探索 OpenStack 之（12）：cinder-api Service 处理 HTTP Request 的过程分析

本文是上一篇 探索 OpenStack 之（11）：cinder-api Service 启动过程分析 以及 WSGI / Paste deploy / Router 等介绍> 的后续篇。

osapi_volume 的 WSGI Service 进程在收到 HTTP Request 后，首先将HTTP request 封装成wsgi request，然后依次执行以下步骤。

1. 按顺序调用已经注册的 middleware （filter） 实例的 __call__ 方法

1.1 filter:keystonecontext

该 filter 会提取 request header 中 token 相关数据，组成一个 ctx， 放入 request.environ 的 cinder.context 项中，供后续 app 使用。

def __call__(self, req):
'keystonecontext : Make a request context from keystone headers.'
ctx = context.RequestContext(user_id,
                                     project_id,
                                     project_name=project_name,
                                     roles=roles,
                                     auth_token=auth_token,
                                     remote_address=remote_address,
                                     service_catalog=service_catalog,
                                     request_id=req_id)

req.environ['cinder.context'] = ctx #添加 cinder.context
return self.application //If you are not modifying the output, you can just return the app.

1.2 filter:authtoken

该 filter 的代码在 https://github.com/openstack/keystonemiddleware。 该 filter 会通过校验request所带的 token 来校验用户身份。当request来的时候，它根据 token 来判断这个请求是否合法。校验失败则直接返回 401 错误，校验通过则提取 token 中的信息，放入到env中，供后续的其它app使用。具体会在 token 相关的文章中详细分析。

1.3  filter:osprofiler

该 filter 的代码在 https://github.com/stackforge/osprofiler， 它用来 enables tracing for an application。似乎是在 enabled = yes 的情况下，记录 request 的如下信息：

"request": {
"host_url": request.host_url,
"path": request.path,
"query": request.query_string,
"method": request.method,
"scheme": request.scheme
}

具体细节待查。

1.4 filter:sizelimit

该 filer 会检查 request 的 conent length，如果超过 cinder.conf 中 osapi_max_request_body_size 定义的最大size，则直接报 HTTPRequestEntityTooLarge

，不再进行后续处理。

def __call__(self, req):
     if req.content_length > CONF.osapi_max_request_body_size: #判断 req.content_length，大于 CONF.osapi_max_request_body_size 则直接失败
            msg = _("Request is too large.")
            raise webob.exc.HTTPRequestEntityTooLarge(explanation=msg)
        if req.content_length is None and req.is_body_readable:
            limiter = LimitingReader(req.body_file, CONF.osapi_max_request_body_size)
            req.body_file = limiter
        return self.application 

1.5 filter:faultwrap

该 filter 不会修改 request 和 respone，而是在一旦之前的App 处理 response 出错的话，它 catch 住错误并调用 _error 方法，返回一个适当的错误信息。

def __call__(self, req):
       try:
           return req.get_response(self.application)
       except Exception as ex:
           return self._error(ex, req) //return wsgi.Fault(outer) 如果application返回response过程出错，则返回Traceback (most recent call last)

1.6 filter:request_id

该 filter 不会修改 request，而是做为第一个修改 APIRouter 生成的 response 的 filter，在 response headers里面添加 x-openstack-request-id 项，其值为一个随机的 UUID.

def __call__(self, req):
        req_id = context.generate_request_id()  //req-<UUID>
        req.environ[ENV_REQUEST_ID] = req_id //添加 'openstack.request_id'
        response = req.get_response(self.application)
        if HTTP_RESP_HEADER_REQUEST_ID not in response.headers: // 'x-openstack-request-id'
            response.headers.add(HTTP_RESP_HEADER_REQUEST_ID, req_id) #在response上添加x-openstack-request-id，比如 x-openstack-request-id: 123567890
        return response //return 修改后的 response

2. 调用已注册的 APIRouter 实例的 __call__ 方法

该方法直接返回 self._router，而它是一个 routes.middleware.RoutesMiddleware 类型的 WSGI app，所以调用其 __call__ 方法。

2.1 以 cinder list 为例分析 Core Resource 基本方法的分发过程

2.1.1 调用 routes.middleware.RoutesMiddleware 的 __call__ 方法

过程见注释部分：

def __call__(self, environ, start_response):
        """Resolves the URL in PATH_INFO, and uses wsgi.routing_args to pass on URL resolver results."""

