官网中关于ReactorNotRestartable的错误描述（摘自：https://twistedmatrix.com/documents/16.1.0/api/twisted.internet.error.html），我们将从scrapy源码分析这个问题
重点要了解scrapy源码下的crawler.py模块的三个类，这三个类是Scrapy的启动核心代码， 由于Scrapy是基于Twisted(一个python网络编程框架)的事件循环写的异步爬虫框架，
所以需要对Twisted模块有一定了解。这里不展开描写了。
- Crawler类： 启动scrapy的爬虫引擎并开始爬取
- CrawlerRunner类：是Crawler子类， 将设置传递给Crawler类并启动
- CrawlerProcesser类： 是CrawlerRunner子类，启动并且控制Twisted的reactor参数，这个reactor参数是全局变量，有且只能有一个，且只能在主线程中运行
import six
import signal
import logging
import warnings
import sys
from twisted.internet import reactor, defer
from zope.interface.verify import verifyClass, DoesNotImplement
from scrapy import Spider
from scrapy.core.engine import ExecutionEngine
from scrapy.resolver import CachingThreadedResolver
from scrapy.interfaces import ISpiderLoader
from scrapy.extension import ExtensionManager
from scrapy.settings import overridden_settings, Settings
from scrapy.signalmanager import SignalManager
from scrapy.exceptions import ScrapyDeprecationWarning
from scrapy.utils.ossignal import install_shutdown_handlers, signal_names
from scrapy.utils.misc import load_object
from scrapy.utils.log import (
LogCounterHandler, configure_logging, log_scrapy_info,
get_scrapy_root_handler, install_scrapy_root_handler)
from scrapy import signals
logger = logging.getLogger(__name__)
class Crawler(object):
def __init__(self, spidercls, settings=None):
if isinstance(spidercls, Spider):
raise ValueError(
'The spidercls argument must be a class, not an object')
if isinstance(settings, dict) or settings is None:
settings = Settings(settings)
self.spidercls = spidercls
self.settings = settings.copy()
self.spidercls.update_settings(self.settings)
self.signals = SignalManager(self)
self.stats = load_object(self.settings['STATS_CLASS'])(self)
handler = LogCounterHandler(self, level=self.settings.get('LOG_LEVEL'))
logging.root.addHandler(handler)
d = dict(overridden_settings(self.settings))
logger.info("Overridden settings: %(settings)r", {'settings': d})
if get_scrapy_root_handler() is not None:
# scrapy root handler already installed: update it with new settings
install_scrapy_root_handler(self.settings)
# lambda is assigned to Crawler attribute because this way it is not
# garbage collected after leaving __init__ scope
self.__remove_handler = lambda: logging.root.removeHandler(handler)
self.signals.connect(self.__remove_handler, signals.engine_stopped)
lf_cls = load_object(self.settings['LOG_FORMATTER'])
self.logformatter = lf_cls.from_crawler(self)
self.extensions = ExtensionManager.from_crawler(self)
self.settings.freeze()
self.crawling = False
self.spider = None
self.engine = None
@property
def spiders(self):
if not hasattr(self, '_spiders'):
warnings.warn("Crawler.spiders is deprecated, use "
"CrawlerRunner.spider_loader or instantiate "
"scrapy.spiderloader.SpiderLoader with your "
"settings.",
category=ScrapyDeprecationWarning, stacklevel=2)
self._spiders = _get_spider_loader(self.settings.frozencopy())
return self._spiders
@defer.inlineCallbacks
def crawl(self, *args, **kwargs):
assert not self.crawling, "Crawling already taking place"
self.crawling = True
try:
self.spider = self._create_spider(*args, **kwargs)
self.engine = self._create_engine()
start_requests = iter(self.spider.start_requests())
yield self.engine.open_spider(self.spider, start_requests)
yield defer.maybeDeferred(self.engine.start)
except Exception:
# In Python 2 reraising an exception after yield discards
# the original traceback (see https://bugs.python.org/issue7563),
# so sys.exc_info() workaround is used.
# This workaround also works in Python 3, but it is not needed,
# and it is slower, so in Python 3 we use native `raise`.
if six.PY2:
exc_info = sys.exc_info()
self.crawling = False
if self.engine is not None:
yield self.engine.close()
if six.PY2:
six.reraise(*exc_info)
raise
def _create_spider(self, *args, **kwargs):
return self.spidercls.from_crawler(self, *args, **kwargs)
def _create_engine(self):
return ExecutionEngine(self, lambda _: self.stop())
@defer.inlineCallbacks
def stop(self):
"""Starts a graceful stop of the crawler and returns a deferred that is
fired when the crawler is stopped."""
if self.crawling:
self.crawling = False
yield defer.maybeDeferred(self.engine.stop)
class CrawlerRunner(object):
"""
This is a convenient helper class that keeps track of, manages and runs
crawlers inside an already setup Twisted `reactor`_.
The CrawlerRunner object must be instantiated with a
:class:`~scrapy.settings.Settings` object.
