前言

这段时间都在读 OpenStack 组件源码，主要依赖官方文档和《OpenStack 设计与实现》，目前这部分的工作告一段落，稍微整理一下阅读源码的方法，主要是如何找到程序的入口。

Kolla-Ansible 搭建环境使用的是 Victoria 版本的源码，主要包含以下几个组件：

Keystone
Glance
Nova
Neutron
Heat

源码阅读

实际上，程序的入口从 setup.cfg 文件就可以看出来了，如何处理请求会涉及到 paste.ini 配置文件。有些组件比较复杂（比如 nova、neutron），组件本身包含多个子组件，有 wsgi 应用、OS-Ken 应用等，启动方式也并不统一，因此需要深入源码才能找到真正的启动位置和启动方式。

此外，组件不仅有对外提供的 RESTful API 接口，组件内部和组件之间还有 RPC 调用，会涉及消息队列（一般是 rabbitmq）和 socket 通信，而这也是需要深入源码才能理清的。

setup.cfg

上述 OpenStack 组件都包含 setup.cfg 文件，Setuptools 工具使用该配置文件设置包的元数据和其他选项，具体的配置项可以在文档中查阅。

这里主要关注的是 entry_points 小节，可以找到代码的入口点，组件启动的方式包括 console_scripts 和 wsgi_scripts ，分别表示命令行脚本和 wsgi 脚本，通常 wsgi 脚本通过 Apache + mod_wsgi 调用。

以 glance 为例，glance-api 可以使用命令行脚本启动也可以使用 wsgi 脚本启动，不过官方建议在生产环境中使用命令行脚本启动。

[entry_points]
console_scripts =
    glance-api = glance.cmd.api:main
    glance-cache-prefetcher = glance.cmd.cache_prefetcher:main
    glance-cache-pruner = glance.cmd.cache_pruner:main
    glance-cache-manage = glance.cmd.cache_manage:main
    glance-cache-cleaner = glance.cmd.cache_cleaner:main
    glance-control = glance.cmd.control:main
    glance-manage = glance.cmd.manage:main
    glance-replicator = glance.cmd.replicator:main
    glance-scrubber = glance.cmd.scrubber:main
    glance-status = glance.cmd.status:main
wsgi_scripts =
    glance-wsgi-api = glance.common.wsgi_app:init_app

等号右边可以理解为调用的函数，以 glance-api = glance.cmd.api:main 为例，定位源码 glance/cmd/api.py 中的 main 函数。

def main():
    try:
        config.parse_args()             # 读取配置
        config.set_config_defaults()    # 设置默认配置
        wsgi.set_eventlet_hub()         # 设置 eventlet.hub
        logging.setup(CONF, 'glance')   # 日志
        gmr.TextGuruMeditation.setup_autorun(version)
        notifier.set_defaults()

        if CONF.profiler.enabled: # OSProfiler
            osprofiler.initializer.init_from_conf(
                conf=CONF,
                context={},
                project="glance",
                service="api",
                host=CONF.bind_host
            )

        # NOTE(danms): Configure system-wide threading model to use eventlet
        glance.async_.set_threadpool_model('eventlet') # 设置同步线程池模型

        # NOTE(abhishekk): Added initialize_prefetcher KW argument to Server
        # object so that prefetcher object should only be initialized in case
        # of API service and ignored in case of registry. Once registry is
        # removed this parameter should be removed as well.
        initialize_prefetcher = False
        if CONF.paste_deploy.flavor == 'keystone+cachemanagement':
            initialize_prefetcher = True
        server = wsgi.Server(initialize_glance_store=True,  # wsgi 应用
                             initialize_prefetcher=initialize_prefetcher)
        server.start(config.load_paste_app('glance-api'), default_port=9292) # 启动 wsgi 应用
        server.wait() # 等待启动完成
    except Exception as e:
        fail(e)

再来看一下 wsgi 脚本 glance-wsgi-api = glance.common.wsgi_app:init_app，定位源码 glance/common/wsgi_app.py 的 init_app 函数。重点关注的是最后使用 Paste Deploy 加载 wsgi 应用。

def init_app():
    config.set_config_defaults()
    config_files = _get_config_files()
    CONF([], project='glance', default_config_files=config_files)
    logging.setup(CONF, "glance")

    # NOTE(danms): We are running inside uwsgi or mod_wsgi, so no eventlet;
    # use native threading instead.
    glance.async_.set_threadpool_model('native')
    atexit.register(drain_threadpools)

    # NOTE(danms): Change the default threadpool size since we
    # are dealing with native threads and not greenthreads.
    # Right now, the only pool of default size is tasks_pool,
    # so if others are created this will need to change to be
    # more specific.
    common.DEFAULT_POOL_SIZE = CONF.wsgi.task_pool_threads

    if CONF.enabled_backends:
        if store_utils.check_reserved_stores(CONF.enabled_backends):
            msg = _("'os_glance_' prefix should not be used in "
                    "enabled_backends config option. It is reserved "
                    "for internal use only.")
            raise RuntimeError(msg)
        glance_store.register_store_opts(CONF, reserved_stores=RESERVED_STORES)
        glance_store.create_multi_stores(CONF, reserved_stores=RESERVED_STORES)
        glance_store.verify_store()
    else:
        glance_store.register_opts(CONF)
        glance_store.create_stores(CONF)
        glance_store.verify_default_store()

    run_staging_cleanup()