        # Run the actual route matching
        # -- Assignment of environ to config triggers route matching
        if self.singleton: #singleton 默认值为 True，所以执行该分支
            config = request_config()
            config.mapper = self.mapper
            config.environ = environ
            match = config.mapper_dict #获取match，（{'action': u'detail', 'controller': <cinder.api.openstack.wsgi.Resource object at 0x7fa137be8950>, 'project_id': u'fa2046aaead44a698de8268f94759fc1'})
            route = config.route #获取route
        ...
        url = URLGenerator(self.mapper, environ) #比如 <routes.util.URLGenerator object at 0x7fa137a15590>
        environ['wsgiorg.routing_args'] = ((url), match)
        environ['routes.route'] = route #比如 'routes.route': <routes.route.Route object at 0x7fa137bef4d0>
        environ['routes.url'] = url #比如 'routes.url': <routes.util.URLGenerator object at 0x7fa137a15590>
        ...

        response = self.app(environ, start_response) #调用 _dispatch WSGI App
        ...
        return response

2.1.2 调用 Router.def _dispatch(req) 方法

其代码如下：

match = req.environ['wsgiorg.routing_args'][1] #match：{'action': u'detail', 'controller': <cinder.api.openstack.wsgi.Resource object at 0x7fa137be8950>, 'project_id': u'fa2046aaead44a698de8268f94759fc1'}
if not match:
     return webob.exc.HTTPNotFound()
app = match['controller'] #<cinder.api.openstack.wsgi.Resource object at 0x7fa137be8950>
return app #调用该 App 的 __call__ 方法

2.1.3 执行 cinder.api.openstack.wsgi.Resource 的 def __call__(self, request) 方法

该方法负责（反）序列化和方法分发。

# Identify the action, its arguments, and the requested content type
action_args = self.get_action_args(request.environ) #从environ 获取 match
action = action_args.pop('action', None) #得到 action 'detail'
content_type, body = self.get_body(request) #得到request 的 content_type 和 body。body 为空。
accept = request.best_match_content_type() #如果 environ 中有 'cinder.best_content_type' 的话，直接返回，比如 'application/json'；没有的话，则找个最合适的content type，并设置到environ。
return self._process_stack(request, action, action_args, content_type, body, accept) #调用 _process_stack

函数 _process_stack 的主要代码如下：

def _process_stack(self, request, action, action_args,content_type, body, accept):

#action_args：{'project_id': u'43f66bb82e684bbe9eb9ef6892bd7fd6'}， action: detail， body:， 

meth, extensions = self.get_method(request, action, content_type, body) 
#找到 action 对应的 Controller 的 method 和 这个 Core Resource 的所有带这个 aciton 的 Resource  extensions（如果 action 是 ‘action’ 的话，根据 request body 中的信息，会把 action 转化为具体的 Controller 的 method）。 
#返回值 meth： <bound method VolumeController.detail of <cinder.api.v2.volumes.VolumeController object at 0x7f2fd96863d0>>。
#返回值 extensions: [<bound method VolumeTenantAttributeController.detail of <cinder.api.contrib.volume_tenant_attribute.VolumeTenantAttributeController object at 0x7f2fd885ac50>>, <bound method VolumeReplicationController.detail of <cinder.api.contrib.volume_replication.VolumeReplicationController object at 0x7f2fd86eacd0>>, <bound method VolumeHostAttributeController.detail of <cinder.api.contrib.volume_host_attribute.VolumeHostAttributeController object at 0x7f2fd87088d0>>, <bound method VolumeMigStatusAttributeController.detail of <cinder.api.contrib.volume_mig_status_attribute.VolumeMigStatusAttributeController object at 0x7f2fd870e090>>, <bound method VolumeImageMetadataController.detail of <cinder.api.contrib.volume_image_metadata.VolumeImageMetadataController object at 0x7f2fd870e790>>]

