openstack ice resize 詳解(一)

感謝朋友支持本博客，歡迎共同探討交流，因爲能力和時間有限，錯誤之處在所不免，歡迎指正！

若有轉載，請保留源做者博客信息。

Better Me的博客：blog.csdn.net/tantexian

如需交流，歡迎你們博客留言。

一：resize總結

    總結，寫在最前面，具體分析，請參照後續步驟。

一、resize與migrate底層接口一致，倘若前端傳入了新的flavor，則是resize，該該新的flavor傳入底層。遷移傳入底層的flavor則爲自身實例相同的flavor。底層根據傳入進來的flavor參數走相同的邏輯。resize與migrate的區別是在遷移的同時，是否改變虛擬機的flavor。 二、resize底層與冷遷移公用接口，所以遷移過程當中會關閉實例。（注：在confirm_resize過程則不會關閉實例）實質是：當resize完成，新虛擬機已經遷移完畢，且能對外提供服務。後續的確認confirm_resize只是清理遷移數據（清理完畢則不能回滾revert_resize操做）（注：熱遷移：整個過程不須要關機） 三、resize/migrate以後，會在數據庫表migration表中增長記錄，同時會增長instance_system_metadata表記錄，記錄老的flavor和新的flavor，若是是resize，也會更新instances表中虛擬機的flavor信息。 四、配置項allow_resize_to_same_host表示是否容許遷移到本機，默認是False。 五、配置項resize_confirm_window>0則表示超過resize_confirm_window時間後倘若用戶沒有確認resize，則系統會自動確認。 六、resize涉及技術比較多，涉及到底層網絡、磁盤遷移、調度算法等，所以分析流程比較複雜繁瑣。因此本文只能是儘量的將重點邏輯代碼進行了講解。

2、resize代碼詳解分析

一、/nova/api/openstack/compute/servers.py

@wsgi.response(202)     @wsgi.serializers(xml=FullServerTe-plate)     @wsgi.deserializers(xml=ActionDeserializer)     @wsgi.action('resize')     def _action_resize(self, req, id, body):         """Resizes a given instance to the flavor size requested."""         try:             flavor_ref = str(body["resize"]["flavorRef"]) #獲取升級flavor             if not flavor_ref:                 msg = _("Resize request has invalid 'flavorRef' attribute.")                 raise exc.HTTPBadRequest(explanation=msg)         except (KeyError, TypeError):             msg = _("Resize requests require 'flavorRef' attribute.")             raise exc.HTTPBadRequest(explanation=msg)         kwargs = {}         if 'auto_disk_config' in body['resize']:             kwargs['auto_disk_config'] = body['resize']['auto_disk_config']         return self. _resize (req, id, flavor_ref, **kwargs) #跟進到2

二、 /nova/api/openstack/compute/servers.py

def _resize(self, req, instance_id, flavor_id, **kwargs):         """Begin the resize process with given instance/flavor."""         context = req.environ["nova.context"]         instance = self._get_server(context, req, instance_id)  #獲取虛擬機數據庫信息         try:             #傳入虛擬機instance 、將要升級的flavor_id              self.compute_api.resize (context, instance, flavor_id, **kwargs)#跟進到3         except exception.QuotaError as error:             raise exc.HTTPRequestEntityTooLarge(                 explanation=error.format_message(),                 headers={'Retry-After': 0})         except exception.FlavorNotFound:             msg = _("Unable to locate requested flavor.")             raise exc.HTTPBadRequest(explanation=msg)         except exception.CannotResizeToSameFlavor:             msg = _("Resize requires a flavor change.")             raise exc.HTTPBadRequest(explanation=msg)         except exception.InstanceIsLocked as e:             raise exc.HTTPConflict(explanation=e.format_message())         except exception.InstanceInvalidState as state_error:             common.raise_http_conflict_for_instance_invalid_state(state_error,                     'resize')         except exception.ImageNotAuthorized:             msg = _("You are not authorized to access the image "                     "the instance was started with.")             raise exc.HTTPUnauthorized(explanation=msg)         except exception.ImageNotFound:             msg = _("Image that the instance was started "                     "with could not be found.")             raise exc.HTTPBadRequest(explanation=msg)         except exception.Invalid:             msg = _("Invalid instance image.")             raise exc.HTTPBadRequest(explanation=msg)         return webob.Response(status_int=202)

