Ejemplo n.º 1
0
    def setup_cache(self,
                    backend='mongo',
                    database=None,
                    use_compression=True,
                    **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')
        self.cache_enabled = True
        mod = __import__('grab.spider.cache_backend.%s' % backend, globals(),
                         locals(), ['foo'])
        cache = mod.CacheBackend(database=database,
                                 use_compression=use_compression,
                                 spider=self,
                                 **kwargs)
        self.cache_pipeline = CachePipeline(self, cache)
Ejemplo n.º 2
0
 def setup_cache(self, backend='mongo', database=None, use_compression=True,
                 **kwargs):
     if database is None:
         raise SpiderMisuseError('setup_cache method requires database '
                                 'option')
     self.cache_enabled = True
     mod = __import__('grab.spider.cache_backend.%s' % backend,
                      globals(), locals(), ['foo'])
     cache = mod.CacheBackend(database=database,
                              use_compression=use_compression,
                              spider=self, **kwargs)
     self.cache_pipeline = CachePipeline(self, cache)
Ejemplo n.º 3
0
 def setup_cache(self, backend='mongo', database=None, use_compression=True,
                 **kwargs):
     if database is None:
         raise SpiderMisuseError('setup_cache method requires database '
                                 'option')
     self.cache_enabled = True
     mod = __import__('grab.spider.cache_backend.%s' % backend,
                      globals(), locals(), ['foo'])
     cache = mod.CacheBackend(database=database,
                              use_compression=use_compression,
                              spider=self, **kwargs)
     self.cache_pipeline = CachePipeline(self, cache)
Ejemplo n.º 4
0
class Spider(object):
    """
    Asynchronous scraping framework.
    """

    # 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 = None

    class Meta:
        # 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 hasattr(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,
                 slave=None,
                 args=None,
                 # New options start here
                 taskq=None,
                 # MP:
                 network_result_queue=None,
                 parser_result_queue=None,
                 is_parser_idle=None,
                 shutdown_event=None,
                 mp_mode=False,
                 parser_pool_size=None,
                 parser_mode=False,
                 parser_requests_per_process=10000,
                 # http api
                 http_api_port=None,
                 transport='multicurl',
                 grab_transport='pycurl',
                 ):
        """
        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
        * network_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
        New options:
        * taskq=None,
        * newtork_response_queue=None,
        """

        if slave is not None:
            raise SpiderConfigurtionError(
                'Slave mode is not supported anymore. '
                'Use `mp_mode=True` option to run multiple HTML'
                ' parser processes.')

        # API:
        self.http_api_port = http_api_port

        assert transport in ('multicurl', 'threaded')
        self.transport_name = transport

        assert grab_transport in ('pycurl', 'urllib3')
        self.grab_transport_name = grab_transport

        # MP:
        self.mp_mode = mp_mode
        if self.mp_mode:
            from multiprocessing import Process, Event, Queue
        else:
            from multiprocessing.dummy import Process, Event, Queue

        if network_result_queue is not None:
            self.network_result_queue = network_result_queue
        else:
            self.network_result_queue = Queue()
        self.parser_result_queue = parser_result_queue
        self.is_parser_idle = is_parser_idle
        if shutdown_event is not None:
            self.shutdown_event = shutdown_event
        else:
            self.shutdown_event = Event()
        if not self.mp_mode and parser_pool_size and parser_pool_size > 1:
            raise SpiderConfigurationError(
                'Parser pool size could be only 1 in '
                'non-multiprocess mode')
        self.parser_pool_size = parser_pool_size
        self.parser_mode = parser_mode
        self.parser_requests_per_process = parser_requests_per_process

        self.stat = Stat()
        self.timer = Timer()
        self.task_queue = taskq

        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.cache_pipeline = None
        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

    def setup_cache(self, backend='mongo', database=None, use_compression=True,
                    **kwargs):
        if database is None:
            raise SpiderMisuseError('setup_cache method requires database '
                                    'option')
        self.cache_enabled = True
        mod = __import__('grab.spider.cache_backend.%s' % backend,
                         globals(), locals(), ['foo'])
        cache = mod.CacheBackend(database=database,
                                 use_compression=use_compression,
                                 spider=self, **kwargs)
        self.cache_pipeline = CachePipeline(self, cache)

    def setup_queue(self, backend='memory', **kwargs):
        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, raise_error=False):
        """
        Add task to the task queue.
        """

