天天看點

Openstack liberty 雲主機遷移源碼分析之靜态遷移1

虛拟機遷移使資源配置更加靈活,尤其是線上遷移,提供了虛拟機的可用性和可靠性。Openstack liberty中提供了兩種類型的遷移實作:靜态遷移(cold migration)和動态遷移(live migration)。在接下來的幾篇文章中,我将詳細分析兩種遷移的實作過程,先來看靜态遷移。

限于篇幅,靜态遷移的源碼分析将包含兩篇文章:

  • 第一篇:主要介紹遷移過程中

    nova-api

    nova-conductor

    所在的工作
  • 第二篇:重點介紹

    nova-compute

    的處理過程

下面請看第一篇的内容:

發起遷移

使用者可以手動通過

nova CLI

指令行發起雲主機遷移動作:

上述指令将

id=52e4d485-6ccf-47f3-a754-b62649e7b256

的雲主機遷移到另外一個最優的

nova-compute

節點上,

--debug

選項用來顯示執行日志:

......

curl -g -i -X POST http://controller:8774/v2/eab72784b36040a186a6b88dac9ac0b2/servers/5a7d302f-f388-4ffb-af37-f1e6964b3a51/action -H "User-Agent: python-novaclient" -H "Content-Type: application/json" -H "Accept: application/json" -H "X-Auth-Token: {SHA1}8e294a111a5deaa45f6cb0f3c58a600d2b1b0493" -d '{"migrate": null}

......
           

上述截取的日志表明:

novaclient

通過http方式将遷移請求發送給

nova-api

并執行

migrate

動作(action),由

nova-api

啟動時建立的路由映射,很容易的知道,該動作的入口函數為

nova/api/openstack/compute/migrate_server.py/MigrateServerController._migrate

,下文具體分析。

源碼分析

nova-api

部分

如上分析,遷移入口如下:

#nova/api/openstack/compute/migrate_server.py/MigrateServerController._migrate, 省略裝飾器定義
def _migrate(self, req, id, body):
    """Permit admins to migrate a server to a new host.
    req 是Request對象,包含該次請求資訊
    id 是待遷移的雲主機id 如:52e4d485-6ccf-47f3-a754-b62649e7b256
    body 是該次請求的參數資訊 {"migrate": null}
    """

    #從Request對象提取請求上下文
    context = req.environ['nova.context']
    """執行權限認證,預設會通過讀取host節點/etc/nova/policy.json檔案
    中的權限規則完成認證,如果沒有定義相關的規則,則表明認證失敗抛抛異
    這裡對應的認證規則是:
    "os_compute_api:os_migrate_server:migrate": rule:admin_api"
    """
    authorize(context, action='migrate')

    #從nova資料庫中擷取id指向的雲主機資訊,傳回一個InstanceV2對象
    instance = common.get_instance(self.compute_api, context, id)
    """省略異常處理代碼
    如果雲主機不存在,找不到合适的目标主機,雲主機處于鎖定狀态,
    資源不足,雲主機狀态不對(隻能是運作或者停止态)則抛異常

    與‘調整雲主機大小’(resize)操作一樣,也是調用
    `/nova/compute/api.py/API.resize`
    執行遷移操作,resize是通過判斷
    是否指定了flavor_id參數來判斷是執行‘調整雲主機大小’還是‘遷移’操作,
    請看下文的具體分析
    """
    self.compute_api.resize(req.environ['nova.context'], instance)

---------------------------------------------------------------
#接上文:/nova/compute/api.py/API.resize, 省略裝飾器定義
def resize(self, context, instance, flavor_id=None, 
                clean_shutdown=True,
               **extra_instance_updates):
    """Resize (ie, migrate) a running instance.

    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.

    上面的注釋是說:如果flavor_id = None, 則用原有的flavor(配置)執行
    遷移操作。如果不為None,則應将雲主機遷移到新的主機并應用flavor_id指
    定的配置

    conext 請求上下文
    instance InstanceV2執行個體對象,包含雲主機的詳細配置資訊
    flavor_id 配置模闆id,這裡為None,因為是遷移操作
    clean_shutdown = True, 靜态遷移時開啟關機重試,如果未能正常關閉雲
    主機會抛異常
    """

    #檢查系統磁盤的‘自動配置磁盤’功能是否打開,否則抛異常
    #遷移完成後,雲主機需要能夠自動配置系統磁盤
    self._check_auto_disk_config(instance, **extra_instance_updates)
    #擷取雲主機配置模闆資訊
    current_instance_type = instance.get_flavor()