contents = self.deserialize(meth, content_type, body) #根据 content_type 反序列化 request body
response, post = self.pre_process_extensions(extensions, request, action_args) # 找出 extensions 的 _call__ 方法，分别调用，得到 response

def pre_process_extensions(self, extensions, request, action_args)
{

   for ext in extensions:
       #遍历所有的Resource extension，比如 <bound method VolumeTenantAttributeController.detail of <cinder.api.contrib.volume_tenant_attribute.VolumeTenantAttributeController object at 0x7f2fd885ac50>>

if inspect.isgeneratorfunction(ext): #如果object是一个生成器函数(任何包含yield表达式的函数即为生成器方法), 则返回True。现在还没有这样的方法。这里会返回false
        response = None

# If it's a generator function, the part before the yield is the preprocessing stage
try:
with ResourceExceptionHandler():
gen = ext(req=request, **action_args)
response = gen.next()
except Fault as ex:
response = ex

# We had a response...
if response:
return response, []

# No response, queue up generator for post-processing
post.append(gen)

else:
# Regular functions only perform post-processing
post.append(ext) #等待执行post-processing

# Run post-processing in the reverse order

#将所有不是generatorfunciton 的 ext 放进 post， 比如 [<bound method VolumeTenantAttributeController.detail of <cinder.api.contrib.volume_tenant_attribute.VolumeTenantAttributeController object at 0x7f2fd885ac50>>, <bound method VolumeReplicationController.detail of <cinder.api.contrib.volume_replication.VolumeReplicationController object at 0x7f2fd86eacd0>>, <bound method VolumeHostAttributeController.detail of <cinder.api.contrib.volume_host_attribute.VolumeHostAttributeController object at 0x7f2fd87088d0>>, <bound method VolumeMigStatusAttributeController.detail of <cinder.api.contrib.volume_mig_status_attribute.VolumeMigStatusAttributeController object at 0x7f2fd870e090>>, <bound method VolumeImageMetadataController.detail of <cinder.api.contrib.volume_image_metadata.VolumeImageMetadataController object at 0x7f2fd870e790>>]
return None, reversed(post)

}

if not response: #response 为 null，调用 method 方法，比如 <bound method VolumeController.detail of <cinder.api.v1.volumes.VolumeController object at 0x7fa137be8650>>
    action_result = self.dispatch(meth, request, action_args) #执行Resource Controller 的基本方法，得到其返回值    

def dispatch(self, method, request, action_args):
        """Dispatch a call to the action-specific method."""
        #method: <bound method VolumeController.detail of <cinder.api.v2.volumes.VolumeController object at 0x7f2fd96863d0>>
        #action_args: {}
        #method执行结果为： {'volumes': [{'status': u'error', 'user_id': u'1dc0db32a936496ebfc50be54924a7cc', 'attachments': [], 'links': [{'href': u'http://controller:8776/v2/43f66bb82e684bbe9eb9ef6892bd7fd6/volumes/42ddb30e-8a36-4301-99e4-9547ce4e860d', 'rel': 'self'}, {'href': u'http://controller:8776/43f66bb82e684bbe9eb9ef6892bd7fd6/volumes/42ddb30e-8a36-4301-99e4-9547ce4e860d', 'rel': 'bookmark'}], 'availability_zone': u'nova', 'bootable': 'false', 'encrypted': False, 'created_at': datetime.datetime(2015, 1, 3, 2, 47, 52), 'description': None, 'volume_type': None, 'name': None, 'replication_status': u'disabled', 'consistencygroup_id': None, 'source_volid': None, 'snapshot_id': None, 'metadata': {}, 'id': u'42ddb30e-8a36-4301-99e4-9547ce4e860d', 'size': 1L}]}
        return method(req=request, **action_args)

