神器 celery 源码解析 - 5

Celery是一款非常简单、灵活、可靠的分布式系统，可用于处理大量消息，并且提供了一整套操作此系统的工具。Celery 也是一款消息队列工具，可用于处理实时数据以及任务调度。

本文是是celery源码解析的第五篇，在前4篇里分别介绍了vine, py-amqp和kombu:

1. 神器 celery 源码解析- vine实现Promise功能
2. 神器 celery 源码解析- py-amqp实现AMQP协议
3. 神器 celery 源码解析- kombu，一个python实现的消息库
4. 神器 celery 源码解析- kombu的企业级算法

基本扫清celery的基础库后，我们正式进入celery的源码解析，本文包括下面几个部分:

• celery应用示例
• celery项目概述
• worker启动流程跟踪
• client启动流程跟踪
• celery的app
• worker模式启动流程
• 小结

celery应用示例

启动celery之前，我们先使用docker启动一个redis服务，作为broker:

$ docker run -p 6379:6379 --name redis -d redis:6.2.3-alpine

使用telnet监控redis服务，观测任务调度情况:

$ telnet 127.0.0.1 6379
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
monitor
+OK

下面是我们的celery服务代码 myapp.py :

# myapp.py
from celery import Celery

app = Celery(
    'myapp',
    broker='redis://localhost:6379/0',
    result_backend='redis://localhost:6379/0'
)

@app.task
def add(x, y):
    print("add", x, y)
    return x + y

if __name__ == '__main__':
    app.start()

打开一个新的终端，使用下面的命令启动celery的worker服务:

$ python myapp.py worker -l DEBUG

正常情况下，可以看到worker正常启动。启动的时候会显示一些banner信息，包括AMQP的实现协议，任务等:

$ celery -A myapp worker -l DEBUG

 -------------- celery@bogon v5.1.2 (sun-harmonics)
--- ***** ----- 
-- ******* ---- macOS-10.16-x86_64-i386-64bit 2021-09-08 20:33:45
- *** --- * --- 
- ** ---------- [config]
- ** ---------- .> app:         myapp:0x7f855079e730
- ** ---------- .> transport:   redis://localhost:6379/0
- ** ---------- .> results:     disabled://
- *** --- * --- .> concurrency: 12 (prefork)
-- ******* ---- .> task events: OFF (enable -E to monitor tasks in this worker)
--- ***** ----- 
 -------------- [queues]
                .> celery           exchange=celery(direct) key=celery


[tasks]
  . myapp.add

[2021-09-08 20:33:46,220: INFO/MainProcess] Connected to redis://localhost:6379/0
[2021-09-08 20:33:46,234: INFO/MainProcess] mingle: searching for neighbors
[2021-09-08 20:33:47,279: INFO/MainProcess] mingle: all alone
[2021-09-08 20:33:47,315: INFO/MainProcess] celery@bogon ready.

再开启一个终端窗口，作为client执行下面的代码, 可以看到add函数正确的执行，获取到计算 16+16 的结果 32。注意: 这个过程是远程执行的，使用的是delay方法，函数的打印print("add", x, y)并没有输出:

$ python
>>> from myapp import add
>>> task = add.delay(16,16)
>>> task
<AsyncResult: 5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b>
>>> task.get()
32

在celery的worker服务窗口，可以看到类似下面的输出。收到一个执行任务 myapp.add 的请求, 请求的uuid是 5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b ，参数数组是 [16, 16] 正常执行后返回结果32。

[2021-11-11 20:13:48,040: INFO/MainProcess] Task myapp.add[5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b] received
[2021-11-11 20:13:48,040: DEBUG/MainProcess] TaskPool: Apply <function fast_trace_task at 0x7fda086baa60> (args:('myapp.add', '5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b', {'lang': 'py', 'task': 'myapp.add', 'id': '5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b', 'shadow': None, 'eta': None, 'expires': None, 'group': None, 'group_index': None, 'retries': 0, 'timelimit': [None, None], 'root_id': '5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b', 'parent_id': None, 'argsrepr': '(16, 16)', 'kwargsrepr': '{}', 'origin': 'gen63119@localhost', 'ignore_result': False, 'reply_to': '97a3e117-c8cf-3d4c-97c0-c0a76aaf9a16', 'correlation_id': '5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b', 'hostname': 'celery@localhost', 'delivery_info': {'exchange': '', 'routing_key': 'celery', 'priority': 0, 'redelivered': None}, 'args': [16, 16], 'kwargs': {}}, b'[[16, 16], {}, {"callbacks": null, "errbacks": null, "chain": null, "chord": null}]', 'application/json', 'utf-8') kwargs:{})
[2021-11-11 20:13:49,059: INFO/ForkPoolWorker-8] Task myapp.add[5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b] succeeded in 1.0166977809999995s: 32