    # If flavor_id is not provided, only migrate the instance.
    #flavor_id = None, 執行遷移操作;列印日志并将目前配置作為遷移後雲主
    #機的配置
    if not flavor_id:
        LOG.debug("flavor_id is None. Assuming migration.",
                      instance=instance)
        new_instance_type = current_instance_type
    else:
        #從nova.instance_types資料表擷取flavor_id指定的配置模闆資訊
        #read_deleted="no",表示讀取資料庫時過濾掉已經删除的配置模闆
        new_instance_type = flavors.get_flavor_by_flavor_id(
                    flavor_id, read_deleted="no")
        #如果雲主機是從鏡像啟動的并且目前的配置模闆中root_gb(根磁盤大
        #小)不為0,而目标配置模闆中的root_gb=0,則不支援resize操作
        #因為不知道怎麼配置設定系統磁盤大小了,抛異常
        if (new_instance_type.get('root_gb') ==  and
            current_instance_type.get('root_gb') !=  and
            not self.is_volume_backed_instance(context, instance)):
            reason = _('Resize to zero disk flavor is not' 
                                                    'allowed.')
            raise exception.CannotResizeDisk(reason=reason)

    #如果沒有找到指定的配置模闆,抛異常
    if not new_instance_type:
        raise exception.FlavorNotFound(flavor_id=flavor_id)

    #列印debug日志
    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)

    #判斷是否是同一配置模闆,遷移操作中肯定是同一配置模闆
    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.
    """
    #如果是resize操作,新的配置模闆被disable了,抛異常
    if not same_instance_type and new_instance_type.get('disabled'):
            raise exception.FlavorNotFound(flavor_id=flavor_id)

    #預設cell關閉,cell_type = None
    #這裡是說resize的時候,新舊配置模闆不能是相同的,因為這樣做沒有意義
    if same_instance_type and flavor_id and 
        self.cell_type != 'compute':
        raise exception.CannotResizeToSameFlavor()

    # ensure there is sufficient headroom for upsizes
    #如果是resize操作,需要先保留資源配額
    if flavor_id:
        #擷取vcpu和memory的增量配額(如果有的話,新舊配置模闆的內插補點)
        deltas = compute_utils.upsize_quota_delta(context,
                                  new_instance_type,
                                   current_instance_type)
        try:
            #為目前使用者和項目保留資源(增量)配額,更新資料庫
            quotas = compute_utils.reserve_quota_delta(context, 
                                                        deltas,
                                                      instance)
        except exception.OverQuota as exc:
            #統計資源不足資訊,并列印日志
            quotas = exc.kwargs['quotas']
            overs = exc.kwargs['overs']
            usages = exc.kwargs['usages']
            headroom = self._get_headroom(quotas, usages, 
                                                    deltas)
            (overs, reqs, total_alloweds,
            useds) = self._get_over_quota_detail(headroom, 
                                         overs, quotas, deltas)
            LOG.warning(_LW("%(overs)s quota exceeded for %"
                    "(pid)s, tried to resize instance."),
                   {'overs': overs, 'pid': context.project_id})
            raise exception.TooManyInstances(overs=overs,
                                                 req=reqs,
                                                 used=useds,
                                      allowed=total_alloweds)
    #遷移操作,沒有額外的資源需要保留
    else:
        quotas = objects.Quotas(context=context)

    #更新與主機狀态:主機狀态:重建/遷移,任務狀态:準備重建或者遷移
    instance.task_state = task_states.RESIZE_PREP
    instance.progress = 
    instance.update(extra_instance_updates)
    instance.save(expected_task_state=[None])

    """為nova-scheduler生成過濾選項,
    CONF.allow_resize_to_same_host = true
    表示允許遷移的目的主機與源主機相同,否則過濾掉源主機
    """
    filter_properties = {'ignore_hosts': []}
    if not CONF.allow_resize_to_same_host:
        filter_properties['ignore_hosts'].append(instance.host)

    #預設cell_type = None, 
    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)

    #flavor_id = None, 執行遷移操作,否則執行resize
    #記錄執行個體操作,更新nova.instance_actions資料表,遷移結束後會更新數
    #據庫記錄,反映遷移結果
    if not flavor_id:
        self._record_action_start(context, instance,
                                      instance_actions.MIGRATE)
    else:
        self._record_action_start(context, instance,
                                      instance_actions.RESIZE)
    """将遷移請求轉發給
    `/nova/conductor/api.py/ComputeTaskAPI.resize_instance`,該
    方法直接調用
    `nova/conductor/rpcapi.py/ComputeTaskAPI.migrate_server`處理
    請求,請看下文的分析
    """
    scheduler_hint = {'filter_properties': filter_properties}
    self.compute_task_api.resize_instance(context, instance,
                extra_instance_updates, 
                scheduler_hint=scheduler_hint,
                flavor=new_instance_type,
                reservations=quotas.reservations or [],
                clean_shutdown=clean_shutdown)