resp_obj = ResponseObject(action_result) #初始化其返回值为一个ResponseObject对象

_set_request_id_header(request, resp_obj) #设置 headers['x-compute-request-id'] = req.environ.get('cinder.context').request_id
serializers = getattr(meth, 'wsgi_serializers', {}) #获取Controller的 action 方法使用的serializers。detail 方法的serizlizer 是 {'xml': <class 'cinder.api.v2.volumes.VolumesTemplate'>}
resp_obj._bind_method_serializers(serializers) #设置 resp_obj 的序列化方法
resp_obj.preserialize(accept, self.default_serializers) # resp_obj 做序列化准备

response = self.post_process_extensions(post, resp_obj, request, action_args) # 对每个 extension 执行操作，获取response：  response = ext(req=request, resp_obj=resp_obj, **action_args)     

def post_process_extensions(self, extensions, resp_obj, request,action_args):

for ext in extensions:

#遍历 Resource extension。以 <bound method VolumeImageMetadataController.detail of <cinder.api.contrib.volume_image_metadata.VolumeImageMetadataController object at 0x7f2fd870e790>> 为例。

response = None

if inspect.isgenerator(ext): #Return true if the object is a generator. 这里返回 false

                # If it's a generator, run the second half of
                # processing
                try:
                    with ResourceExceptionHandler():
                        response = ext.send(resp_obj)
                    except StopIteration:
                    # Normal exit of generator
                    continue
                except Fault as ex:
                    response = ex
            else:
                # Regular functions get post-processing...
                try:
                    with ResourceExceptionHandler():
                        response = ext(req=request, resp_obj=resp_obj, **action_args) #调用 extension method 的方法，它会把 reponse 插进 resp_obj。本例子中将返回none。

@wsgi.extends
    def detail(self, req, resp_obj):
        context = req.environ['cinder.context']
        if authorize(context):
            resp_obj.attach(xml=VolumesImageMetadataTemplate())
            all_meta = self._get_all_images_metadata(context)
            for vol in list(resp_obj.obj.get('volumes', [])):
                image_meta = all_meta.get(vol['id'], {})
                self._add_image_metadata(context, vol, image_meta)

                except Fault as ex:
                    response = ex

            # We had a response...
            if response:
                return response
        return None #返回none

if resp_obj and not response:

response = resp_obj.serialize(request, accept, self.default_serializers) #执行序列化，得到 response string

return response #返回response，后续需要改response的middleware的__call__方法将得到继续执行，指导回到 WSGI Server，它会将 response 返回给cinderclient

2.2 cinder quota-update 的简要过程 （os-quota-set  是Extension resource）

Request:

PUT /43f66bb82e684bbe9eb9ef6892bd7fd6/os-quota-sets/43f66bb82e684bbe9eb9ef6892bd7fd6

body: {"quota_set": {"gigabytes": 1000, "tenant_id": "43f66bb82e684bbe9eb9ef6892bd7fd6", "snapshots": 20, "volumes": 10}}

Route:

Route path: '/{project_id}/os-quota-sets/:(id)', defaults: {'action': u'update', 'controller': <cinder.api.openstack.wsgi.Resource object at 0x7f1662934c90>}