        # MP:
        # ***
        if self.parser_mode:
            self.parser_result_queue.put((task, None))
            return

        if self.task_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 = ('It is not allowed to build Task object with '
                   'relative URL: %s' % task.url)
            ex = SpiderError(msg)
            if raise_error:
                raise ex
            else:
                logger.error('', exc_info=ex)
                return False

        # TODO: keep original task priority if it was set explicitly
        # WTF the previous comment means?
        self.task_queue.put(task, 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.
        """

        logger_verbose.debug('Method `stop` was called')
        self.work_allowed = False

    def load_proxylist(self, source, source_type=None, proxy_type='http',
                       auto_init=True, auto_change=True,
                       **kwargs):
        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=True):
        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)
        elapsed = self.timer.timers['total']
        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'))

        if timing:
            out.append('')
            out.append(self.render_timing())
        return '\n'.join(out) + '\n'

    def render_timing(self):
        out = ['Timers:']
        out.append('  DOM: %.3f' % GLOBAL_STATE['dom_build_time'])
        time_items = [(x, y) for x, y in self.timer.timers.items()]
        time_items = sorted(time_items, key=lambda x: x[1])
        for time_item in time_items:
            out.append('  %s: %.03f' % time_item)
        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 prepare_parser(self):
        """
        You can do additional spider customization here
        before it has started working. Simply redefine
        this method in your Spider class.

        This method is called only from Spider working in parser mode
        that, in turn, is spawned automatically by main spider proces
        working in multiprocess mode.
        """

    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:
            # 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 task_generator_thread_wrapper(self, task_generator):
        """
        Load new tasks from `self.task_generator_object`
        Create new tasks.

        If task queue size is less than some value
        then load new tasks from tasks file.
        """

        while True:
            with self.timer.log_time('task_generator'):
                queue_size = self.task_queue.size()
                min_limit = self.thread_number * 10
            if queue_size < min_limit:
                with self.timer.log_time('task_generator'):
                    logger_verbose.debug(
                        'Task queue contains less tasks (%d) than '
                        'allowed limit (%d). Trying to add '
                        'new tasks.' % (queue_size, min_limit))
                    try:
                        for x in six.moves.range(min_limit - queue_size):
                            item = next(task_generator)
                            logger_verbose.debug('Got new item from generator. '
                                                 'Processing it.')
                            self.process_handler_result(item, None)
                    except StopIteration:
                        # If generator have no values to yield
                        # then disable it
                        logger_verbose.debug('Task generator has no more tasks. '
                                             'Disabling it')
                        break
            else:
                time.sleep(0.1)

    def start_task_generators(self):
        """
        Process `self.initial_urls` list and `self.task_generator`
        method.
        """

        logger_verbose.debug('Processing initial urls')
        if self.initial_urls:
            for url in self.initial_urls:
                self.add_task(Task('initial', url=url))

        self._task_generator_list = []
        th = Thread(target=self.task_generator_thread_wrapper,
                    args=[self.task_generator()])
        th.daemon = True
        th.start()
        self._task_generator_list.append(th)

    def get_task_from_queue(self):
        start = time.time()
        try:
            with self.timer.log_time('task_queue'):
                return self.task_queue.get()
        except queue.Empty:
            size = self.task_queue.size()
            if size:
                logger_verbose.debug(
                    'No ready-to-go tasks, Waiting for '
                    'scheduled tasks (%d)' % size)
                return True
            else:
                logger_verbose.debug('Task queue is empty.')
                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)
        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 faield and should be processed as error.
        """

