예제 #1
0
class Spider(object):
    """
    Asynchronous scraping framework.
    """
    spider_name = None

    # You can define here some urls and initial tasks
    # with name "initial" will be created from these
    # urls
    # If the logic of generating initial tasks is complex
    # then consider to use `task_generator` method instead of
    # `initial_urls` attribute
    initial_urls = []

    class Meta:
        # pylint: disable=no-init
        #
        # Meta.abstract means that this class will not be
        # collected to spider registry by `grab crawl` CLI command.
        # The Meta is inherited by descendant classes BUT
        # Meta.abstract is reset to False in each descendant
        abstract = True

    # *************
    # Class Methods
    # *************

    @classmethod
    def update_spider_config(cls, config):
        pass

    @classmethod
    def get_spider_name(cls):
        if cls.spider_name:
            return cls.spider_name
        else:
            return camel_case_to_underscore(cls.__name__)

    # **************
    # Public Methods
    # **************

    def __init__(
            self,
            thread_number=None,
            network_try_limit=None, task_try_limit=None,
            request_pause=NULL,
            priority_mode='random',
            meta=None,
            config=None,
            args=None,
            parser_requests_per_process=10000,
            parser_pool_size=1,
            http_api_port=None,
            network_service='threaded',
            grab_transport='pycurl',
            # Deprecated
            transport=None,
            only_cache=False,
        ):
        """
        Arguments:
        * thread-number - Number of concurrent network streams
        * network_try_limit - How many times try to send request
            again if network error was occurred, use 0 to disable
        * task_try_limit - Limit of tries to execute some task
            this is not the same as network_try_limit
            network try limit limits the number of tries which
            are performed automatically in case of network timeout
            of some other physical error
            but task_try_limit limits the number of attempts which
            are scheduled manually in the spider business logic
        * priority_mode - could be "random" or "const"
        * meta - arbitrary user data
        * retry_rebuild_user_agent - generate new random user-agent for each
            network request which is performed again due to network error
        * args - command line arguments parsed with `setup_arg_parser` method
        """

        self.fatal_error_queue = Queue()
        self.task_queue_parameters = None
        self.http_api_port = http_api_port
        self._started = None
        assert grab_transport in ('pycurl', 'urllib3')
        self.grab_transport_name = grab_transport
        self.parser_requests_per_process = parser_requests_per_process
        self.stat = Stat()
        self.task_queue = None
        if args is None:
            self.args = {}
        else:
            self.args = args
        if config is not None:
            self.config = config
        else:
            self.config = {}
        if meta:
            self.meta = meta
        else:
            self.meta = {}
        self.thread_number = (
            thread_number or
            int(self.config.get('thread_number',
                                DEFAULT_NETWORK_STREAM_NUMBER)))
        self.task_try_limit = (
            task_try_limit or
            int(self.config.get('task_try_limit', DEFAULT_TASK_TRY_LIMIT)))
        self.network_try_limit = (
            network_try_limit or
            int(self.config.get('network_try_limit',
                                DEFAULT_NETWORK_TRY_LIMIT)))
        self._grab_config = {}
        if priority_mode not in ['random', 'const']:
            raise SpiderMisuseError('Value of priority_mode option should be '
                                    '"random" or "const"')
        else:
            self.priority_mode = priority_mode
        if only_cache:
            raise_feature_is_deprecated('Cache feature')
        self.work_allowed = True
        if request_pause is not NULL:
            warn('Option `request_pause` is deprecated and is not '
                 'supported anymore')
        self.proxylist_enabled = None
        self.proxylist = None
        self.proxy = None
        self.proxy_auto_change = False
        self.interrupted = False
        self.parser_pool_size = parser_pool_size
        self.parser_service = ParserService(
            spider=self,
            pool_size=self.parser_pool_size,
        )
        if transport is not None:
            warn('The "transport" argument of Spider constructor is'
                 ' deprecated. Use "network_service" argument.')
            network_service = transport
        assert network_service in ('threaded',)
        if network_service == 'threaded':
            # pylint: disable=no-name-in-module, import-error
            from grab.spider.network_service.threaded import (
                NetworkServiceThreaded
            )
            self.network_service = NetworkServiceThreaded(
                self, self.thread_number
            )
        self.task_dispatcher = TaskDispatcherService(self)
        if self.http_api_port:
            self.http_api_service = HttpApiService(self)
        else:
            self.http_api_service = None
        self.task_generator_service = TaskGeneratorService(
            self.task_generator(), self,
        )

    def setup_cache(self, *args, **kwargs):
        raise_feature_is_deprecated('Cache feature')

    def setup_queue(self, backend='memory', **kwargs):
        """
        Setup queue.

        :param backend: Backend name
            Should be one of the following: 'memory', 'redis' or 'mongo'.
        :param kwargs: Additional credentials for backend.
        """
        if backend == 'mongo':
            warn('Backend name "mongo" is deprecated. Use "mongodb" instead.')
            backend = 'mongodb'
        logger.debug('Using %s backend for task queue', backend)
        mod = __import__('grab.spider.queue_backend.%s' % backend,
                         globals(), locals(), ['foo'])
        self.task_queue = mod.QueueBackend(spider_name=self.get_spider_name(),
                                           **kwargs)

    def add_task(self, task, queue=None, raise_error=False):
        """
        Add task to the task queue.
        """

        if queue is None:
            queue = self.task_queue
        if queue is None:
            raise SpiderMisuseError('You should configure task queue before '
                                    'adding tasks. Use `setup_queue` method.')
        if task.priority is None or not task.priority_set_explicitly:
            task.priority = self.generate_task_priority()
            task.priority_set_explicitly = False
        else:
            task.priority_set_explicitly = True

        if not task.url.startswith(('http://', 'https://', 'ftp://',
                                    'file://', 'feed://')):
            self.stat.collect('task-with-invalid-url', task.url)
            msg = 'Invalid task URL: %s' % task.url
            if raise_error:
                raise SpiderError(msg)
            else:
                logger.error(
                    '%s\nTraceback:\n%s', msg, ''.join(format_stack()),
                )
                return False
        else:
            # TODO: keep original task priority if it was set explicitly
            # WTF the previous comment means?
            queue.put(
                task, priority=task.priority, schedule_time=task.schedule_time
            )
            return True

    def stop(self):
        """
        This method set internal flag which signal spider
        to stop processing new task and shuts down.
        """
        self.work_allowed = False

    def load_proxylist(self, source, source_type=None, proxy_type='http',
                       auto_init=True, auto_change=True):
        """
        Load proxy list.