在redis的monitor窗口，也可以可以看到类似的输出，展示了过程中一些对redis的操作命令:

+1636632828.304020 [0 172.16.0.117:51127] "SUBSCRIBE" "celery-task-meta-5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b"
+1636632828.304447 [0 172.16.0.117:51129] "PING"
+1636632828.305448 [0 172.16.0.117:51129] "LPUSH" "celery" "{\"body\": \"W1sxNiwgMTZdLCB7fSwgeyJjYWxsYmFja3MiOiBudWxsLCAiZXJyYmFja3MiOiBudWxsLCAiY2hhaW4iOiBudWxsLCAiY2hvcmQiOiBudWxsfV0=\", \"content-encoding\": \"utf-8\", \"content-type\": \"application/json\", \"headers\": {\"lang\": \"py\", \"task\": \"myapp.add\", \"id\": \"5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b\", \"shadow\": null, \"eta\": null, \"expires\": null, \"group\": null, \"group_index\": null, \"retries\": 0, \"timelimit\": [null, null], \"root_id\": \"5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b\", \"parent_id\": null, \"argsrepr\": \"(16, 16)\", \"kwargsrepr\": \"{}\", \"origin\": \"gen63119@localhost\", \"ignore_result\": false}, \"properties\": {\"correlation_id\": \"5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b\", \"reply_to\": \"97a3e117-c8cf-3d4c-97c0-c0a76aaf9a16\", \"delivery_mode\": 2, \"delivery_info\": {\"exchange\": \"\", \"routing_key\": \"celery\"}, \"priority\": 0, \"body_encoding\": \"base64\", \"delivery_tag\": \"20dbd584-b669-4ef0-8a3b-41d19b354690\"}}"
+1636632828.307040 [0 172.16.0.117:52014] "MULTI"
+1636632828.307075 [0 172.16.0.117:52014] "ZADD" "unacked_index" "1636632828.038743" "20dbd584-b669-4ef0-8a3b-41d19b354690"
+1636632828.307088 [0 172.16.0.117:52014] "HSET" "unacked" "20dbd584-b669-4ef0-8a3b-41d19b354690" "[{\"body\": \"W1sxNiwgMTZdLCB7fSwgeyJjYWxsYmFja3MiOiBudWxsLCAiZXJyYmFja3MiOiBudWxsLCAiY2hhaW4iOiBudWxsLCAiY2hvcmQiOiBudWxsfV0=\", \"content-encoding\": \"utf-8\", \"content-type\": \"application/json\", \"headers\": {\"lang\": \"py\", \"task\": \"myapp.add\", \"id\": \"5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b\", \"shadow\": null, \"eta\": null, \"expires\": null, \"group\": null, \"group_index\": null, \"retries\": 0, \"timelimit\": [null, null], \"root_id\": \"5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b\", \"parent_id\": null, \"argsrepr\": \"(16, 16)\", \"kwargsrepr\": \"{}\", \"origin\": \"gen63119@localhost\", \"ignore_result\": false}, \"properties\": {\"correlation_id\": \"5aabfc0b-04b5-4a51-86b0-6a7263e2ef3b\", \"reply_to\": \"97a3e117-c8cf-3d4c-97c0-c0a76aaf9a16\", \"delivery_mode\": 2, \"delivery_info\": {\"exchange\": \"\", \"routing_key\": \"celery\"}, \"priority\": 0, \"body_encoding\": \"base64\", \"delivery_tag\": \"20dbd584-b669-4ef0-8a3b-41d19b354690\"}}, \"\", \"celery\"]"
...

我们再一次回顾下图，对比一下示例，加强理解：

hello-world-example-routing- 我们先启动一个celery的worker服务作为消费者

• 再启动一个窗口作为生产者执行task
• 使用redis作为broker，负责生产者和消费者之间的消息通讯
• 最终生成者的task，作为消息发送到远程的消费者上执行，执行的结果又通过网络回传给生产者

上面示例展示了celery作为一个分布式任务调度系统的执行过程，本地的任务调用，通过AMQP协议的包装，作为消息发送到远程的消费者执行。

celery项目概述

解析celery采用的代码版本5.0.5, 主要模块结构:

从项目结构看，模块较多，功能复杂。不过我们已经搞定了vine, py-amqp和kombu三个库，接下来只需要理解worker，beat和multi三种服务模型，就可以较好的了解celery这个分布式系统如何构建。

worker启动流程跟踪

worker的启动命令 celery -A myapp worker -l DEBUG 使celery作为一个模块，入口在main文件的main函数:

