horizon源码分析(二)
源码版本:H版
一、简要回顾
对于请求:
地址:/dashboard/admin/instances/
URL绑定为:
openstack_dashboard/dashboards/admin/instances/urls.py
二、目录结构
三、请求的响应
接下来主要分析如下代码:
openstack_dashboard/dashboards/admin/instances/views.py
views.AdminIndexView.as_view()
先从Django的generic view说起.... generic view中的as_view()可以返回一个Django的view函数,该view函数会构造出类实例,将as_view()中传入的参数设置为该实例的属性,然后调用dispatch函数,dispatch函数通常会将request请求转给类中的post或get函数。generic view的主要使用方法是用子类重写其中的属性或方法。详细情况可以参考Django官方文档:https://docs.djangoproject.com/en/1.7/topics/class-based-views/。对Django框架的深入了解对于理解Horizon十分必要,as_view函数最终达到的效果还是将处理逻辑放入post函数或get函数中,这点和其他网络框架类似。
分析AdminIndexView.as_view(),由于请求的方式为POST,其会调用该类的post函数。先看看AdminIndexView类中的属性设置如下:
openstack_dashboard/dashboards/admin/instances/views.py
class AdminIndexView(tables.DataTableView): table_class = project_tables.AdminInstancesTable template_name = 'admin/instances/index.html'
由于AdminIndexView -> DataTableView -> MultiTableView,类关系如下图所示。追踪到MultiTableView.post,该post函数会调用该类的get函数。
1、 DataTableView、DataTable、Action三者的说明
这里为了后面分析的方便,先对DataTableView、DataTable、Action进行一番说明,如下:
(参考:http://docs.openstack.org/developer/horizon/topics/tables.html)
1)DataTableView簇有如下属性:
_data={
表名:data(通过get_data函数获得)
...
}
_tables={
表名:table实例
}
table=table实例
说明:本例中data为一个包含instance的list
DataTableView可以通过table_class绑定具体的DataTable,通过get_data函数获取data,该函数通常调用openstack_dashboard/api模块获取数据,最后,DataTableView通过handle_table函数负责将data和table挂钩,或者处理table行为。DataTableView正如其名字一样,拥有table和data,负责处理data的获取,Table的创建,以及二者的绑定等。
2)DataTable簇:
DataTable规定了table的column和action,可以处理和table绑定的data,take_action函数负责处理action。以AdminInstanceTable的创建过程为例,其中使用了metaclass对DataTable及其子类进行修改,具体解释如下:
先观察AdminInstancesTable类和DataTableOptions类:
class AdminInstancesTable(tables.DataTable): ... class Meta: name = "instances" verbose_name = _("Instances") status_columns = ["status", "task"] table_actions = (project_tables.TerminateInstance, AdminInstanceFilterAction) row_class = AdminUpdateRow row_actions = (project_tables.ConfirmResize, project_tables.RevertResize, AdminEditInstance, project_tables.ConsoleLink, project_tables.LogLink, project_tables.CreateSnapshot, project_tables.TogglePause, project_tables.ToggleSuspend, MigrateInstance, project_tables.SoftRebootInstance, project_tables.RebootInstance, project_tables.TerminateInstance)
class DataTableOptions(object): def __init__(self, options): self.name = getattr(options, 'name', self.__class__.__name__) verbose_name = getattr(options, 'verbose_name', None) \ or self.name.title() self.verbose_name = verbose_name self.columns = getattr(options, 'columns', None) self.status_columns = getattr(options, 'status_columns', []) self.table_actions = getattr(options, 'table_actions', []) self.row_actions = getattr(options, 'row_actions', []) self.row_class = getattr(options, 'row_class', Row) self.column_class = getattr(options, 'column_class', Column) self.pagination_param = getattr(options, 'pagination_param', 'marker') ...
