openstack 基础库之 oslo_context

官方使用

• 例1：增加日志的上下文

  from oslo_config import cfg
  from oslo_context import context
  from oslo_log import log as logging
  
  CONF = cfg.CONF
  DOMAIN = "demo"
  
  logging.register_options(CONF)
  logging.setup(CONF, DOMAIN)
  
  LOG = logging.getLogger(__name__)
  
  LOG.info("Message without context")
  # ids in Openstack are 32 characters long
  # For readability a shorter id value is used
  context.RequestContext(user='6ce90b4d',
                         tenant='d6134462',
                         project_domain='a6b9360e')
  LOG.info("Message with context")

• 输出：会加上上下文信息

  2016-01-21 17:30:50.263 12201 INFO __main__ [-] Message without context
  2016-01-21 17:30:50.264 12201 INFO __main__ [req-e591e881-36c3-4627-a5d8-54df200168ef 6ce90b4d d6134462 - - a6b9360e] Message with context

• 思考：context.RequestContext() 仅仅新建一个对象为什么能够改变 Log 的行为?

context 实现原理

oslo.context 包代码比较少，只有一个类，如下：

class RequestContext(object):
    def __init__(
        self,
        auth_token: ty.Optional[str] = None,
        user_id: ty.Optional[str] = None,
        project_id: ty.Optional[str] = None,
        domain_id: ty.Optional[str] = None,
        user_domain_id: ty.Optional[str] = None,
        project_domain_id: ty.Optional[str] = None,
        is_admin: bool = False,
        read_only: bool = False,
        show_deleted: bool = False,
        request_id: ty.Optional[str] = None,
        resource_uuid: ty.Optional[str] = None,
        overwrite: bool = True,
        roles: ty.Optional[ty.List[str]] = None,
        user_name: ty.Optional[str] = None,
        project_name: ty.Optional[str] = None,
        domain_name: ty.Optional[str] = None,
        user_domain_name: ty.Optional[str] = None,
        project_domain_name: ty.Optional[str] = None,
        is_admin_project: bool = True,
        service_token: ty.Optional[str] = None,
        service_user_id: ty.Optional[str] = None,
        service_user_name: ty.Optional[str] = None,
        service_user_domain_id: ty.Optional[str] = None,
        service_user_domain_name: ty.Optional[str] = None,
        service_project_id: ty.Optional[str] = None,
        service_project_name: ty.Optional[str] = None,
        service_project_domain_id: ty.Optional[str] = None,
        service_project_domain_name: ty.Optional[str] = None,
        service_roles: ty.Optional[ty.List[str]] = None,
        global_request_id: ty.Optional[str] = None,
        system_scope: ty.Optional[str] = None,
    ):
        # setting to private variables to avoid triggering subclass properties
        self._user_id = user_id
        self._project_id = project_id
        self._domain_id = domain_id
        self._user_domain_id = user_domain_id
        self._project_domain_id = project_domain_id

        self.auth_token = auth_token
        self.user_name = user_name
        self.project_name = project_name
        self.domain_name = domain_name
        self.system_scope = system_scope
        self.user_domain_name = user_domain_name
        self.project_domain_name = project_domain_name
        self.is_admin = is_admin
        self.is_admin_project = is_admin_project
        self.read_only = read_only
        self.show_deleted = show_deleted
        self.resource_uuid = resource_uuid
        self.roles = roles or []

        self.service_token = service_token
        self.service_user_id = service_user_id
        self.service_user_name = service_user_name
        self.service_user_domain_id = service_user_domain_id
        self.service_user_domain_name = service_user_domain_name
        self.service_project_id = service_project_id
        self.service_project_name = service_project_name
        self.service_project_domain_id = service_project_domain_id
        self.service_project_domain_name = service_project_domain_name
        self.service_roles = service_roles or []

        if not request_id:
            request_id = generate_request_id()
        self.request_id = request_id
        self.global_request_id = global_request_id
        if overwrite or not get_current():
            self.update_store()

• 构造方法：将构造参数设置为实例的属性值，最后再用这一行调用对这些设置的值进行存储 self.update_store()

def update_store(self) -> None:
    """Store the context in the current thread."""
    _request_store.context = self

• _request_store 是一个模块级的变量，在模块加载的时候初始化

  import threading
  _request_store = threading.local()

  线程本地存储（Thread-local Storage）是一种机制，允许每个线程在共享相同全局变量的情况下，拥有自己的独立变量副本。通过线程本地存储，每个线程可以独立地访问和修改自己的变量副本，而不会影响其他线程的副本

_request_store 用于在多线程环境下存储与请求相关的数据，以便每个线程可以访问和修改自己的请求数据，而不会与其他线程的数据发生冲突

• 在一个线程中 _request_store 相当于一个全局的变量
• 在不通的线程中，_request_store 相互隔离

log 使用

oslo_log 有一个ContextFormatter 他就是用来 context 来格式化日志输出的，他会引用线程本地存储。