Software RAID: Get component devices by md UUID

Scanning the output of mdadm commands for RAID members will miss component devices which are currently not part of the RAID. For proper cleaning it is better to scan block devices for a signature of the md device for which we would like to get the components. Story: #2008186 Task: #40947 Change-Id: Ib46612697851e36a16d272ccaeb0115106253863
2020-09-24 16:25:01 +02:00 · 2020-09-24 16:25:01 +02:00 · 044c64dbc0
parent 6cb781b24e
commit 044c64dbc0
3 changed files with 152 additions and 20 deletions
--- a/ironic_python_agent/hardware.py
+++ b/ironic_python_agent/hardware.py
@ -136,10 +136,34 @@ def _check_for_iscsi():
                    "Error: %s", e)


+def _get_md_uuid(raid_device):
+    """Get the md UUID of a Software RAID device.
+
+    :param raid_device: A Software RAID block device name.
+    :returns: A string containing the UUID of an md device.
+    """
+    try:
+        out, _ = utils.execute('mdadm', '--detail', raid_device,
+                               use_standard_locale=True)
+    except processutils.ProcessExecutionError as e:
+        msg = ('Could not get the details of %(dev)s: %(err)s' %
+               {'dev': raid_device, 'err': e})
+        LOG.warning(msg)
+        return
+
+    lines = out.splitlines()
+    # the first line contains the md device itself
+    for line in lines[1:]:
+        match = re.search(r'UUID : ([a-f0-9:]+)', line)
+        if match:
+            return match.group(1)
+
+
 def _get_component_devices(raid_device):
    """Get the component devices of a Software RAID device.

-    Examine an md device and return its constituent devices.
+    Get the UUID of the md device and scan all other devices
+    for the same md UUID.

    :param raid_device: A Software RAID block device name.
    :returns: A list of the component devices.
@ -147,22 +171,35 @@ def _get_component_devices(raid_device):
    if not raid_device:
        return []

-    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})
-        LOG.warning(msg)
+    md_uuid = _get_md_uuid(raid_device)
+    if not md_uuid:
        return []
+    LOG.debug('%s has UUID %s', raid_device, md_uuid)

    component_devices = []
-    lines = out.splitlines()
-    # the first line contains the md device itself
-    for line in lines[1:]:
-        device = re.findall(r'/dev/\w+', line)
-        component_devices += device
+    block_devices = list_all_block_devices()
+    block_devices.extend(list_all_block_devices(block_type='part',
+                                                ignore_raid=True))
+    for bdev in block_devices:
+        try:
+            out, _ = utils.execute('mdadm', '--examine', bdev.name,
+                                   use_standard_locale=True)
+        except processutils.ProcessExecutionError as e:
+            if "No md superblock detected" in str(e):
+                # actually not a component device
+                LOG.debug('Not a component device %s', bdev.name)
+                continue
+            else:
+                LOG.warning("Failed to examine device %s: %s",
+                            bdev.name, e)
+                continue
+        lines = out.splitlines()
+        for line in lines:
+            if md_uuid in line:
+                component_devices.append(bdev.name)

+    LOG.info('Found component devices for %s:',
+             raid_device, component_devices)
    return component_devices


--- a/ironic_python_agent/tests/unit/test_hardware.py
+++ b/ironic_python_agent/tests/unit/test_hardware.py
@ -801,6 +801,64 @@ MDADM_DETAIL_OUTPUT_BROKEN_RAID0 = ("""/dev/md126:
 """)


+MDADM_EXAMINE_OUTPUT_MEMBER = ("""/dev/sda1:
+          Magic : a92b4efc
+        Version : 1.2
+    Feature Map : 0x0
+     Array UUID : 83143055:2781ddf5:2c8f44c7:9b45d92e
+           Name : horse.cern.ch:1  (local to host abc.xyz.com)
+  Creation Time : Tue Jun 11 12:43:37 2019
+     Raid Level : raid1
+   Raid Devices : 2
+
+ Avail Dev Size : 2093056 sectors (1022.00 MiB 1071.64 MB)
+     Array Size : 1046528 KiB (1022.00 MiB 1071.64 MB)
+    Data Offset : 2048 sectors
+   Super Offset : 8 sectors
+   Unused Space : before=1968 sectors, after=0 sectors
+          State : clean
+    Device UUID : 88bf2723:d082f14f:f95e87cf:b7c59b83
+
+    Update Time : Sun Sep 27 01:00:08 2020
+  Bad Block Log : 512 entries available at offset 16 sectors
+       Checksum : 340a1610 - correct
+         Events : 178
+
+
+   Device Role : Active device 0
+   Array State : A. ('A' == active, '.' == missing, 'R' == replacing)
+""")
+
+
+MDADM_EXAMINE_OUTPUT_NON_MEMBER = ("""/dev/sdz1:
+          Magic : a92b4efc
+        Version : 1.2
+    Feature Map : 0x0
+     Array UUID : 83143055:2781ddf5:2c8f44c7:9b45d92f
+           Name : horse.cern.ch:1  (local to host abc.xyz.com)
+  Creation Time : Tue Jun 11 12:43:37 2019
+     Raid Level : raid1
+   Raid Devices : 2
+
+ Avail Dev Size : 2093056 sectors (1022.00 MiB 1071.64 MB)
+     Array Size : 1046528 KiB (1022.00 MiB 1071.64 MB)
+    Data Offset : 2048 sectors
+   Super Offset : 8 sectors
+   Unused Space : before=1968 sectors, after=0 sectors
+          State : clean
+    Device UUID : 88bf2723:d082f14f:f95e87cf:b7c59b84
+
+    Update Time : Sun Sep 27 01:00:08 2020
+  Bad Block Log : 512 entries available at offset 16 sectors
+       Checksum : 340a1610 - correct
+         Events : 178
+
+
+   Device Role : Active device 0
+   Array State : A. ('A' == active, '.' == missing, 'R' == replacing)
+""")
+
+
 class FakeHardwareManager(hardware.GenericHardwareManager):
    def __init__(self, hardware_support):
        self._hardware_support = hardware_support
@ -3908,16 +3966,47 @@ class TestGenericHardwareManager(base.IronicAgentTest):
                               self.node, [])