# ch23-celery/celery-5.0.5/celery/__main__.py
def main():
    """Entrypoint to the ``celery`` umbrella command."""
    """celery命令入口"""
    ...
    # 具体执行的main函数
    from celery.bin.celery import main as _main
    sys.exit(_main())

celery命令作为主命令，加载celery-app的同时，还会启动worker子命令:

# ch23-celery/celery-5.0.5/celery/bin/celery.py
def celery(ctx, app, broker, result_backend, loader, config, workdir,
           no_color, quiet, version):
    """Celery command entrypoint."""
    ...
    ctx.obj = CLIContext(app=app, no_color=no_color, workdir=workdir,
                         quiet=quiet)
    # worker/beat/events三个主要子命令参数
    # User options
    worker.params.extend(ctx.obj.app.user_options.get('worker', []))
    beat.params.extend(ctx.obj.app.user_options.get('beat', []))
    events.params.extend(ctx.obj.app.user_options.get('events', []))

def main() -> int:
    """Start celery umbrella command.

    This function is the main entrypoint for the CLI.

    :return: The exit code of the CLI.
    """
    return celery(auto_envvar_prefix="CELERY")

在worker子命令中创建worker并启动:

# ch23-celery/celery-5.0.5/celery/bin/worker.py
def worker(ctx, hostname=None, pool_cls=None, app=None, uid=None, gid=None,
           loglevel=None, logfile=None, pidfile=None, statedb=None,
           **kwargs):
    # 创建和启动worker
    worker = app.Worker(
        hostname=hostname, pool_cls=pool_cls, loglevel=loglevel,
        logfile=logfile,  # node format handled by celery.app.log.setup
        pidfile=node_format(pidfile, hostname),
        statedb=node_format(statedb, hostname),
        no_color=ctx.obj.no_color,
        quiet=ctx.obj.quiet,
        **kwargs)
    worker.start()

下面是创建worker的方式，创一个 celery.apps.worker:Worker 对象:

# ch23-celery/celery-5.0.5/celery/app/base.py
def Worker(self):
    # 创建worker
    return self.subclass_with_self('celery.apps.worker:Worker')

服务启动过程中，调用链路如下：

                                 +----------+
                             +--->app.celery|
                             |   +----------+
+---------+   +----------+   |
|main.main+--->bin.celery+---+
+---------+   +----------+   |
                             |   +----------+   +-----------+
                             +--->bin.worker+--->apps.worker|
                                 +----------+   +-----------+

在这个服务启动过程中，创建了celery-application和worker-application两个应用程序。至于具体的启动流程，我们暂时跳过，先看看客户端的流程。

client启动流程分析

示例client的启动过程包括下面4步: 1 创建celery-application， 2 创建task 3 调用task的delay方法执行任务得到一个异步结果 4 最后使用异步结果的get方法获取真实结果

task是通过app创建的装饰器创建的Promise对象:

# ch23-celery/celery-5.0.5/celery/app/base.py
task_cls = 'celery.app.task:Task'

def task(self, *args, **opts):
    """Decorator to create a task class out of any callable.
    """
    def inner_create_task_cls(shared=True, filter=None, lazy=True, **opts):

        def _create_task_cls(fun):

            ret = PromiseProxy(self._task_from_fun, (fun,), opts,
                                       __doc__=fun.__doc__)
            return ret

        return _create_task_cls
    return inner_create_task_cls(**opts)

task实际上是一个由Task基类动态创建的子类:

def _task_from_fun(self, fun, name=None, base=None, bind=False, **options):
    base = base or self.Task
    task = type(fun.__name__, (base,), dict({
                'app': self,
                'name': name,
                'run': run,
                '_decorated': True,
                '__doc__': fun.__doc__,
                '__module__': fun.__module__,
                '__annotations__': fun.__annotations__,
                '__header__': staticmethod(head_from_fun(fun, bound=bind)),
                '__wrapped__': run}, **options))
    add_autoretry_behaviour(task, **options)
    # 增加task
    self._tasks[task.name] = task
    task.bind(self)  # connects task to this app
    add_autoretry_behaviour(task, **options)
    return task

任务的执行使用app的send_task方法进行:

# ch23-celery/celery-5.0.5/celery/app/task.py
def delay(self, *args, **kwargs):
    ...
    return app.send_task(
                self.name, args, kwargs, task_id=task_id, producer=producer,
                link=link, link_error=link_error, result_cls=self.AsyncResult,
                shadow=shadow, task_type=self,
                **options
            )