------------------------------------------------------------
#接上文:`nova/conductor/rpcapi.py/ComputeTaskAPI.migrate_server`
def migrate_server(self, context, instance, scheduler_hint, 
                  live, rebuild,
                  flavor, block_migration, disk_over_commit,
                  reservations=None, clean_shutdown=True):

    """輸入參數如下:
    live = False, 靜态遷移
    rebuild = false, 遷移,而不是resize
    block_migration = None, 不是塊遷移
    disk_over_commit = None
    reservations = [] 遷移操作,沒有增量保留資源
    """
    #生成請求參數字典
    kw = {'instance': instance, 'scheduler_hint': 
                                            scheduler_hint,
          'live': live, 'rebuild': rebuild, 'flavor': flavor,
          'block_migration': block_migration,
          'disk_over_commit': disk_over_commit,
          'reservations': reservations,
          'clean_shutdown': clean_shutdown}
    #根據RPCClient的版本相容性,選擇用戶端版本。
    #在初始化rpc的時候會設定版本相容特性
    version = '1.11'
    if not self.client.can_send_version(version):
        del kw['clean_shutdown']
        version = '1.10'
    if not self.client.can_send_version(version):
        kw['flavor'] = objects_base.obj_to_primitive(flavor)
        version = '1.6'
   if not self.client.can_send_version(version):
        kw['instance'] = jsonutils.to_primitive(
                    objects_base.obj_to_primitive(instance))
        version = '1.4'
   #通過同步rpc調用将`migrate_server`消息發送給rabbitmq,
   #消費者`nova-conductor`将會收到該消息
   cctxt = self.client.prepare(version=version)
   return cctxt.call(context, 'migrate_server', **kw)
           

小結:

nova-api

主要完成執行個體狀态、相關條件檢查, 之後更新雲主機狀态及添加

nova.instance_actions

資料庫記錄,最後通過同步rpc将請求轉發給

nova-conductor

處理

nova-conductor

部分

由前述的分析,我們很容易就知道

nova-conductor

處理遷移請求的入口:

#/nova/conductor/manager.py/ComputeTaskManager.migrate_server
def migrate_server(self, context, instance, scheduler_hint, 
            live, rebuild,
            flavor, block_migration, disk_over_commit, 
            reservations=None,
            clean_shutdown=True):
    """各輸入參數來自`nova-api`,如下:
    scheduler_hint 排程選項,{u'filter_properties': 
    {u'ignore_hosts': []}}
    live = False, 靜态遷移
    rebuild = Flase, 遷移而不是調整雲主機大小
    block_migration = None, 非塊遷移
    disk_over_commit = None
    reservations = [] ,遷移操作沒有增量保留資源
    """
    #如果輸入的instance參數不是非法的NovaObject對象,就先從資料庫擷取
    #雲主機資訊,然後生成InstanceV2對象
    if instance and not isinstance(instance, nova_object.NovaObject):
        # 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 = objects.Instance._from_db_object(
                context, objects.Instance(), instance,
                expected_attrs=attrs)
    # NOTE: Remove this when we drop support for v1 of the RPC API
    #如果輸入的flavor參數不是合法的Flavor對象,就先從資料庫提取指定id
    #的配置模闆,然後生成Flavor對象
    if flavor and not isinstance(flavor, objects.Flavor):
        # Code downstream may expect extra_specs to be 
        #populated since it is receiving an object, so lookup 
        #the flavor to ensure this.
        flavor = objects.Flavor.get_by_id(context, flavor['id'])

