+++ b/ironic_python_agent/tests/unit/test_hardware.py
@@ -2955,6 +2955,19 @@ class TestGenericHardwareManager(base.IronicAgentTest):
             mock.call('wipefs', '-af', '/dev/sda'),
             mock.call('wipefs', '-af', '/dev/sdb')])
 
+    @mock.patch.object(hardware, '_get_component_devices', autospec=True)
+    @mock.patch.object(hardware, 'list_all_block_devices', autospec=True)
+    @mock.patch.object(utils, 'execute', autospec=True)
+    def test_delete_configuration_partition(self, mocked_execute, mocked_list,
+                                            mocked_get_component):
+        raid_device1_part1 = hardware.BlockDevice('/dev/md0p1', 'RAID-1',
+                                                  1073741824, True)
+        hardware.list_all_block_devices.return_value = [raid_device1_part1]
+        mocked_get_component.return_value = []
+
+        self.assertIsNone(self.hardware.delete_configuration(self.node, []))
+        mocked_execute.assert_has_calls([])
+
     @mock.patch.object(utils, 'execute', autospec=True)
     def test_validate_configuration_valid_raid1(self, mocked_execute):
         raid_config = {