Burn-in: Add network step

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

Story: #2007523
Task: #42385

Change-Id: I2861696740b2de9ec38f7e9fc2c5e448c009d0bf

6 files changed, 179 insertions, 0 deletions

diff --git a/doc/source/admin/hardware_managers.rst b/doc/source/admin/hardware_managers.rst
index 4228dd85..290fcf0f 100644
--- a/doc/source/admin/hardware_managers.rst
+++ b/doc/source/admin/hardware_managers.rst
@@ -82,6 +82,9 @@ Clean steps
 ``deploy.burnin_memory``
     Stress-test the memory of a node via stress-ng for a configurable
     amount of time. Disabled by default.
+``deploy.burnin_network``
+    Stress-test the network of a pair of nodes via fio for a configurable
+    amount of time. Disabled by default.
 ``deploy.erase_devices``
     Securely erases all information from all recognized disk devices.
     Relatively fast when secure ATA erase is available, otherwise can take
diff --git a/ironic_python_agent/burnin.py b/ironic_python_agent/burnin.py
index 77f83ac4..375f11d3 100644
--- a/ironic_python_agent/burnin.py
+++ b/ironic_python_agent/burnin.py
@@ -10,6 +10,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import time
+
 from ironic_lib import utils
 from oslo_concurrency import processutils
 from oslo_log import log
@@ -19,6 +21,9 @@ from ironic_python_agent import hardware
 
 LOG = log.getLogger(__name__)
 
+NETWORK_BURNIN_ROLES = frozenset(['writer', 'reader'])
+NETWORK_READER_CYCLE = 30
+
 
 def stress_ng_cpu(node):
     """Burn-in the CPU with stress-ng
@@ -115,3 +120,72 @@ def fio_disk(node):
                      {'err': e})
         LOG.error(error_msg)
         raise errors.CommandExecutionError(error_msg)
+
+
+def _do_fio_network(writer, runtime, partner):
+
+    args = ['fio', '--ioengine', 'net', '--port', '9000', '--fill_device', 1,
+            '--group_reporting', '--gtod_reduce', 1, '--numjobs', 16]
+    if writer:
+        xargs = ['--name', 'writer', '--rw', 'write', '--runtime', runtime,
+                 '--time_based', '--listen']
+    else:
+        xargs = ['--name', 'reader', '--rw', 'read', '--hostname', partner]
+    args.extend(xargs)
+
+    while True:
+        LOG.info('Burn-in fio network command: %s', ' '.join(map(str, args)))
+        try:
+            out, err = utils.execute(*args)
+            # fio reports on stdout
+            LOG.info(out)
+            break
+        except (processutils.ProcessExecutionError, OSError) as e:
+            error_msg = ("fio (network) failed with error %(err)s",
+                         {'err': e})
+            LOG.error(error_msg)
+            # while the writer blocks in fio, the reader fails with
+            # 'Connection {refused, timeout}' errors if the partner
+            # is not ready, so we need to wait explicitly
+            if not writer and 'Connection' in str(e):
+                LOG.info("fio (network): reader retrying in %s seconds ...",
+                         NETWORK_READER_CYCLE)
+                time.sleep(NETWORK_READER_CYCLE)
+            else:
+                raise errors.CommandExecutionError(error_msg)
+
+
+def fio_network(node):
+    """Burn-in the network with fio
+
+    Run an fio network job for a pair of nodes for a configurable
+    amount of time. The pair is statically defined in driver_info
+    via 'agent_burnin_fio_network_config'.
+    The writer will wait for the reader to connect, then write to the
+    network. Upon completion, the roles are swapped.
+
+    Note (arne_wiebalck): Initial version. The plan is to make the
+                          match making dynamic by posting availability
+                          on a distributed backend, e.g. via tooz.
+
+    :param node: Ironic node object
+    :raises: CommandExecutionError if the execution of fio fails.
+    :raises: CleaningError if the configuration is incomplete.
+    """
+
+    info = node.get('driver_info', {})
+    runtime = info.get('agent_burnin_fio_network_runtime', 21600)
+
+    # get our role and identify our partner
+    config = info.get('agent_burnin_fio_network_config')
+    if not config:
+        error_msg = ("fio (network) failed to find "
+                     "'agent_burnin_fio_network_config' in driver_info")
+        raise errors.CleaningError(error_msg)
+    LOG.debug("agent_burnin_fio_network_config is %s", str(config))
+    role = config.get('role')
+    partner = config.get('partner')
+
+    _do_fio_network(role == 'writer', runtime, partner)
+    LOG.debug("fio (network): first direction done, swapping roles ...")
+    _do_fio_network(not role == 'writer', runtime, partner)
diff --git a/ironic_python_agent/hardware.py b/ironic_python_agent/hardware.py
index cc75c5ce..04426419 100644
--- a/ironic_python_agent/hardware.py
+++ b/ironic_python_agent/hardware.py
@@ -1418,6 +1418,14 @@ class GenericHardwareManager(HardwareManager):
         """
         burnin.stress_ng_vm(node)
 
+    def burnin_network(self, node, ports):
+        """Burn-in the network
+
+        :param node: Ironic node object
+        :param ports: list of Ironic port objects
+        """
+        burnin.fio_network(node)
+
     def _shred_block_device(self, node, block_device):
         """Erase a block device using shred.
 