可以看到，client作为生产者启动任务，也需要创建celery-application，下面我们就先看celery-application的实现。

celery的app两大功能

Celery的构造函数:

class Celery:

    # 协议类
    amqp_cls = 'celery.app.amqp:AMQP'
    backend_cls = None
    # 事件类
    events_cls = 'celery.app.events:Events'
    loader_cls = None
    log_cls = 'celery.app.log:Logging'
    # 控制类
    control_cls = 'celery.app.control:Control'
    # 任务类
    task_cls = 'celery.app.task:Task'
    # 任务注册中心
    registry_cls = 'celery.app.registry:TaskRegistry'
    ...

    def __init__(self, main=None, loader=None, backend=None,
                 amqp=None, events=None, log=None, control=None,
                 set_as_current=True, tasks=None, broker=None, include=None,
                 changes=None, config_source=None, fixups=None, task_cls=None,
                 autofinalize=True, namespace=None, strict_typing=True,
                 **kwargs):
        # 启动步骤
        self.steps = defaultdict(set)
        # 待执行的task
        self._pending = deque()
        # 所有任务
        self._tasks = self.registry_cls(self._tasks or {})
        ...
        self.__autoset('broker_url', broker)
        self.__autoset('result_backend', backend)
        ...
        self.on_init()
        _register_app(self)

可以看到celery类提供了一些默认模块类的名称，可以根据这些类名动态创建对象。app对象任务的处理使用一个队列作为pending状态的任务容器，使用TaskRegistry来管理任务的注册。

任务通过task装饰器，记录到celery的TaskRegistry中:

def task(self, *args, **opts):
    ...
    # 增加task
    self._tasks[task.name] = task
    task.bind(self)  # connects task to this app
    add_autoretry_behaviour(task, **options)
    ...

celery另外一个核心功能是提供到broker的连接:

def _connection(self, url, userid=None, password=None,
                virtual_host=None, port=None, ssl=None,
                connect_timeout=None, transport=None,
                transport_options=None, heartbeat=None,
                login_method=None, failover_strategy=None, **kwargs):
    conf = self.conf
    return self.amqp.Connection(
        url,
        userid or conf.broker_user,
        password or conf.broker_password,
        virtual_host or conf.broker_vhost,
        port or conf.broker_port,
        transport=transport or conf.broker_transport,
        ssl=self.either('broker_use_ssl', ssl),
        heartbeat=heartbeat,
        login_method=login_method or conf.broker_login_method,
        failover_strategy=(
            failover_strategy or conf.broker_failover_strategy
        ),
        transport_options=dict(
            conf.broker_transport_options, **transport_options or {}
        ),
        connect_timeout=self.either(
            'broker_connection_timeout', connect_timeout
        ),
    )
broker_connection = connection

@cached_property
def amqp(self):
    """AMQP related functionality: :class:`~@amqp`."""
    return instantiate(self.amqp_cls, app=self)

AMQP的实现，是依赖kombu提供的AMQP协议封装:

from kombu import Connection, Consumer, Exchange, Producer, Queue, pools

class AMQP:
    """App AMQP API: app.amqp."""

    Connection = Connection

然后使用我们熟悉的Queue，Consumer，Producer进行消息的生成和消费:

def Queues(self, queues, create_missing=None,
           autoexchange=None, max_priority=None):
    ...
    return self.Queues(
            queues, self.default_exchange, create_missing,
            autoexchange, max_priority, default_routing_key,
        )

def TaskConsumer(self, channel, queues=None, accept=None, **kw):
    ...
    return self.Consumer(
        channel, accept=accept,
        queues=queues or list(self.queues.consume_from.values()),
        **kw
    )

def _create_task_sender(self):
    ...
    producer.publish(
                body,
                exchange=exchange,
                routing_key=routing_key,
                serializer=serializer or default_serializer,
                compression=compression or default_compressor,
                retry=retry, retry_policy=_rp,
                delivery_mode=delivery_mode, declare=declare,
                headers=headers2,
                **properties
            )
    ...

celery-app的两大功能，管理task和管理AMQP连接，我们有一个大概的了解。

worker模式启动流程

worker模式启动在WorkController中，将服务分成不同的阶段，然后将各个阶段组装成一个叫做蓝图(Blueprint)的方式进行管理:

class WorkController:
    # 内部类
    class Blueprint(bootsteps.Blueprint):
        """Worker bootstep blueprint."""