        :param source: Proxy source.
            Accepts string (file path, url) or ``BaseProxySource`` instance.
        :param source_type: The type of the specified source.
            Should be one of the following: 'text_file' or 'url'.
        :param proxy_type:
            Should be one of the following: 'socks4', 'socks5' or'http'.
        :param auto_change:
            If set to `True` then automatical random proxy rotation
            will be used.


        Proxy source format should be one of the following (for each line):
            - ip:port
            - ip:port:login:password

        """
        self.proxylist = ProxyList()
        if isinstance(source, BaseProxySource):
            self.proxylist.set_source(source)
        elif isinstance(source, six.string_types):
            if source_type == 'text_file':
                self.proxylist.load_file(source, proxy_type=proxy_type)
            elif source_type == 'url':
                self.proxylist.load_url(source, proxy_type=proxy_type)
            else:
                raise SpiderMisuseError('Method `load_proxylist` received '
                                        'invalid `source_type` argument: %s'
                                        % source_type)
        else:
            raise SpiderMisuseError('Method `load_proxylist` received '
                                    'invalid `source` argument: %s'
                                    % source)

        self.proxylist_enabled = True
        self.proxy = None
        if not auto_change and auto_init:
            self.proxy = self.proxylist.get_random_proxy()
        self.proxy_auto_change = auto_change

    def process_next_page(self, grab, task, xpath,
                          resolve_base=False, **kwargs):
        """
        Generate task for next page.

        :param grab: Grab instance
        :param task: Task object which should be assigned to next page url
        :param xpath: xpath expression which calculates list of URLS
        :param **kwargs: extra settings for new task object

        Example::

            self.follow_links(grab, 'topic', '//div[@class="topic"]/a/@href')
        """
        try:
            # next_url = grab.xpath_text(xpath)
            next_url = grab.doc.select(xpath).text()
        except IndexError:
            return False
        else:
            url = grab.make_url_absolute(next_url, resolve_base=resolve_base)
            page = task.get('page', 1) + 1
            grab2 = grab.clone()
            grab2.setup(url=url)
            task2 = task.clone(task_try_count=1, grab=grab2,
                               page=page, **kwargs)
            self.add_task(task2)
            return True

    def render_stats(self, timing=None):
        if timing is not None:
            warn('Option timing of method render_stats is deprecated.'
                 ' There is no more timing feature.')
        out = ['------------ Stats: ------------']
        out.append('Counters:')

        # Process counters
        items = sorted(self.stat.counters.items(),
                       key=lambda x: x[0], reverse=True)
        for item in items:
            out.append('  %s: %s' % item)
        out.append('')

        out.append('Lists:')
        # Process collections sorted by size desc
        col_sizes = [(x, len(y)) for x, y in self.stat.collections.items()]
        col_sizes = sorted(col_sizes, key=lambda x: x[1], reverse=True)
        for col_size in col_sizes:
            out.append('  %s: %d' % col_size)
        out.append('')

        # Process extra metrics
        if 'download-size' in self.stat.counters:
            out.append('Network download: %s' %
                       metric.format_traffic_value(
                           self.stat.counters['download-size']))
        out.append('Queue size: %d' % self.task_queue.size()
                   if self.task_queue else 'NA')
        out.append('Network streams: %d' % self.thread_number)
        if self._started:
            elapsed = time.time() - self._started
        else:
            elapsed = 0
        hours, seconds = divmod(elapsed, 3600)
        minutes, seconds = divmod(seconds, 60)
        out.append('Time elapsed: %d:%d:%d (H:M:S)' % (
            hours, minutes, seconds))
        out.append('End time: %s' %
                   datetime.utcnow().strftime('%d %b %Y, %H:%M:%S UTC'))
        return '\n'.join(out) + '\n'

    # ********************************
    # Methods for spider customization
    # ********************************

    def prepare(self):
        """
        You can do additional spider customization here
        before it has started working. Simply redefine
        this method in your Spider class.
        """

    def shutdown(self):
        """
        You can override this method to do some final actions
        after parsing has been done.
        """

        pass

    def update_grab_instance(self, grab):
        """
        Use this method to automatically update config of any
        `Grab` instance created by the spider.
        """
        pass

    def create_grab_instance(self, **kwargs):
        # Back-ward compatibility for deprecated `grab_config` attribute
        # Here I use `_grab_config` to not trigger warning messages
        kwargs['transport'] = self.grab_transport_name
        if self._grab_config and kwargs:
            merged_config = deepcopy(self._grab_config)
            merged_config.update(kwargs)
            grab = Grab(**merged_config)
        elif self._grab_config and not kwargs:
            grab = Grab(**self._grab_config)
        else:
            grab = Grab(**kwargs)
        return grab

    def task_generator(self):
        """
        You can override this method to load new tasks smoothly.