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

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

        if hasattr(ex, 'tb'):
            logger.error('Error in %s function' % func_name)
            logger.error(ex.tb)
        else:
            logger.error('Error in %s function' % func_name, exc_info=ex)

        # 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 is not None else None
        self.stat.collect('fatal', '%s|%s|%s|%s' % (
            func_name, ex.__class__.__name__, ex_str, task_url))
        if isinstance(ex, FatalError):
            #raise FatalError()
            #six.reraise(FatalError, ex)
            #logger.error(ex.tb)
            raise ex

    def find_data_handler(self, data):
        try:
            return getattr(data, 'handler')
        except AttributeError:
            try:
                handler = getattr(self, 'data_%s' % data.handler_key)
            except AttributeError:
                raise NoDataHandler('No handler defined for Data %s'
                                    % data.handler_key)
            else:
                return handler

    def run_parser(self):
        """
        Main work cycle of spider process working in parser-mode.
        """
        self.is_parser_idle.clear()
        # Use Stat instance that does not print any logging messages
        if self.parser_mode:
            self.stat = Stat(logging_period=None)
        self.prepare_parser()
        process_request_count = 0
        try:
            recent_task_time = time.time()
            while True:
                try:
                    result = self.network_result_queue.get(block=False)
                except queue.Empty:
                    self.is_parser_idle.set()
                    time.sleep(0.1)
                    self.is_parser_idle.clear()
                    logger_verbose.debug('Network result queue is empty')
                    # Set `waiting_shutdown_event` only after 1 seconds
                    # of waiting for tasks to avoid
                    # race-condition issues
                    #if time.time() - recent_task_time > 1:
                    #    self.waiting_shutdown_event.set()
                    if self.shutdown_event.is_set():
                        logger_verbose.debug('Got shutdown event')
                        return
                else:
                    process_request_count += 1
                    recent_task_time = time.time()
                    if self.parser_mode:
                        self.stat.reset()
                    #if self.waiting_shutdown_event.is_set():
                    #    self.waiting_shutdown_event.clear()
                    try:
                        handler = self.find_task_handler(result['task'])
                    except NoTaskHandler as ex:
                        ex.tb = format_exc()
                        self.parser_result_queue.put((ex, result['task']))
                        self.stat.inc('parser:handler-not-found')
                    else:
                        self.process_network_result_with_handler(
                            result, handler)
                        self.stat.inc('parser:handler-processed')
                    finally:
                        if self.parser_mode:
                            data = {
                                'type': 'stat',
                                'counters': self.stat.counters,
                                'collections': self.stat.collections,
                            }
                            self.parser_result_queue.put((data,
                                                          result['task']))
                        if self.parser_mode:
                            if self.parser_requests_per_process:
                                if (process_request_count >=
                                        self.parser_requests_per_process):
                                    break
        except Exception as ex:
            logging.error('', exc_info=ex)
            raise
        #finally:
        #    self.waiting_shutdown_event.set()


    def process_network_result_with_handler(self, result, handler):
        handler_name = getattr(handler, '__name__', 'NONE')
        try:
            with self.timer.log_time('response_handler'):
                with self.timer.log_time('response_handler.%s' % handler_name):
                    handler_result = handler(result['grab'], result['task'])
                    if handler_result is None:
                        pass
                    else:
                        for something in handler_result:
                            self.parser_result_queue.put((something,
                                                          result['task']))
        except NoDataHandler as ex:
            ex.tb = format_exc()
            self.parser_result_queue.put((ex, result['task']))
        except Exception as ex:
            ex.tb = format_exc()
            self.parser_result_queue.put((ex, result['task']))