        name = 'Worker'
        default_steps = {
            'celery.worker.components:Hub',
            'celery.worker.components:Pool',
            'celery.worker.components:Beat',
            'celery.worker.components:Timer',
            'celery.worker.components:StateDB',
            'celery.worker.components:Consumer',
            'celery.worker.autoscale:WorkerComponent',
        }

    def __init__(self, app=None, hostname=None, **kwargs):
        self.blueprint = self.Blueprint(
            steps=self.app.steps['worker'],
            on_start=self.on_start,
            on_close=self.on_close,
            on_stopped=self.on_stopped,
        )
        self.blueprint.apply(self, **kwargs)

启动蓝图:

def start(self):
    try:
        # 启动worker
        self.blueprint.start(self)
    except WorkerTerminate:
        self.terminate()
    except Exception as exc:
        logger.critical('Unrecoverable error: %r', exc, exc_info=True)
        self.stop(exitcode=EX_FAILURE)
    except SystemExit as exc:
        self.stop(exitcode=exc.code)
    except KeyboardInterrupt:
        self.stop(exitcode=EX_FAILURE)

启动步骤，比较简单，大概代码如下:

class StepType(type):
    """Meta-class for steps."""

    name = None
    requires = None

class Step(metaclass=StepType):
    ...

    def instantiate(self, name, *args, **kwargs):
        return symbol_by_name(name)(*args, **kwargs)

    def include_if(self, parent):
        return self.enabled

    def _should_include(self, parent):
        if self.include_if(parent):
            return True, self.create(parent)
        return False, None

    def create(self, parent):
        """Create the step."""

从Step大概可以看出:

• 每个步骤，可以有依赖requires
• 每个步骤，可以有具体的动作instantiate
• 步骤具有树状的父子结构，可以自动创建上级步骤

比如一个消费者步骤, 依赖Connection步骤。启动的时候对Connection进行消费。两者代码如下:

class ConsumerStep(StartStopStep):
    """Bootstep that starts a message consumer."""

    requires = ('celery.worker.consumer:Connection',)
    consumers = None

    def start(self, c):
        channel = c.connection.channel()
        self.consumers = self.get_consumers(channel)
        for consumer in self.consumers or []:
            consumer.consume()

class Connection(bootsteps.StartStopStep):
    """Service managing the consumer broker connection."""

    def __init__(self, c, **kwargs):
        c.connection = None
        super().__init__(c, **kwargs)

    def start(self, c):
        c.connection = c.connect()
        info('Connected to %s', c.connection.as_uri())

在Blueprint中创建和管理这些step:

class Blueprint:

    def __init__(self, steps=None, name=None,
                 on_start=None, on_close=None, on_stopped=None):
        self.name = name or self.name or qualname(type(self))
        # 并集
        self.types = set(steps or []) | set(self.default_steps)
        ...
        self.steps = {}

    def apply(self, parent, **kwargs):
        steps = self.steps = dict(symbol_by_name(step) for step in self.types)

        self._debug('Building graph...')
        for S in self._finalize_steps(steps):
            step = S(parent, **kwargs)
            steps[step.name] = step
            order.append(step)
        self._debug('New boot order: {%s}',
                    ', '.join(s.alias for s in self.order))
        for step in order:
            step.include(parent)
        return self

启动Blueprint:

def start(self, parent):
    self.state = RUN
    if self.on_start:
        self.on_start()
    for i, step in enumerate(s for s in parent.steps if s is not None):
        self._debug('Starting %s', step.alias)
        self.started = i + 1
        step.start(parent)
        logger.debug('^-- substep ok')

通过将启动过程拆分成多个step单元，然后组合单元构建成graph，逐一启动。

小结

本篇我们正式学习了一下celery的使用流程，了解celery如果使用redis作为broker，利用服务作为消费者，使用客户端作为生成者，完成一次远程任务的执行。简单探索worker服务模式的启动流程，重点分析celery-application的管理task和管理连接两大功能实现。

小技巧

celery中展示了一种动态创建类和对象的方法:

task = type(fun.__name__, (Task,), dict({
                'app': self,
                'name': name,
                'run': run,
                '_decorated': True,
                '__doc__': fun.__doc__,
                '__module__': fun.__module__,
                '__annotations__': fun.__annotations__,
                '__header__': staticmethod(head_from_fun(fun, bound=bind)),
                '__wrapped__': run}, **options))()

通过type函数创了一个动态的task子类，然后执行 () 实例化一个task子对象。

参考链接

• 以编程方式定义类 https://python3-cookbook.readthedocs.io/zh_CN/latest/c09/p18_define_classes_programmatically.html

本文由哈喽比特于3年以前收录，如有侵权请联系我们。
文章来源：https://mp.weixin.qq.com/s/SM4P07BVdhXwasAE9KmkYw