    @mock.patch.object(utils, 'execute', autospec=True)
-    def test__get_component_devices(self, mocked_execute):
+    def test__get_md_uuid(self, mocked_execute):
        mocked_execute.side_effect = [(MDADM_DETAIL_OUTPUT, '')]
-        component_devices = hardware._get_component_devices('/dev/md0')
-        self.assertEqual(['/dev/vde1', '/dev/vdf1'], component_devices)
+        md_uuid = hardware._get_md_uuid('/dev/md0')
+        self.assertEqual('83143055:2781ddf5:2c8f44c7:9b45d92e', md_uuid)

+    @mock.patch.object(hardware, '_get_md_uuid', autospec=True)
+    @mock.patch.object(hardware, 'list_all_block_devices', autospec=True)
    @mock.patch.object(utils, 'execute', autospec=True)
-    def test__get_component_devices_broken_raid0(self, mocked_execute):
-        mocked_execute.side_effect = [(MDADM_DETAIL_OUTPUT_BROKEN_RAID0, '')]
-        component_devices = hardware._get_component_devices('/dev/md126')
-        self.assertEqual(['/dev/sda2'], component_devices)
+    def test__get_component_devices(self, mocked_execute,
+                                    mocked_list_all_block_devices,
+                                    mocked_md_uuid):
+        raid_device1 = hardware.BlockDevice('/dev/md0', 'RAID-1',
+                                            107374182400, True)
+        sda = hardware.BlockDevice('/dev/sda', 'model12', 21, True)
+        sdz = hardware.BlockDevice('/dev/sdz', 'model12', 21, True)
+        sda1 = hardware.BlockDevice('/dev/sda1', 'model12', 21, True)
+        sdz1 = hardware.BlockDevice('/dev/sdz1', 'model12', 21, True)
+
+        mocked_md_uuid.return_value = '83143055:2781ddf5:2c8f44c7:9b45d92e'
+        hardware.list_all_block_devices.side_effect = [
+            [sda, sdz],    # list_all_block_devices
+            [sda1, sdz1],  # list_all_block_devices partitions
+        ]
+        mocked_execute.side_effect = [
+            ['mdadm --examine output for sda', '_'],
+            ['mdadm --examine output for sdz', '_'],
+            [MDADM_EXAMINE_OUTPUT_MEMBER, '_'],
+            [MDADM_EXAMINE_OUTPUT_NON_MEMBER, '_'],
+        ]
+
+        component_devices = hardware._get_component_devices(raid_device1)
+        self.assertEqual(['/dev/sda1'], component_devices)
+        mocked_execute.assert_has_calls([
+            mock.call('mdadm', '--examine', '/dev/sda',
+                      use_standard_locale=True),
+            mock.call('mdadm', '--examine', '/dev/sdz',
+                      use_standard_locale=True),
+            mock.call('mdadm', '--examine', '/dev/sda1',
+                      use_standard_locale=True),
+            mock.call('mdadm', '--examine', '/dev/sdz1',
+                      use_standard_locale=True)])

    @mock.patch.object(utils, 'execute', autospec=True)
    def test_get_holder_disks(self, mocked_execute):
--- a/releasenotes/notes/get_md_components_by_uuid-7f08d423ea9e7c94.yaml
+++ b/releasenotes/notes/get_md_components_by_uuid-7f08d423ea9e7c94.yaml
@ -0,0 +1,6 @@
+---
+fixes:
+  - |
+    Detects md component devices by their UUID, rather than by scanning the
+    output of mdadm. This will prevent that devices miss md superblock
+    cleanup when they are currently not part of an array.