        It will be used each time as number of tasks
        in task queue is less then number of threads multiplied on 2
        This allows you to not overload all free memory if total number of
        tasks is big.
        """

        if False: # pylint: disable=using-constant-test
            # Some magic to make this function empty generator
            yield ':-)'
        return

    # ***************
    # Private Methods
    # ***************

    def check_task_limits(self, task):
        """
        Check that task's network & try counters do not exceed limits.

        Returns:
        * if success: (True, None)
        * if error: (False, reason)

        """

        if task.task_try_count > self.task_try_limit:
            return False, 'task-try-count'

        if task.network_try_count > self.network_try_limit:
            return False, 'network-try-count'

        return True, None

    def generate_task_priority(self):
        if self.priority_mode == 'const':
            return DEFAULT_TASK_PRIORITY
        else:
            return randint(*RANDOM_TASK_PRIORITY_RANGE)

    def process_initial_urls(self):
        if self.initial_urls:
            for url in self.initial_urls:
                self.add_task(Task('initial', url=url))

    def get_task_from_queue(self):
        try:
            return self.task_queue.get()
        except Empty:
            size = self.task_queue.size()
            if size:
                return True
            else:
                return None

    def setup_grab_for_task(self, task):
        grab = self.create_grab_instance()
        if task.grab_config:
            grab.load_config(task.grab_config)
        else:
            grab.setup(url=task.url)

        # Generate new common headers
        grab.config['common_headers'] = grab.common_headers()
        self.update_grab_instance(grab)
        grab.setup_transport(self.grab_transport_name)
        return grab

    def is_valid_network_response_code(self, code, task):
        """
        Answer the question: if the response could be handled via
        usual task handler or the task failed and should be processed as error.
        """

        return (code < 400 or code == 404 or
                code in task.valid_status)

    def process_parser_error(self, func_name, task, exc_info):
        _, ex, _ = exc_info
        self.stat.inc('spider:error-%s' % ex.__class__.__name__.lower())

        logger.error(
            'Task handler [%s] error\n%s',
            func_name,
            ''.join(format_exception(*exc_info)),
        )

        # Looks strange but I really have some problems with
        # serializing exception into string
        try:
            ex_str = six.text_type(ex)
        except TypeError:
            try:
                ex_str = ex.decode('utf-8', 'ignore')
            except TypeError:
                ex_str = str(ex)

        task_url = task.url if task else None
        self.stat.collect('fatal', '%s|%s|%s|%s' % (
            func_name, ex.__class__.__name__, ex_str, task_url
        ))

    def find_task_handler(self, task):
        callback = task.get('callback')
        if callback:
            return callback
        else:
            try:
                handler = getattr(self, 'task_%s' % task.name)
            except AttributeError:
                raise NoTaskHandler('No handler or callback defined for '
                                    'task %s' % task.name)
            else:
                return handler

    def log_network_result_stats(self, res, task):
        # Increase stat counters
        self.stat.inc('spider:request-processed')
        self.stat.inc('spider:task')
        self.stat.inc('spider:task-%s' % task.name)
        if (task.network_try_count == 1 and
                task.task_try_count == 1):
            self.stat.inc('spider:task-%s-initial' % task.name)

        # Update traffic statistics
        if res['grab'] and res['grab'].doc:
            doc = res['grab'].doc
            self.stat.inc('spider:download-size', doc.download_size)
            self.stat.inc('spider:upload-size', doc.upload_size)

    def process_grab_proxy(self, task, grab):
        """Assign new proxy from proxylist to the task"""

        if task.use_proxylist:
            if self.proxylist_enabled:
                if self.proxy_auto_change:
                    self.change_active_proxy(task, grab)
                if self.proxy:
                    grab.setup(proxy=self.proxy.get_address(),
                               proxy_userpwd=self.proxy.get_userpwd(),
                               proxy_type=self.proxy.proxy_type)

    # pylint: disable=unused-argument
    def change_active_proxy(self, task, grab):
        self.proxy = self.proxylist.get_random_proxy()
    # pylint: enable=unused-argument

    def submit_task_to_transport(self, task, grab):
        grab_config_backup = grab.dump_config()
        self.process_grab_proxy(task, grab)
        self.stat.inc('spider:request-network')
        self.stat.inc('spider:task-%s-network' % task.name)
        try:
            # pylint: disable=no-member
            self.network_service.start_task_processing(
                task, grab, grab_config_backup)
            # pylint: enable=no-member
        except GrabInvalidUrl:
            # TODO: log error
            # TODO: show traceback
            logger.debug('Task %s has invalid URL: %s',
                         task.name, task.url)
            self.stat.collect('invalid-url', task.url)

    def run(self):
        self._started = time.time()
        services = []
        try:
            self.prepare()
            if self.task_queue is None:
                self.setup_queue()
            self.process_initial_urls()
            services = [
                self.task_dispatcher,
                self.task_generator_service,
                self.parser_service,
                self.network_service,
            ]
            if self.http_api_service:
                self.http_api_service.start()
            for srv in services:
                srv.start()
            while self.work_allowed:
                try:
                    exc_info = self.fatal_error_queue.get(True, 0.5)
                except Empty:
                    pass
                else:
                    # The trackeback of fatal error MUST BE
                    # rendered by the sender
                    raise exc_info[1]
                if self.is_idle():
                    for srv in services:
                        srv.pause()
                    if self.is_idle():
                        break
                    for srv in services:
                        srv.resume()
        except KeyboardInterrupt:
            self.interrupted = True
            raise
        finally:
            # TODO:
            if self.task_queue:
                self.task_queue.close()
            #print('Start stopping services')
            for srv in services:
                # Resume service if it has been paused
                # to allow service to process stop signal
                srv.resume()
                srv.stop()
            #print('Called .stop() for all services')
            start = time.time()
            while any(x.is_alive() for x in services):
                time.sleep(0.1)
                if time.time() - start > 10:
                    break
            for srv in services:
                if srv.is_alive():
                    print('The %s has not stopped :(' % srv)
            self.stat.print_progress_line()
            self.shutdown()
            if self.task_queue:
                self.task_queue.clear()
            logger.debug('Work done')