    #動态遷移,在另外一篇文章中詳述
    if live and not rebuild and not flavor:
        self._live_migrate(context, instance, scheduler_hint,
                               block_migration, disk_over_commit)
    #調用_cold_migrate執行靜态遷移,下文具體分析
    elif not live and not rebuild and flavor:
        instance_uuid = instance.uuid
        #with語句,在遷移前記錄遷移事件記錄到資料庫
        #(nova.instance_actions_events),遷移後更新資料庫遷移記錄
        with compute_utils.EventReporter(context, 'cold_migrate',
                                             instance_uuid):
        self._cold_migrate(context, instance, flavor,
                            scheduler_hint['filter_properties'],
                                   reservations, clean_shutdown)
    #未知類型                               
    else:
        raise NotImplementedError()
-------------------------------------------------------------
#接上文:
def _cold_migrate(self, context, instance, flavor, 
                        filter_properties,
                      reservations, clean_shutdown):
    #從執行個體對象中擷取所使用的鏡像資訊,示例如下:
    """
    {u'min_disk': u'20', u'container_format': u'bare', 
    u'min_ram': u'0', u'disk_format': u'raw', 'properties': 
    {u'base_image_ref': u'e0cc468f-6501-4a85-9b19-
    70e782861387'}}
    """
    image = utils.get_image_from_system_metadata(
            instance.system_metadata)
    #通過鏡像屬性、雲主機屬性、雲主機配置模闆生成請求參數字典,格式如下:
    """
    request_spec = {
            'image': image,
            'instance_properties': instance,
            'instance_type': flavor,
            'num_instances': 1}
    """
    request_spec = scheduler_utils.build_request_spec(
            context, image, [instance], instance_type=flavor)
    #生成遷移任務對象
    #`/nova/conductor/tasks/migrate.py/MigrationTask
    task = self._build_cold_migrate_task(context, instance, 
                                            flavor,
                                             filter_properties, 
                                             request_spec,
                                             reservations, 
                                             clean_shutdown)

    """省略異常處理代碼
    如果未找到合适的目标主機,政策不合法等異常,則退出
    在退出前會更新資料庫,設定雲主機的狀态并列印日志及發送
    `compute_task.migrate_server`通知
    """
    #執行遷移,下文具體分析
    task.execute()

---------------------------------------------------------------
#接上文:`nova/conductor/tasks/migrate.py/MigrationTask._execute
def _execute(self):
    #從請求參數中擷取所使用的鏡像資訊
    image = self.request_spec.get('image')
    #根據self.reservations保留配額生成配額對象,
    #遷移操作沒有保留配額 self.reservations = []
    self.quotas = objects.Quotas.from_reservations(self.context,
                                       self.reservations,
                                     instance=self.instance)
    #添加組(group_hosts)及組政策(group_polices)資訊到過濾屬性(如果有
    #的話)
    scheduler_utils.setup_instance_group(self.context, 
                                            self.request_spec,
                                             self.filter_properties)
    """添加重試參數到過濾屬性(如果配置的重試次數 
    CONF.scheduler_max_attempts 〉1的話),修改後的過濾屬性如下:
    {'retry': {'num_attempts': 1, 'hosts': []}, 
    u'ignore_hosts': []}

    如果是`nova-compute`發送過來的重試請求,輸入的filter_properties過
    濾屬性中的retry字典中包含
    前一次請求的異常資訊,再次選擇目标主機的時候會排除`hosts`中的主機,在
    populate_retry過程中,會列印該條異常日志;如果重試超過了最大重試次
    數,也會抛異常
    """
    scheduler_utils.populate_retry(
                                    self.filter_properties,
                                       self.instance.uuid)
    #發送請求給`nova-scheduler`,根據過濾規則選擇合适的目标主機,
    #如果逾時會根據前文的重試參數重試。如果成功,傳回合适的目标主機清單
    #如果找不到合适的目标主機,抛異常
    hosts = self.scheduler_client.select_destinations(
            self.context, self.request_spec, self.filter_properties)
    #選取第一個
    host_state = hosts[]
    #添加目标主機到過濾屬性的重試清單(重試的時候'hosts'中的主機被忽
    略),示例如下:
    """
    {'retry': {'num_attempts': 1, 'hosts': [[u'devstack', 
    u'devstack']]}, 'limits': {u'memory_mb': 11733.0, 
    u'disk_gb': 1182.0}, u'ignore_hosts': []}
    """
    scheduler_utils.populate_filter_properties(
                                        self.filter_properties,
                                                   host_state)
    # context is not serializable
    self.filter_properties.pop('context', None)

    #通過異步rpc調用發送`prep_resize`消息到消息隊列,`nova-compute`會
    #處理該請求(`nova/compute/rpcapi.py/ComputeAPI`)
    (host, node) = (host_state['host'], host_state['nodename'])
    self.compute_rpcapi.prep_resize(
            self.context, image, self.instance, self.flavor, host,
            self.reservations, request_spec=self.request_spec,
            filter_properties=self.filter_properties, node=node,
            clean_shutdown=self.clean_shutdown)
           

小結:

nova-conductor

主要是借助

nova-scheduler

選擇合适的目标主機,同時也會更新

nova.instance_actions_events

資料表,最後發起異步rpc調用将遷移請求轉交給

nova-compute

處理

到這裡靜态遷移的前篇就介紹完成了,過程還是比較簡單的:主要完成一些條件判斷,更新資料庫記錄,通過

nova-scheduler

選主,最後将請求轉交給

nova-compute

處理。敬請期待:

Openstack liberty 雲主機遷移源碼分析之靜态遷移2

繼續閱讀