    _setup_os_profiler()
    return config.load_paste_app('glance-api')  # Paste Deploy

paste.ini

paste.ini 文件是 wsgi 应用的配置文件，根据该文件可以知道应用程序是如何映射 URL 以及如何处理请求。

paste-ini 配置文件类似 ini 配置，每个 Section 的格式均为 [type:name] ，有以下几个小节

composite：收到请求后通过的第一个 Section，表示需要将 HTTP URL Request 调度到一个或者多个应用中
app：实现主要功能的具体应用
pipeline：过滤器管道，最后一个必须是 app 类型
filter：实现过滤器功能的中间件，用于过滤请求和响应

仍然以 glance 为例，启动 glance 服务时需要指定 paste.ini 配置文件，源码中的 etc/glance-api-paste.ini 配置文件如下：

# Use this pipeline for no auth or image caching - DEFAULT
[pipeline:glance-api]
pipeline = cors healthcheck http_proxy_to_wsgi versionnegotiation osprofiler unauthenticated-context rootapp

# Use this pipeline for image caching and no auth
[pipeline:glance-api-caching]
pipeline = cors healthcheck http_proxy_to_wsgi versionnegotiation osprofiler unauthenticated-context cache rootapp

# Use this pipeline for caching w/ management interface but no auth
[pipeline:glance-api-cachemanagement]
pipeline = cors healthcheck http_proxy_to_wsgi versionnegotiation osprofiler unauthenticated-context cache cachemanage rootapp

# Use this pipeline for keystone auth
[pipeline:glance-api-keystone]
pipeline = cors healthcheck http_proxy_to_wsgi versionnegotiation osprofiler authtoken context  rootapp

# Use this pipeline for keystone auth with image caching
[pipeline:glance-api-keystone+caching]
pipeline = cors healthcheck http_proxy_to_wsgi versionnegotiation osprofiler authtoken context cache rootapp

# Use this pipeline for keystone auth with caching and cache management
[pipeline:glance-api-keystone+cachemanagement]
pipeline = cors healthcheck http_proxy_to_wsgi versionnegotiation osprofiler authtoken context cache cachemanage rootapp

# Use this pipeline for authZ only. This means that the registry will treat a
# user as authenticated without making requests to keystone to reauthenticate
# the user.
[pipeline:glance-api-trusted-auth]
pipeline = cors healthcheck http_proxy_to_wsgi versionnegotiation osprofiler context rootapp

# Use this pipeline for authZ only. This means that the registry will treat a
# user as authenticated without making requests to keystone to reauthenticate
# the user and uses cache management
[pipeline:glance-api-trusted-auth+cachemanagement]
pipeline = cors healthcheck http_proxy_to_wsgi versionnegotiation osprofiler context cache cachemanage rootapp

[composite:rootapp]
paste.composite_factory = glance.api:root_app_factory
/: apiversions
/v2: apiv2app

[app:apiversions]
paste.app_factory = glance.api.versions:create_resource

[app:apiv2app]
paste.app_factory = glance.api.v2.router:API.factory

[filter:healthcheck]
paste.filter_factory = oslo_middleware:Healthcheck.factory
backends = disable_by_file
disable_by_file_path = /etc/glance/healthcheck_disable

[filter:versionnegotiation]
paste.filter_factory = glance.api.middleware.version_negotiation:VersionNegotiationFilter.factory

[filter:cache]
paste.filter_factory = glance.api.middleware.cache:CacheFilter.factory

[filter:cachemanage]
paste.filter_factory = glance.api.middleware.cache_manage:CacheManageFilter.factory

[filter:context]
paste.filter_factory = glance.api.middleware.context:ContextMiddleware.factory

[filter:unauthenticated-context]
paste.filter_factory = glance.api.middleware.context:UnauthenticatedContextMiddleware.factory

[filter:authtoken]
paste.filter_factory = keystonemiddleware.auth_token:filter_factory
delay_auth_decision = true

[filter:gzip]
paste.filter_factory = glance.api.middleware.gzip:GzipMiddleware.factory

[filter:osprofiler]
paste.filter_factory = osprofiler.web:WsgiMiddleware.factory
hmac_keys = SECRET_KEY  #DEPRECATED
enabled = yes  #DEPRECATED

[filter:cors]
paste.filter_factory =  oslo_middleware.cors:filter_factory
oslo_config_project = glance
oslo_config_program = glance-api

[filter:http_proxy_to_wsgi]
paste.filter_factory = oslo_middleware:HTTPProxyToWSGI.factory

可以看到 pipeline 最后的总是 rootapp 应用，paste.composite_factory 设置应用的工厂函数

[composite:rootapp]
paste.composite_factory = glance.api:root_app_factory
/: apiversions
/v2: apiv2app

定位源码 glance/api.py 的 root_app_factory 函数，显然是用于设置 url 映射的。

1 2	def root_app_factory(loader, global_conf, local_conf): return paste.urlmap.urlmap_factory(loader, global_conf, local_conf)

由于我主要关注的是服务的启动方式，因此 URL 映射具体是如何实现的并没有了解。通过 paste.ini 文件可以知道请求到达真正的应用前经过了什么中间件（过滤器）的处理，这些中间件的源码也可以看一看。

OpenStack 组件源码阅读

前言

源码阅读

setup.cfg

paste.ini

参阅