    def is_idle(self):
        return (
            not self.task_generator_service.is_alive()
            and not self.task_queue.size()
            and not self.task_dispatcher.input_queue.qsize()
            and not self.parser_service.input_queue.qsize()
            and not self.parser_service.is_busy()
            and not self.network_service.get_active_threads_number()
            and not self.network_service.is_busy()
        )

    def log_failed_network_result(self, res):
        if res['ok']:
            msg = 'http-%s' % res['grab'].doc.code
        else:
            msg = res['error_abbr']
        self.stat.inc('error:%s' % msg)

    def log_rejected_task(self, task, reason):
        if reason == 'task-try-count':
            self.stat.collect('task-count-rejected',
                              task.url)
        elif reason == 'network-try-count':
            self.stat.collect('network-count-rejected',
                              task.url)
        else:
            raise SpiderError('Unknown response from '
                              'check_task_limits: %s'
                              % reason)

    # ################
    # Deprecated Things
    # #################

    @property
    def cache_reader_service(self):
        raise_feature_is_deprecated('Cache feature')

    @cache_reader_service.setter
    def cache_reader_service(self, val):
        raise_feature_is_deprecated('Cache feature')

    @property
    def cache_writer_service(self):
        raise_feature_is_deprecated('Cache feature')

    @cache_writer_service.setter
    def cache_writer_service(self, val):
        raise_feature_is_deprecated('Cache feature')
예제 #2
0
파일: base.py 프로젝트: lorien/grab
class Spider(object):
    """
    Asynchronous scraping framework.
    """
    spider_name = None

    # You can define here some urls and initial tasks
    # with name "initial" will be created from these
    # urls
    # If the logic of generating initial tasks is complex
    # then consider to use `task_generator` method instead of
    # `initial_urls` attribute
    initial_urls = []

    class Meta:
        # pylint: disable=no-init
        #
        # Meta.abstract means that this class will not be
        # collected to spider registry by `grab crawl` CLI command.
        # The Meta is inherited by descendant classes BUT
        # Meta.abstract is reset to False in each descendant
        abstract = True

    # *************
    # Class Methods
    # *************

    @classmethod
    def update_spider_config(cls, config):
        pass

    @classmethod
    def get_spider_name(cls):
        if cls.spider_name:
            return cls.spider_name
        else:
            return camel_case_to_underscore(cls.__name__)

    # **************
    # Public Methods
    # **************

    def __init__(
            self,
            thread_number=None,
            network_try_limit=None, task_try_limit=None,
            request_pause=NULL,
            priority_mode='random',
            meta=None,
            only_cache=False,
            config=None,
            args=None,
            parser_requests_per_process=10000,
            parser_pool_size=1,
            http_api_port=None,
            network_service='threaded',
            grab_transport='pycurl',
            # Deprecated
            transport=None):
        """
        Arguments:
        * thread-number - Number of concurrent network streams
        * network_try_limit - How many times try to send request
            again if network error was occurred, use 0 to disable
        * task_try_limit - Limit of tries to execute some task
            this is not the same as network_try_limit
            network try limit limits the number of tries which
            are performed automatically in case of network timeout
            of some other physical error
            but task_try_limit limits the number of attempts which
            are scheduled manually in the spider business logic
        * priority_mode - could be "random" or "const"
        * meta - arbitrary user data
        * retry_rebuild_user_agent - generate new random user-agent for each
            network request which is performed again due to network error
        * args - command line arguments parsed with `setup_arg_parser` method
        """

        self.fatal_error_queue = Queue()
        self.task_queue_parameters = None
        self.http_api_port = http_api_port
        self._started = None
        assert grab_transport in ('pycurl', 'urllib3')
        self.grab_transport_name = grab_transport
        self.parser_requests_per_process = parser_requests_per_process
        self.stat = Stat()
        self.task_queue = None
        if args is None:
            self.args = {}
        else:
            self.args = args
        if config is not None:
            self.config = config
        else:
            self.config = {}
        if meta:
            self.meta = meta
        else:
            self.meta = {}
        self.thread_number = (
            thread_number or
            int(self.config.get('thread_number',
                                DEFAULT_NETWORK_STREAM_NUMBER)))
        self.task_try_limit = (
            task_try_limit or
            int(self.config.get('task_try_limit', DEFAULT_TASK_TRY_LIMIT)))
        self.network_try_limit = (
            network_try_limit or
            int(self.config.get('network_try_limit',
                                DEFAULT_NETWORK_TRY_LIMIT)))
        self._grab_config = {}
        if priority_mode not in ['random', 'const']:
            raise SpiderMisuseError('Value of priority_mode option should be '
                                    '"random" or "const"')
        else:
            self.priority_mode = priority_mode
        self.only_cache = only_cache
        self.work_allowed = True
        if request_pause is not NULL:
            warn('Option `request_pause` is deprecated and is not '
                 'supported anymore')
        self.proxylist_enabled = None
        self.proxylist = None
        self.proxy = None
        self.proxy_auto_change = False
        self.interrupted = False
        self.cache_reader_service = None
        self.cache_writer_service = None
        self.parser_pool_size = parser_pool_size
        self.parser_service = ParserService(
            spider=self,
            pool_size=self.parser_pool_size,
        )
        if transport is not None:
            warn('The "transport" argument of Spider constructor is'
                 ' deprecated. Use "network_service" argument.')
            network_service = transport
        assert network_service in ('threaded',)
        if network_service == 'threaded':
            # pylint: disable=no-name-in-module, import-error
            from grab.spider.network_service.threaded import (
                NetworkServiceThreaded
            )
            self.network_service = NetworkServiceThreaded(
                self, self.thread_number
            )
        self.task_dispatcher = TaskDispatcherService(self)
        if self.http_api_port:
            self.http_api_service = HttpApiService(self)
        else:
            self.http_api_service = None
        self.task_generator_service = TaskGeneratorService(
            self.task_generator(), self,
        )

    def setup_cache(self, backend='mongodb', database=None,
                    **kwargs):
        """
        Setup cache.