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

        # Update traffic statistics
        if res['grab'] and res['grab'].response:
            resp = res['grab'].response
            self.timer.inc_timer('network-name-lookup', resp.name_lookup_time)
            self.timer.inc_timer('network-connect', resp.connect_time)
            self.timer.inc_timer('network-total', resp.total_time)
            if from_cache:
                self.stat.inc('spider:download-size-with-cache',
                              resp.download_size)
                self.stat.inc('spider:upload-size-with-cache',
                              resp.upload_size)
            else:
                self.stat.inc('spider:download-size', resp.download_size)
                self.stat.inc('spider:upload-size', resp.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:
                # Need this to work around
                # pycurl feature/bug: 
                # pycurl instance uses previously connected proxy server
                # even if `proxy` options is set with another proxy server
                grab.setup(connection_reuse=False)
                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)

    def change_active_proxy(self, task, grab):
        self.proxy = self.proxylist.get_random_proxy()

    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)
            with self.timer.log_time('network_transport'):
                logger_verbose.debug('Submitting task to the transport '
                                     'layer')
                try:
                    self.transport.start_task_processing(
                        task, grab, grab_config_backup)
                except GrabInvalidUrl:
                    logger.debug('Task %s has invalid URL: %s' % (
                        task.name, task.url))
                    self.stat.collect('invalid-url', task.url)

    def start_api_thread(self):
        from grab.spider.http_api import HttpApiThread

        proc = HttpApiThread(self)
        proc.start()
        return proc

    def is_ready_to_shutdown(self):
        # Things should be true to shutdown spider
        # 1) No active task handlers (task_* functions)
        # 2) All task generators has completed work
        # 3) No active network threads
        # 4) Task queue is empty
        # 5) Network result queue is empty
        # 6) Cache is disabled or is in idle mode

        #print('parser result queue', self.parser_result_queue.qsize())
        #print('all parsers are idle', all(x['is_parser_idle'].is_set()
        #                                  for x in self.parser_pipeline.parser_pool))
        #print('alive task generators', any(x.isAlive() for x in self._task_generator_list))
        #print('active network threads', self.transport.get_active_threads_number())
        return (
            not self.parser_result_queue.qsize()
            and all(x['is_parser_idle'].is_set()
                    for x in self.parser_pipeline.parser_pool)
            and not any(x.isAlive() for x in self._task_generator_list) # (2)
            and not self.transport.get_active_threads_number() # (3)
            and not self.task_queue.size() # (4)
            and not self.network_result_queue.qsize() # (5)
            and (self.cache_pipeline is None
                 or (self.cache_pipeline.is_idle()
                     and self.cache_pipeline.input_queue.qsize() == 0
                     and self.cache_pipeline.result_queue.qsize() == 0))
        )

    def run(self):
        """
        Main method. All work is done here.
        """
        if self.mp_mode:
            from multiprocessing import Process, Event, Queue
        else:
            from multiprocessing.dummy import Process, Event, Queue

        self.timer.start('total')

        if self.transport_name == 'multicurl':
            self.transport = MulticurlTransport(self, self.thread_number)
        elif self.transport_name == 'threaded':
            self.transport = ThreadedTransport(self, self.thread_number)

        if self.http_api_port:
            http_api_proc = self.start_api_thread()
        else:
            http_api_proc = None

        self.parser_result_queue = Queue()
        self.parser_pipeline = ParserPipeline(
            bot=self,
            mp_mode=self.mp_mode,
            pool_size=self.parser_pool_size,
            shutdown_event=self.shutdown_event,
            network_result_queue=self.network_result_queue,
            parser_result_queue=self.parser_result_queue,
            requests_per_process=self.parser_requests_per_process,
        )
        network_result_queue_limit = max(10, self.thread_number * 2)
        
        try:
            # Run custom things defined by this specific spider
            # By defaut it does nothing
            self.prepare()

            # Setup task queue if it has not been configured yet
            if self.task_queue is None:
                self.setup_queue()

            # Initiate task generator. Only in main process!
            with self.timer.log_time('task_generator'):
                self.start_task_generators()