This class shouldn't be needed (since Scrapy is responsible of using it
accordingly) unless writing scripts that manually handle the crawling
process. See :ref:`run-from-script` for an example.
"""
crawlers = property(
lambda self: self._crawlers,
doc="Set of :class:`crawlers <scrapy.crawler.Crawler>` started by "
":meth:`crawl` and managed by this class."
)
def __init__(self, settings=None):
if isinstance(settings, dict) or settings is None:
settings = Settings(settings)
self.settings = settings
self.spider_loader = _get_spider_loader(settings)
self._crawlers = set()
self._active = set()
self.bootstrap_failed = False
@property
def spiders(self):
warnings.warn("CrawlerRunner.spiders attribute is renamed to "
"CrawlerRunner.spider_loader.",
category=ScrapyDeprecationWarning, stacklevel=2)
return self.spider_loader
def crawl(self, crawler_or_spidercls, *args, **kwargs):
"""
Run a crawler with the provided arguments.
It will call the given Crawler's :meth:`~Crawler.crawl` method, while
keeping track of it so it can be stopped later.
If ``crawler_or_spidercls`` isn't a :class:`~scrapy.crawler.Crawler`
instance, this method will try to create one using this parameter as
the spider class given to it.
Returns a deferred that is fired when the crawling is finished.
:param crawler_or_spidercls: already created crawler, or a spider class
or spider's name inside the project to create it
:type crawler_or_spidercls: :class:`~scrapy.crawler.Crawler` instance,
:class:`~scrapy.spiders.Spider` subclass or string
:param list args: arguments to initialize the spider
:param dict kwargs: keyword arguments to initialize the spider
"""
if isinstance(crawler_or_spidercls, Spider):
raise ValueError(
'The crawler_or_spidercls argument cannot be a spider object, '
'it must be a spider class (or a Crawler object)')
crawler = self.create_crawler(crawler_or_spidercls)
return self._crawl(crawler, *args, **kwargs)
def _crawl(self, crawler, *args, **kwargs):
self.crawlers.add(crawler)
d = crawler.crawl(*args, **kwargs)
self._active.add(d)
def _done(result):
self.crawlers.discard(crawler)
self._active.discard(d)
self.bootstrap_failed |= not getattr(crawler, 'spider', None)
return result
return d.addBoth(_done)
def create_crawler(self, crawler_or_spidercls):
"""
Return a :class:`~scrapy.crawler.Crawler` object.
* If ``crawler_or_spidercls`` is a Crawler, it is returned as-is.
* If ``crawler_or_spidercls`` is a Spider subclass, a new Crawler
is constructed for it.
* If ``crawler_or_spidercls`` is a string, this function finds
a spider with this name in a Scrapy project (using spider loader),
then creates a Crawler instance for it.
"""
if isinstance(crawler_or_spidercls, Spider):
raise ValueError(
'The crawler_or_spidercls argument cannot be a spider object, '
'it must be a spider class (or a Crawler object)')
if isinstance(crawler_or_spidercls, Crawler):
return crawler_or_spidercls
return self._create_crawler(crawler_or_spidercls)
def _create_crawler(self, spidercls):
if isinstance(spidercls, six.string_types):
spidercls = self.spider_loader.load(spidercls)
return Crawler(spidercls, self.settings)
def stop(self):
"""
Stops simultaneously all the crawling jobs taking place.
Returns a deferred that is fired when they all have ended.
"""
return defer.DeferredList([c.stop() for c in list(self.crawlers)])
@defer.inlineCallbacks
def join(self):
"""
join()
Returns a deferred that is fired when all managed :attr:`crawlers` have
completed their executions.
"""
while self._active:
yield defer.DeferredList(self._active)
class CrawlerProcess(CrawlerRunner):
"""
A class to run multiple scrapy crawlers in a process simultaneously.
This class extends :class:`~scrapy.crawler.CrawlerRunner` by adding support
for starting a Twisted `reactor`_ and handling shutdown signals, like the
keyboard interrupt command Ctrl-C. It also configures top-level logging.
This utility should be a better fit than
:class:`~scrapy.crawler.CrawlerRunner` if you aren't running another
Twisted `reactor`_ within your application.
The CrawlerProcess object must be instantiated with a
:class:`~scrapy.settings.Settings` object.
:param install_root_handler: whether to install root logging handler
(default: True)
This class shouldn't be needed (since Scrapy is responsible of using it
accordingly) unless writing scripts that manually handle the crawling
process. See :ref:`run-from-script` for an example.
