Enable start/stop of instances with accelerators.

. Do not delete accelerator requests in stop code paths.
. In the start code path, get the list of accelerator requests from
  Cyborg in the compute manager 'power_on'.
. Pass accel_info (said list) to the virt driver power_on.
. In libvirt driver, pass that accel_info to driver power_on.

Change-Id: I8c94504b87aa4450d163fe2b33f6aa0eb5dae5ff
Blueprint: nova-cyborg-interaction

nova/compute/manager.py

@@ -3099,9 +3099,10 @@ def _power_on(self, context, instance):
         network_info = self.network_api.get_instance_nw_info(context, instance)
         block_device_info = self._get_instance_block_device_info(context,
                                                                  instance)
+        accel_info = self._get_accel_info(context, instance)
         self.driver.power_on(context, instance,
                              network_info,
-                             block_device_info)
+                             block_device_info, accel_info)
 
     def _delete_snapshot_of_shelved_instance(self, context, instance,
                                              snapshot_id):

nova/tests/functional/test_servers.py

@@ -7782,8 +7782,10 @@ def setUp(self):
         self._setup_compute_nodes_and_device_rps()
 
     def _setup_compute_nodes_and_device_rps(self):
+        self.compute_services = []
         for i in range(self.NUM_HOSTS):
-            self._start_compute(host='accel_host' + str(i))
+            svc = self._start_compute(host='accel_host' + str(i))
+            self.compute_services.append(svc)
         self.compute_rp_uuids = [
            rp['uuid'] for rp in self._get_all_providers()
            if rp['uuid'] == rp['root_provider_uuid']]
@@ -7947,6 +7949,30 @@ def throw_error(*args, **kwargs):
         # Verify that no allocations/usages remain after deletion
         self._check_no_allocs_usage(server_uuid)
 
+    def test_create_server_with_local_delete(self):
+        """Delete the server when compute service is down."""
+        server = self._get_server()
+        server_uuid = server['id']
+
+        # Stop the server.
+        self.api.post_server_action(server_uuid, {'os-stop': {}})
+        self._wait_for_state_change(server, 'SHUTOFF')
+        self._check_allocations_usage(server)
+        # Stop and force down the compute service.
+        compute_id = self.admin_api.get_services(
+            host='accel_host0', binary='nova-compute')[0]['id']
+        self.compute_services[0].stop()
+        self.admin_api.put_service(compute_id, {'forced_down': 'true'})
+
+        # Delete the server with compute service down.
+        self.api.delete_server(server_uuid)
+        self.cyborg.mock_del_arqs.assert_called_once_with(server_uuid)
+        self._check_no_allocs_usage(server_uuid)
+
+        # Restart the compute service to see if anything fails.
+        self.admin_api.put_service(compute_id, {'forced_down': 'false'})
+        self.compute_services[0].start()
+
 
 class AcceleratorServerReschedTest(AcceleratorServerBase):
 

nova/tests/unit/compute/test_compute.py

@@ -2493,7 +2493,7 @@ def test_power_on(self):
         called = {'power_on': False}
 
         def fake_driver_power_on(self, context, instance, network_info,
-                                 block_device_info):
+                                 block_device_info, accel_device_info=None):
             called['power_on'] = True
 
         self.stub_out('nova.virt.fake.FakeDriver.power_on',
@@ -2512,6 +2512,25 @@ def fake_driver_power_on(self, context, instance, network_info,
         self.assertTrue(called['power_on'])
         self.compute.terminate_instance(self.context, inst_obj, [])
 
+    @mock.patch.object(compute_manager.ComputeManager,
+                       '_get_instance_block_device_info')
+    @mock.patch('nova.network.neutron.API.get_instance_nw_info')
+    @mock.patch.object(fake.FakeDriver, 'power_on')
+    @mock.patch('nova.accelerator.cyborg._CyborgClient.get_arqs_for_instance')
+    def test_power_on_with_accels(self, mock_get_arqs,
+            mock_power_on, mock_nw_info, mock_blockdev):
+        instance = self._create_fake_instance_obj()
+        instance.flavor.extra_specs = {'accel:device_profile': 'mydp'}
+        accel_info = [{'k1': 'v1', 'k2': 'v2'}]
+        mock_get_arqs.return_value = accel_info
+        mock_nw_info.return_value = 'nw_info'
+        mock_blockdev.return_value = 'blockdev_info'
+
+        self.compute._power_on(self.context, instance)
+        mock_get_arqs.assert_called_once_with(instance['uuid'])
+        mock_power_on.assert_called_once_with(self.context,
+            instance, 'nw_info', 'blockdev_info', accel_info)
+
     def test_power_off(self):
         # Ensure instance can be powered off.
 

nova/virt/driver.py

@@ -889,10 +889,14 @@ def power_off(self, instance, timeout=0, retry_interval=0):
         raise NotImplementedError()
 
     def power_on(self, context, instance, network_info,
-                 block_device_info=None):
+                 block_device_info=None, accel_info=None):
         """Power on the specified instance.
 