三、/nova/compute/api.py

@wrap_check_policy     @check_instance_lock     @check_instance_cell     @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED],                           task_state=[None])     def resize(self, context, instance, flavor_id=None,                **extra_instance_updates):         """Resize (ie, migrate) a running instance.          #若是 flavor一致，不然爲resize         If flavor_id is None, the process is considered a migration, keeping         the original flavor_id. If flavor_id is not None, the instance should         be migrated to a new host and resized to the new flavor_id.         """         self._check_auto_disk_config(instance, **extra_instance_updates)         current_instance_type = flavors.extract_flavor(instance)         # If flavor_id is not provided, only migrate the instance. #若是前端傳遞flavor_id，則new_instance_type 爲該flavor_id type #不然爲instance自身type         if not flavor_id:              LOG.debug(_("flavor_id is None. Assuming migration."),                       instance=instance)             new_instance_type = current_instance_type         else:             new_instance_type = flavors.get_flavor_by_flavor_id(                     flavor_id, read_deleted="no")         current_instance_type_name = current_instance_type['name']         new_instance_type_name = new_instance_type['name']         LOG.debug(_("Old instance type %(current_instance_type_name)s, "                     " new instance type %(new_instance_type_name)s"),                   {'current_instance_type_name': current_instance_type_name,                    'new_instance_type_name': new_instance_type_name},                   instance=instance)         if not new_instance_type:             raise exception.FlavorNotFound(flavor_id=flavor_id)         same_instance_type = (current_instance_type['id'] ==                               new_instance_type['id'])         # NOTE(sirp): We don't want to force a customer to change their flavor         # when Ops is migrating off of a failed host.         if not same_instance_type and new_instance_type.get('disabled'):             raise exception.FlavorNotFound(flavor_id=flavor_id)         if same_instance_type and flavor_id and self.cell_type != 'compute':             raise exception.CannotResizeToSameFlavor()         # ensure there is sufficient headroom for upsizes         deltas = self._upsize_quota_delta(context, new_instance_type,                                           current_instance_type)         try: #更新配置相關信息             project_id, user_id = quotas_obj.ids_from_instance(context,                                                                instance)             quotas = self._reserve_quota_delta(context, deltas,                                                project_id=project_id)         except exception.OverQuota as exc:             quotas = exc.kwargs['quotas']             overs = exc.kwargs['overs']             headroom = exc.kwargs['headroom']             resource = overs[0]             used = quotas[resource] - headroom[resource]             total_allowed = used + headroom[resource]             overs = ','.join(overs)             LOG.warn(_("%(overs)s quota exceeded for %(pid)s,"                        " tried to resize instance."),                      {'overs': overs, 'pid': context.project_id})             raise exception.TooManyInstances(overs=overs,                                              req=deltas[resource],                                              used=used, allowed=total_allowed,                                              resource=resource) #將instance task_state 狀態設置爲RESIZE_PREP         instance.task_state = task_states.RESIZE_PREP         instance.progress = 0         instance.update(extra_instance_updates)         instance.save(expected_task_state=[None]) #更新實例狀態到數據庫         filter_properties = {'ignore_hosts': []} #調度用的忽略列表 #若是nova.conf中allow_resize_to_same_host==false，則將自身host添加到 #ignore_hosts列表中，後續調度則不會將給實例resize到本機         if not CONF.allow_resize_to_same_host:             filter_properties['ignore_hosts'].append(instance['host'])         # Here when flavor_id is None, the process is considered as migrate. #此處原理同上，針對遷移的配置         if (not flavor_id and not CONF.allow_migrate_to_same_host):             filter_properties['ignore_hosts'].append(instance['host'])         if self.cell_type == 'api':             # Commit reservations early and create migration record.             self._resize_cells_support(context, quotas, instance,                                        current_instance_type,                                        new_instance_type) #將instance的一些信息及狀態，保存到數據庫 表instance_actions中         self._record_action_start(context, instance, instance_actions.RESIZE)         scheduler_hint = {'filter_properties': filter_properties}#構造後續調度用參數         self. compute_task_api.resize_instance (context, instance,#跟進到4                 extra_instance_updates, scheduler_hint=scheduler_hint,                 flavor=new_instance_type,                 reservations=quotas.reservations or [])

跟進代碼可知上述：compute_task_api==conductor.ComputeTaskAPI()

四、/nova/conductor/api.py

def resize_instance(self, context, instance, extra_instance_updates,                         scheduler_hint, flavor, reservations):         # NOTE(comstud): 'extra_instance_updates' is not used here but is         # needed for compatibility with the cells_rpcapi version of this         # method.         self. conductor_compute_rpcapi.migrate_server (  #跟進到5             context, instance, scheduler_hint, False, False, flavor,             None, None, reservations)

五、/nova/conductor/rpcapi.py

def migrate_server(self, context, instance, scheduler_hint, live, rebuild,                   flavor, block_migration, disk_over_commit,                   reservations=None): #此處由4傳入的參數live==false，非熱遷移         if self.client.can_send_version('1.6'):             version = '1.6'         else:             instance = jsonutils.to_primitive(                     objects_base.obj_to_primitive(instance))             version = '1.4'         flavor_p = jsonutils.to_primitive(flavor)         cctxt = self.client.prepare(version=version)         return cctxt.call(context, ' migrate_server ',  #rpc調用，跟進到6                           instance=instance, scheduler_hint=scheduler_hint,                           live=live, rebuild=rebuild, flavor=flavor_p,                           block_migration=block_migration,                           disk_over_commit=disk_over_commit,                           reservations=reservations)