"""
def __init__(self, settings=None, install_root_handler=True):
super(CrawlerProcess, self).__init__(settings)
install_shutdown_handlers(self._signal_shutdown)
configure_logging(self.settings, install_root_handler)
log_scrapy_info(self.settings)
def _signal_shutdown(self, signum, _):
install_shutdown_handlers(self._signal_kill)
signame = signal_names[signum]
logger.info("Received %(signame)s, shutting down gracefully. Send again to force ",
{'signame': signame})
reactor.callFromThread(self._graceful_stop_reactor)
def _signal_kill(self, signum, _):
install_shutdown_handlers(signal.SIG_IGN)
signame = signal_names[signum]
logger.info('Received %(signame)s twice, forcing unclean shutdown',
{'signame': signame})
reactor.callFromThread(self._stop_reactor)
def start(self, stop_after_crawl=True):
"""
This method starts a Twisted `reactor`_, adjusts its pool size to
:setting:`REACTOR_THREADPOOL_MAXSIZE`, and installs a DNS cache based
on :setting:`DNSCACHE_ENABLED` and :setting:`DNSCACHE_SIZE`.
If ``stop_after_crawl`` is True, the reactor will be stopped after all
crawlers have finished, using :meth:`join`.
:param boolean stop_after_crawl: stop or not the reactor when all
crawlers have finished
"""
if stop_after_crawl:
d = self.join()
# Don't start the reactor if the deferreds are already fired
if d.called:
return
d.addBoth(self._stop_reactor)
reactor.installResolver(self._get_dns_resolver())
tp = reactor.getThreadPool()
tp.adjustPoolsize(maxthreads=self.settings.getint('REACTOR_THREADPOOL_MAXSIZE'))
reactor.addSystemEventTrigger('before', 'shutdown', self.stop)
reactor.run(installSignalHandlers=False) # blocking call
def _get_dns_resolver(self):
if self.settings.getbool('DNSCACHE_ENABLED'):
cache_size = self.settings.getint('DNSCACHE_SIZE')
else:
cache_size = 0
return CachingThreadedResolver(
reactor=reactor,
cache_size=cache_size,
timeout=self.settings.getfloat('DNS_TIMEOUT')
)
def _graceful_stop_reactor(self):
d = self.stop()
d.addBoth(self._stop_reactor)
return d
def _stop_reactor(self, _=None):
try:
reactor.stop()
except RuntimeError: # raised if already stopped or in shutdown stage
pass
def _get_spider_loader(settings):
""" Get SpiderLoader instance from settings """
cls_path = settings.get('SPIDER_LOADER_CLASS')
loader_cls = load_object(cls_path)
try:
verifyClass(ISpiderLoader, loader_cls)
except DoesNotImplement:
warnings.warn(
'SPIDER_LOADER_CLASS (previously named SPIDER_MANAGER_CLASS) does '
'not fully implement scrapy.interfaces.ISpiderLoader interface. '
'Please add all missing methods to avoid unexpected runtime errors.',
category=ScrapyDeprecationWarning, stacklevel=2
)
return loader_cls.from_settings(settings.frozencopy())
我是在djcelery项目中调用scrapy，是在worker的线程中调用，直接使用CrawlerProcess，会报错，因为CrawlerProcess自带reactor的启动关闭过程，而这个过程是在其他线程中发生的，
所以重复运行会报 ReactorNotRestartable、ReactorNotRestartable、ReactorNotRunning等一系列问题，只能使用CrawlerRunner类来启动Scrapy，但是如果按照官网例子（如下）直接运行：
from twisted.internet import reactor
import scrapy
from scrapy.crawler import CrawlerRunner
from scrapy.utils.log import configure_logging
class MySpider(scrapy.Spider):
# Your spider definition
...
configure_logging({'LOG_FORMAT': '%(levelname)s: %(message)s'})
runner = CrawlerRunner()
d = runner.crawl(MySpider)
d.addBoth(lambda _: reactor.stop()) # twisted的回调函数，当爬虫运行结束时会自动调用并且stop reactor
reactor.run() # the script will block here until the crawling is finished
这个在Scrapy项目里面运行(主线程运行)没有问题，但是要在djcelery中的worker线程中运行会报错ReactorNotRestartable，表示reactor已经关闭了不能重新启动，这个问题不知道后续twisted会不会解决掉，目前需要屏蔽掉下面这两行代码，
d.addBoth(lambda _: reactor.stop()) reactor.run() # the script will block here until the crawling is finished
但是如果不启动reactor又不行，又不能重复启动reactor，可以用到另一个模块 Crochet，下面是他的官网描述，写的很清楚，可以代替我们运行reactor和启动事件循环（主要是在setup()函数中），并且全程不需要start、stop reactor所以避免了上面类似的问题
Use Twisted Anywhere! Crochet is an MIT-licensed library that makes it easier for blocking and threaded applications like Flask or Django to use the Twisted networking framework. Here’s an example of a program using Crochet. Notice that you get a completely blocking interface to Twisted and do not need to run the Twisted reactor, the event loop, yourself.
综合以上，再其他线程中调用Twisted reactor，可以如下调用，reactor会在setup函数中自动调用，并全程使用：
from crochet import setup setup() from scrapy.utils.project import get_project_settings from scrapy.crawler import CrawlerRunner runner = CrawlerRunner(get_project_settings()) runner.crawl(SpiderName) d = runner.join()