         :param instance: nova.objects.instance.Instance
+        :param network_info: instance network information
+        :param block_device_info: instance volume block device info
+        :param accel_info: List of accelerator request dicts. The exact
+            data struct is doc'd in nova/virt/driver.py::spawn().
         """
         raise NotImplementedError()
 

nova/virt/fake.py

@@ -276,7 +276,7 @@ def power_off(self, instance, timeout=0, retry_interval=0):
             raise exception.InstanceNotFound(instance_id=instance.uuid)
 
     def power_on(self, context, instance, network_info,
-                 block_device_info=None):
+                 block_device_info=None, accel_info=None):
         if instance.uuid in self.instances:
             self.instances[instance.uuid].state = power_state.RUNNING
         else:

nova/virt/hyperv/driver.py

@@ -224,7 +224,7 @@ def power_off(self, instance, timeout=0, retry_interval=0):
         self._vmops.power_off(instance, timeout, retry_interval)
 
     def power_on(self, context, instance, network_info,
-                 block_device_info=None):
+                 block_device_info=None, accel_info=None):
         self._vmops.power_on(instance, block_device_info, network_info)
 
     def resume_state_on_host_boot(self, context, instance, network_info,

nova/virt/ironic/driver.py

@@ -1474,7 +1474,7 @@ def power_off(self, instance, timeout=0, retry_interval=0):
                  node.uuid, instance=instance)
 
     def power_on(self, context, instance, network_info,
-                 block_device_info=None):
+                 block_device_info=None, accel_info=None):
         """Power on the specified instance.
 
         NOTE: Unlike the libvirt driver, this method does not delete
@@ -1486,7 +1486,8 @@ def power_on(self, context, instance, network_info,
             this driver.
         :param block_device_info: Instance block device
             information. Ignored by this driver.
-
+        :param accel_info: List of accelerator requests for this instance.
+                           Ignored by this driver.
         """
         LOG.debug('Power on called for instance', instance=instance)
         node = self._validate_instance_and_node(instance)

nova/virt/libvirt/driver.py

@@ -3297,12 +3297,13 @@ def power_off(self, instance, timeout=0, retry_interval=0):
         self._destroy(instance)
 
     def power_on(self, context, instance, network_info,
-                 block_device_info=None):
+                 block_device_info=None, accel_info=None):
         """Power on the specified instance."""
         # We use _hard_reboot here to ensure that all backing files,
         # network, and block device connections, etc. are established
         # and available before we attempt to start the instance.
-        self._hard_reboot(context, instance, network_info, block_device_info)
+        self._hard_reboot(context, instance, network_info, block_device_info,
+                          accel_info)
 
     def trigger_crash_dump(self, instance):
 

nova/virt/powervm/driver.py

@@ -464,7 +464,7 @@ def power_off(self, instance, timeout=0, retry_interval=0):
                      timeout=timeout)
 
     def power_on(self, context, instance, network_info,
-                 block_device_info=None):
+                 block_device_info=None, accel_info=None):
         """Power on the specified instance.
 
         :param instance: nova.objects.instance.Instance

nova/virt/vmwareapi/driver.py

@@ -658,7 +658,7 @@ def power_off(self, instance, timeout=0, retry_interval=0):
         self._vmops.power_off(instance, timeout, retry_interval)
 
     def power_on(self, context, instance, network_info,
-                 block_device_info=None):
+                 block_device_info=None, accel_info=None):
         """Power on the specified instance."""
         self._vmops.power_on(instance)
 

nova/virt/xenapi/driver.py

@@ -331,7 +331,7 @@ def power_off(self, instance, timeout=0, retry_interval=0):
         self._vmops.power_off(instance)
 
     def power_on(self, context, instance, network_info,
-                 block_device_info=None):
+                 block_device_info=None, accel_info=None):
         """Power on the specified instance."""
         self._vmops.power_on(instance)
 

nova/virt/zvm/driver.py

@@ -395,7 +395,7 @@ def power_off(self, instance, timeout=0, retry_interval=0):
             self._hypervisor.guest_softstop(instance.name)
 
     def power_on(self, context, instance, network_info,
-                 block_device_info=None):
+                 block_device_info=None, accel_info=None):
         self._hypervisor.guest_start(instance.name)
 
     def pause(self, instance):