六、 /nova/conductor/manage.py

@messaging.expected_exceptions(exception.NoValidHost,                                    exception.ComputeServiceUnavailable,                                    exception.InvalidHypervisorType,                                    exception.UnableToMigrateToSelf,                                    exception.DestinationHypervisorTooOld,                                    exception.InvalidLocalStorage,                                    exception.InvalidSharedStorage,                                    exception.MigrationPreCheckError)     def migrate_server(self, context, instance, scheduler_hint, live, rebuild,             flavor, block_migration, disk_over_commit, reservations=None):         if instance and not isinstance(instance, instance_obj.Instance):             # NOTE(danms): Until v2 of the RPC API, we need to tolerate             # old-world instance objects here             attrs = ['metadata', 'system_metadata', 'info_cache',                      'security_groups']             instance = instance_obj.Instance._from_db_object(                 context, instance_obj.Instance(), instance,                 expected_attrs=attrs)         if live and not rebuild and not flavor:         #在線遷移、不是重建且沒有傳遞flavor（即不是resize），走在線遷移             self._live_migrate(context, instance, scheduler_hint,                                block_migration, disk_over_commit)         elif not live and not rebuild and flavor:  #若是是resize或者冷遷移             instance_uuid = instance['uuid']             with compute_utils.EventReporter(context, self.db,                                          'cold_migrate', instance_uuid):                 self. _cold_migrate (context, instance, flavor,  #跟進到7                                    scheduler_hint['filter_properties'],                                    reservations)         else:             raise NotImplementedError()

注：從上述代碼能夠看到resize和冷遷移在此處往下走的同一個邏輯代碼

七、 nova/conductor/manage.py

def _cold_migrate(self, context, instance, flavor, filter_properties,                       reservations):         image_ref = instance.image_ref #獲取image         image = compute_utils.get_image_metadata(             context, self.image_service, image_ref, instance)         request_spec = scheduler_utils.build_request_spec(             context, image, [instance], instance_type=flavor)         quotas = quotas_obj.Quotas.from_reservations(context,                                                      reservations,                                                      instance=instance)         try:             #此處爲重點代碼，調度選擇出合適的hosts             hosts =  self.scheduler_rpcapi.select_destinations( #代碼 跟進到7-1                     context, request_spec, filter_properties)             #從hosts中選擇第一個host進行冷遷移（第一個即爲權重最大者）             host_state = hosts[0]         except exception.NoValidHost as ex:             vm_state = instance['vm_state']             if not vm_state:                 vm_state = vm_states.ACTIVE             updates = {'vm_state': vm_state, 'task_state': None}             self._set_vm_state_and_notify(context, 'migrate_server',                                           updates, ex, request_spec)             quotas.rollback() #出現異常則回滾，本博文此處不作更進一步講解             LOG.warning(_("No valid host found for cold migrate"),                         instance=instance)             return         try:             scheduler_utils.populate_filter_properties(filter_properties,                                                        host_state)             # context is not serializable             filter_properties.pop('context', None)             # TODO(timello): originally, instance_type in request_spec             # on compute.api.resize does not have 'extra_specs', so we             # remove it for now to keep tests backward compatibility.             request_spec['instance_type'].pop('extra_specs')             (host, node) = (host_state['host'], host_state['nodename'])             self. compute_rpcapi.prep_resize(  #resize的前序工做，跟進代碼至8                 context, image, instance,                 flavor, host,                 reservations, request_spec=request_spec,                 filter_properties=filter_properties, node=node)         except Exception as ex:             with excutils.save_and_reraise_exception():                 updates = {'vm_state': instance['vm_state'],                            'task_state': None}                 self._set_vm_state_and_notify(context, 'migrate_server',                                               updates, ex, request_spec)                 quotas.rollback()#出現異常則回滾，本博文此處不作更進一步講解

7-1 、/nova/scheduler/rpcapi.py

def select_destinations(self, ctxt, request_spec, filter_properties):         cctxt = self.client.prepare()         return  cctxt.call(ctxt, 'select_destinations ',  #遠程調用跟進到7-2             request_spec=request_spec, filter_properties=filter_properties)

7-2 、 /nova/scheduler/manage.py

@messaging.expected_exceptions(exception.NoValidHost)     def select_destinations(self, context, request_spec, filter_properties):         """Returns destinations(s) best suited for this request_spec and         filter_properties.         The result should be a list of dicts with 'host', 'nodename' and         'limits' as keys.         """         dests =  self.driver.select_destinations (context, request_spec,  #跟進到7-3             filter_properties)         return jsonutils.to_primitive(dests)