接着分析metaclass对类的修改...
class DataTable(object): __metaclass__ = DataTableMetaclass
class DataTableMetaclass(type): def __new__(mcs, name, bases, attrs): # Process options from Meta class_name = name """将类中的Meta转变为DataTableOptions,添加为类的_meta属性""" attrs["_meta"] = opts = DataTableOptions(attrs.get("Meta", None)) # Gather columns; this prevents the column from being an attribute # on the DataTable class and avoids naming conflicts. """将类中的column属性聚集作为新的列属性,阻止其作为类属性""" columns = [] for attr_name, obj in attrs.items(): if issubclass(type(obj), (opts.column_class, Column)): column_instance = attrs.pop(attr_name) column_instance.name = attr_name column_instance.classes.append('normal_column') columns.append((attr_name, column_instance)) columns.sort(key=lambda x: x[1].creation_counter) # Iterate in reverse to preserve final order for base in bases[::-1]: if hasattr(base, 'base_columns'): columns = base.base_columns.items() + columns attrs['base_columns'] = SortedDict(columns) ... """收集row_action和table_action对象""" actions = list(set(opts.row_actions) | set(opts.table_actions)) actions.sort(key=attrgetter('name')) actions_dict = SortedDict([(action.name, action()) for action in actions]) attrs['base_actions'] = actions_dict if opts._filter_action: # Replace our filter action with the instantiated version opts._filter_action = actions_dict[opts._filter_action.name] # Create our new class! return type.__new__(mcs, name, bases, attrs)
总结概况如下图:
说明:使用metaclass对类进行修改,这样极大地增加了程序的可扩展性和灵活性,但同时复杂度也增大。metaclass的理解可以参考:
http://blog.csdn.net/psh2009/article/details/10330747
http://jianpx.iteye.com/blog/908121
3)Action簇:
利用action函数进行处理
继续分析MultiTableView类的get函数,如下:
horizon/tables/views.py
MultiTableView类: def get(self, request, *args, **kwargs): handled = self.construct_tables() if handled: return handled """如果handled不为空则表明只是处理table,无需再次用table渲染模板并返回;否则的话就需要渲染模板。具体渲染操作如下""" context = self.get_context_data(**kwargs) return self.render_to_response(context) def construct_tables(self): """根据类中的table_class属性绑定的DataTable类,创建或返回DataTable对象,此处为AdminInstancesTable对象 """ tables = self.get_tables().values() # Early out before data is loaded for table in tables: """如果当前请求需要预处理或者是AJAX更新操作,将在如下函数中进行,特别注意,此处正是AJAX发送行更新请求的响应处""" preempted = table.maybe_preempt() if preempted: return preempted # Load data into each table and check for action handlers for table in tables: handled = self.handle_table(table) if handled: return handled return None
MultiTableMixin类: def handle_table(self, table): name = table.name """获取数据,此处暂且不深入分析""" data = self._get_data_dict() """获取与该DataTable相关的数据,并将数据和该DataTable挂钩""" self._tables[name].data = data[table._meta.name] """有关翻页的设置,此处暂且不管""" self._tables[name]._meta.has_more_data = self.has_more_data(table) """此处为调用AdminInstancesTable.maybe_handle函数""" handled = self._tables[name].maybe_handle() return handled
horizon/tables/base.py
DataTable类: def maybe_handle(self): """ Determine whether the request should be handled by any action on this table after data has been loaded. """ request = self.request """获取request中的数据,这里为 table_name=’instances’ action_name=’soft_reboot’ obj_id=’89a8849b-a3cd-4ce0-9158-c3dd69e8508e’ """ table_name, action_name, obj_id = self.check_handler(request) if table_name == self.name and action_name: action_names = [action.name for action in self.base_actions.values() if not action.preempt] # do not run preemptive actions here if action_name in action_names: return self.take_action(action_name, obj_id) return None
为了后面的继续分析,先看Action簇的类关系如下:
继续分析take_action函数...
