Add a configurable sleep before IPA starts working

dtantsur · dtantsur · commit c09753bb60be · 2016-05-24T10:39:52.000+02:00
Some kernel modules take substantial time to initialize. For example,
with mpt2sas RAID driver inspection and deployment randomly fail
due to IPA starting before the driver finishes initialization.

As much as I hate it, the only way to guarantee that the hardware is
truely initalized is to wait for it. Apparently all hardware in Linux
is treated as hotplugged, so there is no such thing as "hardware
initialization is finished". Operators can add a sleep based on their
knowledge of their hardware.

The default behaviour remains the same.

Change-Id: I0446ae81d760dacaf31eea6ad9f9eaa098cf5e93
Partial-Bug: #1582797
diff --git a/ironic_python_agent/agent.py b/ironic_python_agent/agent.py
@@ -147,7 +147,8 @@ class IronicPythonAgent(base.ExecuteCommandMixin):
 
     def __init__(self, api_url, advertise_address, listen_address,
                  ip_lookup_attempts, ip_lookup_sleep, network_interface,
-                 lookup_timeout, lookup_interval, driver_name, standalone):
+                 lookup_timeout, lookup_interval, driver_name, standalone,
+                 hardware_initialization_delay=0):
         super(IronicPythonAgent, self).__init__()
         self.ext_mgr = extension.ExtensionManager(
             namespace='ironic_python_agent.extensions',
@@ -175,6 +176,7 @@ def __init__(self, api_url, advertise_address, listen_address,
         self.ip_lookup_sleep = ip_lookup_sleep
         self.network_interface = network_interface
         self.standalone = standalone
+        self.hardware_initialization_delay = hardware_initialization_delay
 
     def get_status(self):
         """Retrieve a serializable status.
@@ -289,6 +291,12 @@ def run(self):
 
         # Cached hw managers at runtime, not load time. See bug 1490008.
         hardware.load_managers()
+        # Operator-settable delay before hardware actually comes up.
+        # Helps with slow RAID drivers - see bug 1582797.
+        if self.hardware_initialization_delay > 0:
+            LOG.info('Waiting %d seconds before proceeding',
+                     self.hardware_initialization_delay)
+            time.sleep(self.hardware_initialization_delay)
 
         if not self.standalone:
             # Inspection should be started before call to lookup, otherwise
diff --git a/ironic_python_agent/cmd/agent.py b/ironic_python_agent/cmd/agent.py
@@ -129,6 +129,11 @@
                 help='Whether to wait for all interfaces to get their IP '
                      'addresses before inspection. If set to false '
                      '(the default), only waits for the PXE interface.'),
+
+    cfg.IntOpt('hardware_initialization_delay',
+               default=APARAMS.get('ipa-hardware-initialization-delay', 0),
+               help='How much time (in seconds) to wait for hardware to '
+                    'initialize before proceeding with any actions.'),
 ]
 
 CONF.register_cli_opts(cli_opts)
@@ -153,4 +158,5 @@ def run():
                             CONF.lookup_timeout,
                             CONF.lookup_interval,
                             CONF.driver_name,
-                            CONF.standalone).run()
+                            CONF.standalone,
+                            CONF.hardware_initialization_delay).run()
diff --git a/ironic_python_agent/tests/unit/test_agent.py b/ironic_python_agent/tests/unit/test_agent.py
@@ -232,6 +232,37 @@ def test_run_with_inspection(self, mocked_list_hardware, wsgi_server_cls,
 
         self.agent.heartbeater.start.assert_called_once_with()
 
+    @mock.patch.object(time, 'sleep', autospec=True)
+    @mock.patch('wsgiref.simple_server.make_server', autospec=True)
+    @mock.patch.object(hardware.HardwareManager, 'list_hardware_info')
+    def test_run_with_sleep(self, mocked_list_hardware, wsgi_server_cls,
+                            mocked_sleep):
+        CONF.set_override('inspection_callback_url', '', enforce_type=True)
+        wsgi_server = wsgi_server_cls.return_value
+        wsgi_server.start.side_effect = KeyboardInterrupt()
+
+        self.agent.hardware_initialization_delay = 10
+        self.agent.heartbeater = mock.Mock()
+        self.agent.api_client.lookup_node = mock.Mock()
+        self.agent.api_client.lookup_node.return_value = {
+            'node': {
+                'uuid': 'deadbeef-dabb-ad00-b105-f00d00bab10c'
+            },
+            'heartbeat_timeout': 300
+        }
+        self.agent.run()
+
+        listen_addr = ('192.0.2.1', 9999)
+        wsgi_server_cls.assert_called_once_with(
+            listen_addr[0],
+            listen_addr[1],
+            self.agent.api,
+            server_class=simple_server.WSGIServer)
+        wsgi_server.serve_forever.assert_called_once_with()
+
+        self.agent.heartbeater.start.assert_called_once_with()
+        mocked_sleep.assert_called_once_with(10)
+
     def test_async_command_success(self):
         result = base.AsyncCommandResult('foo_command', {'fail': False},
                                          foo_execute)
diff --git a/releasenotes/notes/sleep-ebe58fbe07d30219.yaml b/releasenotes/notes/sleep-ebe58fbe07d30219.yaml
@@ -0,0 +1,4 @@
+---
+features:
+  - Add a configurable (and disabled by default) sleep before IPA does any
+    actions.