        :param backend: Backend name
            Should be one of the following: 'mongo', 'mysql' or 'postgresql'.
        :param database: Database name.
        :param kwargs: Additional credentials for backend.

        """
        if database is None:
            raise SpiderMisuseError('setup_cache method requires database '
                                    'option')
        if backend == 'mongo':
            warn('Backend name "mongo" is deprecated. Use "mongodb" instead.')
            backend = 'mongodb'
        mod = __import__('grab.spider.cache_backend.%s' % backend,
                         globals(), locals(), ['foo'])
        backend = mod.CacheBackend(
            database=database, spider=self, **kwargs
        )
        self.cache_reader_service = CacheReaderService(self, backend)
        backend = mod.CacheBackend(
            database=database, spider=self, **kwargs
        )
        self.cache_writer_service = CacheWriterService(self, backend)

    def setup_queue(self, backend='memory', **kwargs):
        """
        Setup queue.

        :param backend: Backend name
            Should be one of the following: 'memory', 'redis' or 'mongo'.
        :param kwargs: Additional credentials for backend.
        """
        if backend == 'mongo':
            warn('Backend name "mongo" is deprecated. Use "mongodb" instead.')
            backend = 'mongodb'
        logger.debug('Using %s backend for task queue', backend)
        mod = __import__('grab.spider.queue_backend.%s' % backend,
                         globals(), locals(), ['foo'])
        self.task_queue = mod.QueueBackend(spider_name=self.get_spider_name(),
                                           **kwargs)

    def add_task(self, task, queue=None, raise_error=False):
        """
        Add task to the task queue.
        """

        if queue is None:
            if self.cache_reader_service:
                queue = self.cache_reader_service.input_queue
            else:
                queue = self.task_queue
        if queue is None:
            raise SpiderMisuseError('You should configure task queue before '
                                    'adding tasks. Use `setup_queue` method.')
        if task.priority is None or not task.priority_set_explicitly:
            task.priority = self.generate_task_priority()
            task.priority_set_explicitly = False
        else:
            task.priority_set_explicitly = True

        if not task.url.startswith(('http://', 'https://', 'ftp://',
                                    'file://', 'feed://')):
            self.stat.collect('task-with-invalid-url', task.url)
            msg = 'Invalid task URL: %s' % task.url
            if raise_error:
                raise SpiderError(msg)
            else:
                logger.error(
                    '%s\nTraceback:\n%s', msg, ''.join(format_stack()),
                )
                return False
        else:
            # TODO: keep original task priority if it was set explicitly
            # WTF the previous comment means?
            queue.put(
                task, priority=task.priority, schedule_time=task.schedule_time
            )
            return True

    def stop(self):
        """
        This method set internal flag which signal spider
        to stop processing new task and shuts down.
        """
        self.work_allowed = False

    def load_proxylist(self, source, source_type=None, proxy_type='http',
                       auto_init=True, auto_change=True):
        """
        Load proxy list.

        :param source: Proxy source.
            Accepts string (file path, url) or ``BaseProxySource`` instance.
        :param source_type: The type of the specified source.
            Should be one of the following: 'text_file' or 'url'.
        :param proxy_type:
            Should be one of the following: 'socks4', 'socks5' or'http'.
        :param auto_change:
            If set to `True` then automatical random proxy rotation
            will be used.


        Proxy source format should be one of the following (for each line):
            - ip:port
            - ip:port:login:password

        """
        self.proxylist = ProxyList()
        if isinstance(source, BaseProxySource):
            self.proxylist.set_source(source)
        elif isinstance(source, six.string_types):
            if source_type == 'text_file':
                self.proxylist.load_file(source, proxy_type=proxy_type)
            elif source_type == 'url':
                self.proxylist.load_url(source, proxy_type=proxy_type)
            else:
                raise SpiderMisuseError('Method `load_proxylist` received '
                                        'invalid `source_type` argument: %s'
                                        % source_type)
        else:
            raise SpiderMisuseError('Method `load_proxylist` received '
                                    'invalid `source` argument: %s'
                                    % source)

        self.proxylist_enabled = True
        self.proxy = None
        if not auto_change and auto_init:
            self.proxy = self.proxylist.get_random_proxy()
        self.proxy_auto_change = auto_change

    def process_next_page(self, grab, task, xpath,
                          resolve_base=False, **kwargs):
        """
        Generate task for next page.

        :param grab: Grab instance
        :param task: Task object which should be assigned to next page url
        :param xpath: xpath expression which calculates list of URLS
        :param **kwargs: extra settings for new task object

        Example::

            self.follow_links(grab, 'topic', '//div[@class="topic"]/a/@href')
        """
        try:
            # next_url = grab.xpath_text(xpath)
            next_url = grab.doc.select(xpath).text()
        except IndexError:
            return False
        else:
            url = grab.make_url_absolute(next_url, resolve_base=resolve_base)
            page = task.get('page', 1) + 1
            grab2 = grab.clone()
            grab2.setup(url=url)
            task2 = task.clone(task_try_count=1, grab=grab2,
                               page=page, **kwargs)
            self.add_task(task2)
            return True

    def render_stats(self, timing=None):
        if timing is not None:
            warn('Option timing of method render_stats is deprecated.'
                 ' There is no more timing feature.')
        out = ['------------ Stats: ------------']
        out.append('Counters:')