Dispatch:

method: <bound method QuotaSetsController.update of <cinder.api.contrib.quotas.QuotaSetsController object at 0x7f1662983c50>>

body: {"quota_set": {"gigabytes": 1000, "tenant_id": "43f66bb82e684bbe9eb9ef6892bd7fd6", "snapshots": 20, "volumes": 10}}

action_args: {'body': {u'quota_set': {u'gigabytes': 1000, u'tenant_id': u'43f66bb82e684bbe9eb9ef6892bd7fd6', u'volumes': 10, u'snapshots': 20}}, 'project_id': u'43f66bb82e684bbe9eb9ef6892bd7fd6', 'id': u'43f66bb82e684bbe9eb9ef6892bd7fd6'}

action_result: {'quota_set': {'snapshots': 20L, u'gigabytes_type-network': -1, u'snapshots_type-network': -1, u'volumes_newtype': -1, 'backups': 10, u'snapshots_typelvm': -1, u'volumes_type-b1': -1, u'volumes_type-b2': -1, u'snapshots_lvmtype': -1, u'volumes_lvmtype': -1, u'gigabytes_lvmtype': -1, u'volumes_typelvm': -1, u'snapshots_newtype': -1, 'gigabytes': 1000L, 'backup_gigabytes': 1000, u'gigabytes_type-b2': -1, u'snapshots_type-b1': -1, u'snapshots_type-b2': -1, u'gigabytes_type-b1': -1, u'gigabytes_newtype': -1, u'volumes_type-network': -1, u'gigabytes_typelvm': -1, 'volumes': 10L}}

Response:

serializers: {'xml': <class 'cinder.api.contrib.quotas.QuotaTemplate'>}

response = ['{"quota_set": {"snapshots": 20, "gigabytes_type-network": -1, "snapshots_type-network": -1, "volumes_newtype": -1, "backups": 10, "snapshots_typelvm": -1, "volumes_type-b1": -1, "volumes_type-b2": -1, "snapshots_lvmtype": -1, "volumes_lvmtype": -1, "gigabytes_lvmtype": -1, "volumes_typelvm": -1, "snapshots_newtype": -1, "gigabytes": 1000, "backup_gigabytes": 1000, "gigabytes_type-b2": -1, "snapshots_type-b1": -1, "snapshots_type-b2": -1, "gigabytes_type-b1": -1, "gigabytes_newtype": -1, "volumes_type-network": -1, "gigabytes_typelvm": -1, "volumes": 10}}']

2.3 以 cinder extend 的简要过程 （os-extend 是 volumes 资源扩展的 wsgi.action）

Request：

POST http://9.123.245.88:8776/v2/2f07ad0f1beb4b629e42e1113196c04b/volumes/8ec62cc2-2f88-409c-a249-bfc1614eb8ab/action

{"os-extend": {"new_size": 2}}

Route：

Route path: '/{project_id}/volumes/:(id)/action', defaults: {'action': u'action', 'controller': <cinder.api.openstack.wsgi.Resource object at 0x7f459d3d4d90>}

match： {'action': u'action', 'controller': <cinder.api.openstack.wsgi.Resource object at 0x7f459d3d4d90>, 'project_id': u'2f07ad0f1beb4b629e42e1113196c04b', 'id': u'8ec62cc2-2f88-409c-a249-bfc1614eb8ab'}

Dispatch：

method：<bound method VolumeActionsController._extend of <cinder.api.contrib.volume_actions.VolumeActionsController object at 0x7f459d254d10>>

action_args: {'body': {u'os-extend': {u'new_size': 2}}, 'id': u'42ddb30e-8a36-4301-99e4-9547ce4e860d'}

Result：Invalid volume: Volume status must be available to extend. （因为此volume 的状态为 error）

3. 小结

小结 osapi-volume 处理 HTTP Request 的全过程：：

1. Middleware filters 处理 HTTP Request header 的子过程，这其中主要是用户校验。

2. APIRouter 生成 HTTP Request 的 response 的子过程。依次： RoutesMiddleware 产生 route -> Resource 的 request body 反序列化、消息派发给 Resource Controller 的具体方法、Resource controller 的具体方法执行、结果的序列化等。

3. 个别 Middleware 还要处理一下 response，包括 RoutesMiddleware。

4. 最终的 response 发还给客户端。