注：根據下述代碼，能夠知道 driver == nova.scheduler.filter_scheduler.FilterScheduler

        self.driver = importutils.import_object(scheduler_driver)

        scheduler_driver = CONF.scheduler_driver

        cfg.StrOpt('scheduler_driver',

               default='nova.scheduler.filter_scheduler.FilterScheduler',

               help='Default driver to use for the scheduler'),

7-3 、/ nova/scheduler/filter_scheduler.py

def select_destinations(self, context, request_spec, filter_properties):         """Selects a filtered set of hosts and nodes."""         num_instances = request_spec['num_instances']         instance_uuids = request_spec.get('instance_uuids')         selected_hosts = self. _schedule (context, request_spec,  #代碼跟進到7-4                                         filter_properties, instance_uuids)         # Couldn't fulfill the request_spec         if len(selected_hosts) < num_instances:             raise exception.NoValidHost(reason='')         dests = [dict(host=host.obj.host, nodename=host.obj.nodename,                       limits=host.obj.limits) for host in selected_hosts]         return dests

7-4 、/ nova/scheduler/filter_scheduler.py

下述代碼爲調度出最合適selected_hosts，更多進一步詳解，請參考博文：nova-scheduler詳解 openstack-ice版

所以調度代碼則再也不作進一步詳解。

def _schedule(self, context, request_spec, filter_properties,                   instance_uuids=None):         """Returns a list of hosts that meet the required specs,         ordered by their fitness.         """         elevated = context.elevated()         instance_properties = request_spec['instance_properties']         instance_type = request_spec.get("instance_type", None)         update_group_hosts = self._setup_instance_group(context,                 filter_properties)         config_options = self._get_configuration_options()         # check retry policy.  Rather ugly use of instance_uuids[0]...         # but if we've exceeded max retries... then we really only         # have a single instance.         properties = instance_properties.copy()         if instance_uuids:             properties['uuid'] = instance_uuids[0]         self._populate_retry(filter_properties, properties)         filter_properties.update({'context': context,                                   'request_spec': request_spec,                                   'config_options': config_options,                                   'instance_type': instance_type})         self.populate_filter_properties(request_spec,                                         filter_properties)         # Find our local list of acceptable hosts by repeatedly         # filtering and weighing our options. Each time we choose a         # host, we virtually consume resources on it so subsequent         # selections can adjust accordingly.         # Note: remember, we are using an iterator here. So only         # traverse this list once. This can bite you if the hosts         # are being scanned in a filter or weighing function.         hosts = self._get_all_host_states(elevated)         selected_hosts = []         if instance_uuids:             num_instances = len(instance_uuids)         else:             num_instances = request_spec.get('num_instances', 1) &n"sp;       for num in xrange(num_instances):             # Fil4er local hosts based on requirements ...             hosts = self.host_manager.get_filtered_hosts(hosts,                     filter_properties, index=num)             if not hosts:                 # Can't get any more locally.                 break             LOG.debug(_("Filtered %(hosts)s"), {'hosts': hosts})             weighed_hosts = self.host_manager.get_weighed_hosts(hosts,                     filter_properties)             LOG.debug(_("Weighed %(hosts)s"), {'hosts': weighed_hosts})             scheduler_host_subset_size = CONF.scheduler_host_subset_size             if scheduler_host_subset_size > len(weighed_hosts):                 scheduler_host_subset_size = len(weighed_hosts)             if scheduler_host_subset_size < 1:                 scheduler_host_subset_size = 1             chosen_host = random.choice(                 weighed_hosts[0:scheduler_host_subset_size])             selected_hosts.append(chosen_host)             # Now consume the resources so the filter/weights             # will change for the next instance.             chosen_host.obj.consume_from_instance(instance_properties)             if update_group_hosts is True:                 filter_properties['group_hosts'].add(chosen_host.obj.host)         return selected_hosts

八、/nova/compute/rpcapi.py

def prep_resize(self, ctxt, image, instance, instance_type, host,                     reservations=None, request_spec=None,                     filter_properties=None, node=None):         # NOTE(russellb) Havana compat         version = self._get_compat_version('3.0', '2.43')         instance_type_p = jsonutils.to_primitive(instance_type)         image_p = jsonutils.to_primitive(image)         cctxt = self.client.prepare(server=host, version=version)          cctxt.cast(ctxt, 'prep_resize' , #rpc調用，跟進代碼至9                    instance=instance,                    instance_type=instance_type_p,                    image=image_p, reservations=reservations,                    request_spec=request_spec,                    filter_properties=filter_properties,                    node=node)

因爲csdn限制了文章字數，所以接下來的步驟，請參考博文： openstack ice resize 詳解(二)