Burn-in: Add disk step

Add a clean step for disk burn-in via fio. Get basic
run parameters from the node's driver_info.

Story: #2007523
Task: #42384

Change-Id: I5f5e336bd629846b3d779fd0fc7a2060b385b035

doc/source/admin/hardware_managers.rst

@@ -77,6 +77,8 @@ Clean steps
 ``deploy.burnin_cpu``
     Stress-test the CPUs of a node via stress-ng for a configurable
     amount of time. Disabled by default.
+``deploy.burnin_disk``
+    Stress-test the disks of a node via fio. Disabled by default.
 ``deploy.burnin_memory``
     Stress-test the memory of a node via stress-ng for a configurable
     amount of time. Disabled by default.

ironic_python_agent/burnin.py

@@ -15,6 +15,7 @@ from oslo_concurrency import processutils
 from oslo_log import log
 
 from ironic_python_agent import errors
+from ironic_python_agent import hardware
 
 LOG = log.getLogger(__name__)
 
@@ -78,3 +79,39 @@ def stress_ng_vm(node):
                      {'err': e})
         LOG.error(error_msg)
         raise errors.CommandExecutionError(error_msg)
+
+
+def fio_disk(node):
+    """Burn-in the disks with fio
+
+    Run an fio randrw job for a configurable number of iterations
+    or a given amount of time.
+
+    :param node: Ironic node object
+    :raises: CommandExecutionError if the execution of fio fails.
+    """
+    info = node.get('driver_info', {})
+    # 4 iterations, same as badblock's default
+    loops = info.get('agent_burnin_fio_disk_loops', 4)
+    runtime = info.get('agent_burnin_fio_disk_runtime', 0)
+
+    args = ['fio', '--rw', 'readwrite', '--bs', '4k', '--direct', 1,
+            '--ioengine', 'libaio', '--iodepth', '32', '--verify',
+            'crc32c', '--verify_dump', 1, '--continue_on_error', 'verify',
+            '--loops', loops, '--runtime', runtime, '--time_based']
+
+    devices = hardware.list_all_block_devices()
+    for device in devices:
+        args.extend(['--name', device.name])
+
+    LOG.debug('Burn-in fio disk command: %s', ' '.join(map(str, args)))
+
+    try:
+        out, _ = utils.execute(*args)
+        # fio reports on stdout
+        LOG.info(out)
+    except (processutils.ProcessExecutionError, OSError) as e:
+        error_msg = ("fio (disk) failed with error %(err)s",
+                     {'err': e})
+        LOG.error(error_msg)
+        raise errors.CommandExecutionError(error_msg)

ironic_python_agent/hardware.py

@@ -1402,6 +1402,14 @@ class GenericHardwareManager(HardwareManager):
         """
         burnin.stress_ng_cpu(node)
 
+    def burnin_disk(self, node, ports):
+        """Burn-in the disk
+
+        :param node: Ironic node object
+        :param ports: list of Ironic port objects
+        """
+        burnin.fio_disk(node)
+
     def burnin_memory(self, node, ports):
         """Burn-in the memory
 
@@ -1890,6 +1898,13 @@ class GenericHardwareManager(HardwareManager):
                 'reboot_requested': False,
                 'abortable': True
             },
+            {
+                'step': 'burnin_disk',
+                'priority': 0,
+                'interface': 'deploy',
+                'reboot_requested': False,
+                'abortable': True
+            },
             {
                 'step': 'burnin_memory',
                 'priority': 0,

ironic_python_agent/tests/unit/test_burnin.py

@@ -17,6 +17,7 @@ from oslo_concurrency import processutils
 
 from ironic_python_agent import burnin
 from ironic_python_agent import errors
+from ironic_python_agent import hardware
 from ironic_python_agent.tests.unit import base
 
 
@@ -63,6 +64,7 @@ class TestBurnin(base.IronicAgentTest):
         burnin.stress_ng_vm(node)
 
         mock_execute.assert_called_once_with(
+
             'stress-ng', '--vm', 0, '--vm-bytes', '98%',
             '--timeout', 86400, '--metrics-brief')
 
@@ -89,3 +91,56 @@ class TestBurnin(base.IronicAgentTest):
 
         self.assertRaises(errors.CommandExecutionError,
                           burnin.stress_ng_vm, node)
+
+    @mock.patch.object(hardware, 'list_all_block_devices', autospec=True)
+    def test_fio_disk_default(self, mock_list, mock_execute):
+
+        node = {'driver_info': {}}
+
+        mock_list.return_value = [
+            hardware.BlockDevice('/dev/sdj', 'big', 1073741824, True),
+            hardware.BlockDevice('/dev/hdaa', 'small', 65535, False),
+        ]
+        mock_execute.return_value = (['out', 'err'])
+
+        burnin.fio_disk(node)
+
+        mock_execute.assert_called_once_with(
+            'fio', '--rw', 'readwrite', '--bs', '4k', '--direct', 1,
+            '--ioengine', 'libaio', '--iodepth', '32', '--verify',
+            'crc32c', '--verify_dump', 1, '--continue_on_error', 'verify',
+            '--loops', 4, '--runtime', 0, '--time_based', '--name',
+            '/dev/sdj', '--name', '/dev/hdaa')
+
+    @mock.patch.object(hardware, 'list_all_block_devices', autospec=True)
+    def test_fio_disk_no_default(self, mock_list, mock_execute):
+
+        node = {'driver_info': {'agent_burnin_fio_disk_runtime': 600,
+                                'agent_burnin_fio_disk_loops': 5}}
+
+        mock_list.return_value = [
+            hardware.BlockDevice('/dev/sdj', 'big', 1073741824, True),
+            hardware.BlockDevice('/dev/hdaa', 'small', 65535, False),
+        ]
+        mock_execute.return_value = (['out', 'err'])
+
+        burnin.fio_disk(node)
+
+        mock_execute.assert_called_once_with(
+            'fio', '--rw', 'readwrite', '--bs', '4k', '--direct', 1,
+            '--ioengine', 'libaio', '--iodepth', '32', '--verify',
+            'crc32c', '--verify_dump', 1, '--continue_on_error', 'verify',
+            '--loops', 5, '--runtime', 600, '--time_based', '--name',
+            '/dev/sdj', '--name', '/dev/hdaa')
+
+    @mock.patch.object(hardware, 'list_all_block_devices', autospec=True)
+    def test_fio_disk_no_fio(self, mock_list, mock_execute):
+
+        node = {'driver_info': {}}
+        mock_execute.side_effect = (['out', 'err'],
+                                    processutils.ProcessExecutionError())
+
+        burnin.fio_disk(node)
+
+        self.assertRaises(errors.CommandExecutionError,
+                          burnin.fio_disk, node)

ironic_python_agent/tests/unit/test_hardware.py

@@ -157,6 +157,13 @@ class TestGenericHardwareManager(base.IronicAgentTest):
                 'reboot_requested': False,
                 'abortable': True
             },
+            {
+                'step': 'burnin_disk',
+                'priority': 0,
+                'interface': 'deploy',
+                'reboot_requested': False,
+                'abortable': True
+            },
             {
                 'step': 'burnin_memory',
                 'priority': 0,

releasenotes/notes/add_burnin_disk-12adb5735a41af47.yaml

@@ -0,0 +1,6 @@
+---
+features:
+  - |
+    Adds a burn-in cleaning step 'burnin_disk' to stress test disks for a
+    configurable number of iterations or a configurable amount of time with
+    fio. To use this step, stress-ng needs to be installed on the RAM disk.