@@ -1912,6 +1920,13 @@ class GenericHardwareManager(HardwareManager):
                 'reboot_requested': False,
                 'abortable': True
             },
+            {
+                'step': 'burnin_network',
+                'priority': 0,
+                'interface': 'deploy',
+                'reboot_requested': False,
+                'abortable': True
+            },
         ]
 
     def get_deploy_steps(self, node, ports):
diff --git a/ironic_python_agent/tests/unit/test_burnin.py b/ironic_python_agent/tests/unit/test_burnin.py
index 316aaebd..18025b41 100644
--- a/ironic_python_agent/tests/unit/test_burnin.py
+++ b/ironic_python_agent/tests/unit/test_burnin.py
@@ -144,3 +144,77 @@ class TestBurnin(base.IronicAgentTest):
 
         self.assertRaises(errors.CommandExecutionError,
                           burnin.fio_disk, node)
+
+    def test_fio_network_reader(self, mock_execute):
+
+        node = {'driver_info': {'agent_burnin_fio_network_runtime': 600,
+                                'agent_burnin_fio_network_config':
+                                    {'partner': 'host-002',
+                                     'role': 'reader'}}}
+        mock_execute.return_value = (['out', 'err'])
+
+        burnin.fio_network(node)
+
+        expected_calls = [
+            mock.call('fio', '--ioengine', 'net', '--port', '9000',
+                      '--fill_device', 1, '--group_reporting',
+                      '--gtod_reduce', 1, '--numjobs', 16, '--name',
+                      'reader', '--rw', 'read', '--hostname', 'host-002'),
+            mock.call('fio', '--ioengine', 'net', '--port', '9000',
+                      '--fill_device', 1, '--group_reporting',
+                      '--gtod_reduce', 1, '--numjobs', 16, '--name', 'writer',
+                      '--rw', 'write', '--runtime', 600, '--time_based',
+                      '--listen')]
+        mock_execute.assert_has_calls(expected_calls)
+
+    def test_fio_network_writer(self, mock_execute):
+
+        node = {'driver_info': {'agent_burnin_fio_network_runtime': 600,
+                                'agent_burnin_fio_network_config':
+                                    {'partner': 'host-001',
+                                     'role': 'writer'}}}
+        mock_execute.return_value = (['out', 'err'])
+
+        burnin.fio_network(node)
+
+        expected_calls = [
+            mock.call('fio', '--ioengine', 'net', '--port', '9000',
+                      '--fill_device', 1, '--group_reporting',
+                      '--gtod_reduce', 1, '--numjobs', 16, '--name', 'writer',
+                      '--rw', 'write', '--runtime', 600, '--time_based',
+                      '--listen'),
+            mock.call('fio', '--ioengine', 'net', '--port', '9000',
+                      '--fill_device', 1, '--group_reporting',
+                      '--gtod_reduce', 1, '--numjobs', 16, '--name',
+                      'reader', '--rw', 'read', '--hostname', 'host-001')]
+        mock_execute.assert_has_calls(expected_calls)
+
+    def test_fio_network_no_fio(self, mock_execute):
+
+        node = {'driver_info': {'agent_burnin_fio_network_config':
+                                {'partner': 'host-003', 'role': 'reader'}}}
+        mock_execute.side_effect = processutils.ProcessExecutionError('boom')
+
+        self.assertRaises(errors.CommandExecutionError,
+                          burnin.fio_network, node)
+
+    @mock.patch('time.sleep', autospec=True)
+    def test_fio_network_reader_loop(self, mock_time, mock_execute):
+
+        node = {'driver_info': {'agent_burnin_fio_network_config':
+                                {'partner': 'host-004', 'role': 'reader'}}}
+        # mock the infinite loop
+        mock_execute.side_effect = (processutils.ProcessExecutionError(
+                                    'Connection timeout'),
+                                    processutils.ProcessExecutionError(
+                                    'Connection timeout'),
+                                    processutils.ProcessExecutionError(
+                                    'Connection refused'),
+                                    ['out', 'err'],  # connected!
+                                    ['out', 'err'])  # reversed roles
+
+        burnin.fio_network(node)
+
+        # we loop 3 times, then do the 2 fio calls
+        self.assertEqual(5, mock_execute.call_count)
+        self.assertEqual(3, mock_time.call_count)
diff --git a/ironic_python_agent/tests/unit/test_hardware.py b/ironic_python_agent/tests/unit/test_hardware.py
index fbb8e650..61001260 100644
--- a/ironic_python_agent/tests/unit/test_hardware.py
+++ b/ironic_python_agent/tests/unit/test_hardware.py
@@ -170,6 +170,13 @@ class TestGenericHardwareManager(base.IronicAgentTest):
                 'interface': 'deploy',
                 'reboot_requested': False,
                 'abortable': True
+            },
+            {
+                'step': 'burnin_network',
+                'priority': 0,
+                'interface': 'deploy',
+                'reboot_requested': False,
+                'abortable': True
             }
         ]
         clean_steps = self.hardware.get_clean_steps(self.node, [])
diff --git a/releasenotes/notes/add_burnin_network-4856153d21c25f4a.yaml b/releasenotes/notes/add_burnin_network-4856153d21c25f4a.yaml
new file mode 100644
index 00000000..e197624a
--- /dev/null
+++ b/releasenotes/notes/add_burnin_network-4856153d21c25f4a.yaml
@@ -0,0 +1,6 @@
+---
+features:
+  - |
+    Adds a burn-in cleaning step 'burnin_network' to stress test the
+    network interface for a configurable amount of time with fio. To
+    use this step, fio needs to be installed on the RAM disk.