        # Process counters
        items = sorted(self.stat.counters.items(),
                       key=lambda x: x[0], reverse=True)
        for item in items:
            out.append('  %s: %s' % item)
        out.append('')

        out.append('Lists:')
        # Process collections sorted by size desc
        col_sizes = [(x, len(y)) for x, y in self.stat.collections.items()]
        col_sizes = sorted(col_sizes, key=lambda x: x[1], reverse=True)
        for col_size in col_sizes:
            out.append('  %s: %d' % col_size)
        out.append('')

        # Process extra metrics
        if 'download-size' in self.stat.counters:
            out.append('Network download: %s' %
                       metric.format_traffic_value(
                           self.stat.counters['download-size']))
        out.append('Queue size: %d' % self.task_queue.size()
                   if self.task_queue else 'NA')
        out.append('Network streams: %d' % self.thread_number)
        if self._started:
            elapsed = time.time() - self._started
        else:
            elapsed = 0
        hours, seconds = divmod(elapsed, 3600)
        minutes, seconds = divmod(seconds, 60)
        out.append('Time elapsed: %d:%d:%d (H:M:S)' % (
            hours, minutes, seconds))
        out.append('End time: %s' %
                   datetime.utcnow().strftime('%d %b %Y, %H:%M:%S UTC'))
        return '\n'.join(out) + '\n'

    # ********************************
    # Methods for spider customization
    # ********************************

    def prepare(self):
        """
        You can do additional spider customization here
        before it has started working. Simply redefine
        this method in your Spider class.
        """

    def shutdown(self):
        """
        You can override this method to do some final actions
        after parsing has been done.
        """

        pass

    def update_grab_instance(self, grab):
        """
        Use this method to automatically update config of any
        `Grab` instance created by the spider.
        """
        pass

    def create_grab_instance(self, **kwargs):
        # Back-ward compatibility for deprecated `grab_config` attribute
        # Here I use `_grab_config` to not trigger warning messages
        kwargs['transport'] = self.grab_transport_name
        if self._grab_config and kwargs:
            merged_config = deepcopy(self._grab_config)
            merged_config.update(kwargs)
            grab = Grab(**merged_config)
        elif self._grab_config and not kwargs:
            grab = Grab(**self._grab_config)
        else:
            grab = Grab(**kwargs)
        return grab

    def task_generator(self):
        """
        You can override this method to load new tasks smoothly.

        It will be used each time as number of tasks
        in task queue is less then number of threads multiplied on 2
        This allows you to not overload all free memory if total number of
        tasks is big.
        """

        if False: # pylint: disable=using-constant-test
            # Some magic to make this function empty generator
            yield ':-)'
        return

    # ***************
    # Private Methods
    # ***************

    def check_task_limits(self, task):
        """
        Check that task's network & try counters do not exceed limits.

        Returns:
        * if success: (True, None)
        * if error: (False, reason)

        """

        if task.task_try_count > self.task_try_limit:
            return False, 'task-try-count'

        if task.network_try_count > self.network_try_limit:
            return False, 'network-try-count'

        return True, None

    def generate_task_priority(self):
        if self.priority_mode == 'const':
            return DEFAULT_TASK_PRIORITY
        else:
            return randint(*RANDOM_TASK_PRIORITY_RANGE)

    def process_initial_urls(self):
        if self.initial_urls:
            for url in self.initial_urls:
                self.add_task(Task('initial', url=url))

    def get_task_from_queue(self):
        try:
            return self.task_queue.get()
        except Empty:
            size = self.task_queue.size()
            if size:
                return True
            else:
                return None

    def setup_grab_for_task(self, task):
        grab = self.create_grab_instance()
        if task.grab_config:
            grab.load_config(task.grab_config)
        else:
            grab.setup(url=task.url)

        # Generate new common headers
        grab.config['common_headers'] = grab.common_headers()
        self.update_grab_instance(grab)
        grab.setup_transport(self.grab_transport_name)
        return grab

    def is_valid_network_response_code(self, code, task):
        """
        Answer the question: if the response could be handled via
        usual task handler or the task failed and should be processed as error.
        """

        return (code < 400 or code == 404 or
                code in task.valid_status)

    def process_parser_error(self, func_name, task, exc_info):
        _, ex, _ = exc_info
        self.stat.inc('spider:error-%s' % ex.__class__.__name__.lower())

        logger.error(
            'Task handler [%s] error\n%s',
            func_name,
            ''.join(format_exception(*exc_info)),
        )

        # Looks strange but I really have some problems with
        # serializing exception into string
        try:
            ex_str = six.text_type(ex)
        except TypeError:
            try:
                ex_str = ex.decode('utf-8', 'ignore')
            except TypeError:
                ex_str = str(ex)

        task_url = task.url if task else None
        self.stat.collect('fatal', '%s|%s|%s|%s' % (
            func_name, ex.__class__.__name__, ex_str, task_url
        ))

    def find_task_handler(self, task):
        callback = task.get('callback')
        if callback:
            return callback
        else:
            try:
                handler = getattr(self, 'task_%s' % task.name)
            except AttributeError:
                raise NoTaskHandler('No handler or callback defined for '
                                    'task %s' % task.name)
            else:
                return handler

    def log_network_result_stats(self, res, task):
        # Increase stat counters
        self.stat.inc('spider:request-processed')
        self.stat.inc('spider:task')
        self.stat.inc('spider:task-%s' % task.name)
        if (task.network_try_count == 1 and
                task.task_try_count == 1):
            self.stat.inc('spider:task-%s-initial' % task.name)