horizon/tables/base.py
DataTable类: """ action_name=’soft_reboot’ obj_id=’89a8849b-a3cd-4ce0-9158-c3dd69e8508e’ """ def take_action(self, action_name, obj_id=None, obj_ids=None): obj_ids = obj_ids or self.request.POST.getlist('object_ids') """得到SoftRebootInstance实例""" action = self.base_actions.get(action_name, None) if not action or action.method != self.request.method: return None if not action.requires_input or obj_id or obj_ids: if obj_id: obj_id = self.sanitize_id(obj_id) if obj_ids: obj_ids = [self.sanitize_id(i) for i in obj_ids] """SoftRebootInstance->RebootInstance->BatchAction->Action,由于BatchAction有handle函数,所以在Action的__init__()中将属性handles_multiple设置为True
""" if not action.handles_multiple: response = action.single(self, self.request, obj_id) else:#进入此项 if obj_id: obj_ids = [obj_id] response = action.multiple(self, self.request, obj_ids) return response elif action and action.requires_input and not (obj_id or obj_ids): messages.info(self.request, _("Please select a row before taking that action.")) return None
注意,这里使用了一个trick,如下:
horizon/tables/actions.py
Action类: def __init__(...): ... if not has_multiple and self.handles_multiple: def multiple(self, data_table, request, object_ids): return self.handle(data_table, request, object_ids) """为该实例动态绑定multiple方法,其实质为调用handle方法""" self.multiple = new.instancemethod(multiple, self)
所以,接下来分析BatchAction中的handle函数...
horizon/tables/actions.py
BatchAction类: def handle(self, table, request, obj_ids): action_success = [] action_failure = [] action_not_allowed = [] for datum_id in obj_ids: datum = table.get_object_by_id(datum_id) datum_display = table.get_object_display(datum) or _("N/A") if not table._filter_action(self, request, datum): action_not_allowed.append(datum_display) LOG.info('Permission denied to %s: "%s"' % (self._conjugate(past=True).lower(), datum_display)) continue try: self.action(request, datum_id) self.update(request, datum) action_success.append(datum_display) self.success_ids.append(datum_id) LOG.info('%s: "%s"' % (self._conjugate(past=True), datum_display)) except Exception as ex: if getattr(ex, "_safe_message", None): ignore = False else: ignore = True action_failure.append(datum_display) exceptions.handle(request, ignore=ignore) ... return shortcuts.redirect(self.get_success_url(request))
openstack_dashboard/dashboards/project/instances/tables.py
SoftRebootInstance类: class SoftRebootInstance(RebootInstance): name = "soft_reboot" action_present = _("Soft Reboot") action_past = _("Soft Rebooted") def action(self, request, obj_id): api.nova.server_reboot(request, obj_id, soft_reboot=True)
在此总结一下,处理的流程大概是DataTableView首先获取Data和Table,然后将Data和Table绑定,如果有对Table的处理则调用Table的函数进行处理,通常最终会落实到Table中Row所对应的Action。补充一下关于返回Table的渲染,首先在template中使用Table对象进行模板渲染,然后Table使用Row进行渲染,Row使用Cell进行渲染,和表格的形式一致。在Row的构造中会绑定Ajax信息,用来对Row进行轮询更新。
四、workflows处理流程
一般Dashboard都不只包含DataTableView,还有很多其他View类,其中WorkflowView比较常见。这里简单说明一下,主要以POST请求为例。经过对DataTableView的分析,很容易明白WorkflowView的处理流程,主要见下图。其中普遍存在用类属性来表明绑定关系的特点,所以图中上面一排的虚线表示类的相互绑定关系,下面的虚线则表明类的调用关系。注意Workflow的finalize函数会先依次调用各个Step的Action的handle方法,然后会调用自己的handle方法做最后的处理!更加详细的说明可以参考:http://docs.openstack.org/developer/horizon/ref/workflows.html
参考文档:
http://docs.openstack.org/developer/horizon/