        # Update traffic statistics
        if res['grab'] and res['grab'].doc:
            doc = res['grab'].doc
            if res.get('from_cache'):
                self.stat.inc('spider:download-size-with-cache',
                              doc.download_size)
                self.stat.inc('spider:upload-size-with-cache',
                              doc.upload_size)
            else:
                self.stat.inc('spider:download-size', doc.download_size)
                self.stat.inc('spider:upload-size', doc.upload_size)

    def process_grab_proxy(self, task, grab):
        """Assign new proxy from proxylist to the task"""

        if task.use_proxylist:
            if self.proxylist_enabled:
                if self.proxy_auto_change:
                    self.change_active_proxy(task, grab)
                if self.proxy:
                    grab.setup(proxy=self.proxy.get_address(),
                               proxy_userpwd=self.proxy.get_userpwd(),
                               proxy_type=self.proxy.proxy_type)

    # pylint: disable=unused-argument
    def change_active_proxy(self, task, grab):
        self.proxy = self.proxylist.get_random_proxy()
    # pylint: enable=unused-argument

    def submit_task_to_transport(self, task, grab):
        if self.only_cache:
            self.stat.inc('spider:request-network-disabled-only-cache')
        else:
            grab_config_backup = grab.dump_config()
            self.process_grab_proxy(task, grab)
            self.stat.inc('spider:request-network')
            self.stat.inc('spider:task-%s-network' % task.name)
            try:
                # pylint: disable=no-member
                self.network_service.start_task_processing(
                    task, grab, grab_config_backup)
                # pylint: enable=no-member
            except GrabInvalidUrl:
                # TODO: log error
                # TODO: show traceback
                logger.debug('Task %s has invalid URL: %s',
                             task.name, task.url)
                self.stat.collect('invalid-url', task.url)

    def run(self):
        self._started = time.time()
        services = []
        try:
            self.prepare()
            if self.task_queue is None:
                self.setup_queue()
            self.process_initial_urls()
            services = [
                self.task_dispatcher,
                self.task_generator_service,
                self.parser_service,
                self.network_service,
            ]
            if self.http_api_service:
                self.http_api_service.start()
            if self.cache_reader_service:
                services.insert(0, self.cache_reader_service)
            if self.cache_writer_service:
                services.insert(0, self.cache_writer_service)
            for srv in services:
                srv.start()
            while self.work_allowed:
                try:
                    exc_info = self.fatal_error_queue.get(True, 0.5)
                except Empty:
                    pass
                else:
                    # The trackeback of fatal error MUST BE
                    # rendered by the sender
                    raise exc_info[1]
                if self.is_idle():
                    for srv in services:
                        srv.pause()
                    if self.is_idle():
                        break
                    for srv in services:
                        srv.resume()
        except KeyboardInterrupt:
            self.interrupted = True
            raise
        finally:
            # TODO:
            if self.task_queue:
                self.task_queue.close()
            #print('Start stopping services')
            for srv in services:
                # Resume service if it has been paused
                # to allow service to process stop signal
                srv.resume()
                srv.stop()
            #print('Called .stop() for all services')
            start = time.time()
            while any(x.is_alive() for x in services):
                time.sleep(0.1)
                if time.time() - start > 10:
                    break
            for srv in services:
                if srv.is_alive():
                    print('The %s has not stopped :(' % srv)
            self.stat.print_progress_line()
            self.shutdown()
            if self.task_queue:
                self.task_queue.clear()
            logger.debug('Work done')

    def is_idle(self):
        result = (
            not self.task_generator_service.is_alive()
            and not self.task_queue.size()
            and not self.task_dispatcher.input_queue.qsize()
            and not self.parser_service.input_queue.qsize()
            and not self.parser_service.is_busy()
            and not self.network_service.get_active_threads_number()
            and not self.network_service.is_busy()
        )
        if result and self.cache_reader_service:
            result = result and (
                not self.cache_reader_service.input_queue.size()
                and not self.cache_writer_service.input_queue.qsize()
            )
        return result

    def log_failed_network_result(self, res):
        if res['ok']:
            msg = 'http-%s' % res['grab'].doc.code
        else:
            msg = res['error_abbr']
        self.stat.inc('error:%s' % msg)

    def log_rejected_task(self, task, reason):
        if reason == 'task-try-count':
            self.stat.collect('task-count-rejected',
                              task.url)
        elif reason == 'network-try-count':
            self.stat.collect('network-count-rejected',
                              task.url)
        else:
            raise SpiderError('Unknown response from '
                              'check_task_limits: %s'
                              % reason)
예제 #3
0
    def __init__(
            self,
            thread_number=None,
            network_try_limit=None, task_try_limit=None,
            request_pause=NULL,
            priority_mode='random',
            meta=None,
            config=None,
            args=None,
            parser_requests_per_process=10000,
            parser_pool_size=1,
            http_api_port=None,
            network_service='threaded',
            grab_transport='pycurl',
            # Deprecated
            transport=None,
            only_cache=False,
        ):
        """
        Arguments:
        * thread-number - Number of concurrent network streams
        * network_try_limit - How many times try to send request
            again if network error was occurred, use 0 to disable
        * task_try_limit - Limit of tries to execute some task
            this is not the same as network_try_limit
            network try limit limits the number of tries which
            are performed automatically in case of network timeout
            of some other physical error
            but task_try_limit limits the number of attempts which
            are scheduled manually in the spider business logic
        * priority_mode - could be "random" or "const"
        * meta - arbitrary user data
        * retry_rebuild_user_agent - generate new random user-agent for each
            network request which is performed again due to network error
        * args - command line arguments parsed with `setup_arg_parser` method
        """

        self.fatal_error_queue = Queue()
        self.task_queue_parameters = None
        self.http_api_port = http_api_port
        self._started = None
        assert grab_transport in ('pycurl', 'urllib3')
        self.grab_transport_name = grab_transport
        self.parser_requests_per_process = parser_requests_per_process
        self.stat = Stat()
        self.task_queue = None
        if args is None:
            self.args = {}
        else:
            self.args = args
        if config is not None:
            self.config = config
        else:
            self.config = {}
        if meta:
            self.meta = meta
        else:
            self.meta = {}
        self.thread_number = (
            thread_number or
            int(self.config.get('thread_number',
                                DEFAULT_NETWORK_STREAM_NUMBER)))
        self.task_try_limit = (
            task_try_limit or
            int(self.config.get('task_try_limit', DEFAULT_TASK_TRY_LIMIT)))
        self.network_try_limit = (
            network_try_limit or
            int(self.config.get('network_try_limit',
                                DEFAULT_NETWORK_TRY_LIMIT)))
        self._grab_config = {}
        if priority_mode not in ['random', 'const']:
            raise SpiderMisuseError('Value of priority_mode option should be '
                                    '"random" or "const"')
        else:
            self.priority_mode = priority_mode
        if only_cache:
            raise_feature_is_deprecated('Cache feature')
        self.work_allowed = True
        if request_pause is not NULL:
            warn('Option `request_pause` is deprecated and is not '
                 'supported anymore')
        self.proxylist_enabled = None
        self.proxylist = None
        self.proxy = None
        self.proxy_auto_change = False
        self.interrupted = False
        self.parser_pool_size = parser_pool_size
        self.parser_service = ParserService(
            spider=self,
            pool_size=self.parser_pool_size,
        )
        if transport is not None:
            warn('The "transport" argument of Spider constructor is'
                 ' deprecated. Use "network_service" argument.')
            network_service = transport
        assert network_service in ('threaded',)
        if network_service == 'threaded':
            # pylint: disable=no-name-in-module, import-error
            from grab.spider.network_service.threaded import (
                NetworkServiceThreaded
            )
            self.network_service = NetworkServiceThreaded(
                self, self.thread_number
            )
        self.task_dispatcher = TaskDispatcherService(self)
        if self.http_api_port:
            self.http_api_service = HttpApiService(self)
        else:
            self.http_api_service = None
        self.task_generator_service = TaskGeneratorService(
            self.task_generator(), self,
        )
예제 #4
0
파일: base.py 프로젝트: lorien/grab
    def __init__(
            self,
            thread_number=None,
            network_try_limit=None, task_try_limit=None,
            request_pause=NULL,
            priority_mode='random',
            meta=None,
            only_cache=False,
            config=None,
            args=None,
            parser_requests_per_process=10000,
            parser_pool_size=1,
            http_api_port=None,
            network_service='threaded',
            grab_transport='pycurl',
            # Deprecated
            transport=None):
        """
        Arguments:
        * thread-number - Number of concurrent network streams
        * network_try_limit - How many times try to send request
            again if network error was occurred, use 0 to disable
        * task_try_limit - Limit of tries to execute some task
            this is not the same as network_try_limit
            network try limit limits the number of tries which
            are performed automatically in case of network timeout
            of some other physical error
            but task_try_limit limits the number of attempts which
            are scheduled manually in the spider business logic
        * priority_mode - could be "random" or "const"
        * meta - arbitrary user data
        * retry_rebuild_user_agent - generate new random user-agent for each
            network request which is performed again due to network error
        * args - command line arguments parsed with `setup_arg_parser` method
        """

        self.fatal_error_queue = Queue()
        self.task_queue_parameters = None
        self.http_api_port = http_api_port
        self._started = None
        assert grab_transport in ('pycurl', 'urllib3')
        self.grab_transport_name = grab_transport
        self.parser_requests_per_process = parser_requests_per_process
        self.stat = Stat()
        self.task_queue = None
        if args is None:
            self.args = {}
        else:
            self.args = args
        if config is not None:
            self.config = config
        else:
            self.config = {}
        if meta:
            self.meta = meta
        else:
            self.meta = {}
        self.thread_number = (
            thread_number or
            int(self.config.get('thread_number',
                                DEFAULT_NETWORK_STREAM_NUMBER)))
        self.task_try_limit = (
            task_try_limit or
            int(self.config.get('task_try_limit', DEFAULT_TASK_TRY_LIMIT)))
        self.network_try_limit = (
            network_try_limit or
            int(self.config.get('network_try_limit',
                                DEFAULT_NETWORK_TRY_LIMIT)))
        self._grab_config = {}
        if priority_mode not in ['random', 'const']:
            raise SpiderMisuseError('Value of priority_mode option should be '
                                    '"random" or "const"')
        else:
            self.priority_mode = priority_mode
        self.only_cache = only_cache
        self.work_allowed = True
        if request_pause is not NULL:
            warn('Option `request_pause` is deprecated and is not '
                 'supported anymore')
        self.proxylist_enabled = None
        self.proxylist = None
        self.proxy = None
        self.proxy_auto_change = False
        self.interrupted = False
        self.cache_reader_service = None
        self.cache_writer_service = None
        self.parser_pool_size = parser_pool_size
        self.parser_service = ParserService(
            spider=self,
            pool_size=self.parser_pool_size,
        )
        if transport is not None:
            warn('The "transport" argument of Spider constructor is'
                 ' deprecated. Use "network_service" argument.')
            network_service = transport
        assert network_service in ('threaded',)
        if network_service == 'threaded':
            # pylint: disable=no-name-in-module, import-error
            from grab.spider.network_service.threaded import (
                NetworkServiceThreaded
            )
            self.network_service = NetworkServiceThreaded(
                self, self.thread_number
            )
        self.task_dispatcher = TaskDispatcherService(self)
        if self.http_api_port:
            self.http_api_service = HttpApiService(self)
        else:
            self.http_api_service = None
        self.task_generator_service = TaskGeneratorService(
            self.task_generator(), self,
        )