            # Work in infinite cycle untill
            # `self.work_allowed` flag is True
            #shutdown_countdown = 0 # !!!
            pending_tasks = deque()
            while self.work_allowed:
                free_threads = self.transport.get_free_threads_number()
                # Load new task only if:
                # 1) network transport has free threads
                # 2) network result queue is not full
                # 3) cache is disabled OR cache has free resources
                if (self.transport.get_free_threads_number()
                        and (self.network_result_queue.qsize()
                             < network_result_queue_limit)
                        and (self.cache_pipeline is None
                             or self.cache_pipeline.has_free_resources())):
                    if pending_tasks:
                        task = pending_tasks.popleft()
                    else:
                        task = self.get_task_from_queue()
                    if task is None:
                        # If received task is None then
                        # check if spider is ready to be shut down
                        if not pending_tasks and self.is_ready_to_shutdown():
                            # I am afraid there is a bug in `is_ready_to_shutdown`
                            # because it tries to evaluate too many things
                            # includig things that are being set from other threads,
                            # so to ensure we are really ready to shutdown I call
                            # is_ready_to_shutdown a few more times.
                            # Without this hack some times really rarely times
                            # the Grab fails to do its job
                            # A good way to see this bug is to disable this hack
                            # and run:
                            # while ./runtest.py -t test.spider_data; do echo "ok"; done;
                            # And wait a few minutes
                            really_ready = True
                            for x in range(10):
                                if not self.is_ready_to_shutdown():
                                    really_ready = False
                                    break
                                time.sleep(0.001)
                            if really_ready:
                                self.shutdown_event.set()
                                self.stop()
                                break # Break from `while self.work_allowed` cycle
                    elif isinstance(task, bool) and (task is True):
                        # If received task is True
                        # and there is no active network threads then
                        # take some sleep
                        if not self.transport.get_active_threads_number():
                            time.sleep(0.01)
                    else:
                        logger_verbose.debug('Got new task from task queue: %s'
                                             % task)
                        task.network_try_count += 1
                        is_valid, reason = self.check_task_limits(task)
                        if is_valid:
                            task_grab = self.setup_grab_for_task(task)
                            if self.cache_pipeline:
                                self.cache_pipeline.input_queue.put(
                                    ('load', (task, task_grab)),
                                )
                            else:
                                self.submit_task_to_transport(task, task_grab)
                        else:
                            self.log_rejected_task(task, reason)
                            handler = task.get_fallback_handler(self)
                            if handler:
                                handler(task)

                with self.timer.log_time('network_transport'):
                    logger_verbose.debug('Asking transport layer to do '
                                         'something')
                    self.transport.process_handlers()

                logger_verbose.debug('Processing network results (if any).')

                # Collect completed network results
                # Each result could be valid or failed
                # Result is dict {ok, grab, grab_config_backup, task, emsg}
                results = [(x, False) for x in
                           self.transport.iterate_results()]
                if self.cache_pipeline:
                    while True:
                        try:
                            action, result = self.cache_pipeline\
                                                 .result_queue.get(False)
                        except queue.Empty:
                            break
                        else:
                            assert action in ('network_result', 'task')
                            if action == 'network_result':
                                results.append((result, True))
                            elif action == 'task':
                                task = result
                                task_grab = self.setup_grab_for_task(task)
                                if (self.transport.get_free_threads_number()
                                        and (self.network_result_queue.qsize()
                                             < network_result_queue_limit)):
                                    self.submit_task_to_transport(task, task_grab)
                                else:
                                    pending_tasks.append(task)

                # Take sleep to avoid millions of iterations per second.
                # 1) If no results from network transport
                # 2) If task queue is empty (or if there are only delayed tasks)
                # 3) If no network activity
                # 4) If parser result queue is empty
                if (not results
                    and (task is None or bool(task) == True)
                    and not self.transport.get_active_threads_number()
                    and not self.parser_result_queue.qsize()
                    and (self.cache_pipeline is None
                         or (self.cache_pipeline.input_queue.qsize() == 0
                             and self.cache_pipeline.is_idle()
                             and self.cache_pipeline.result_queue.qsize() == 0))
                    ):
                        time.sleep(0.001)

                for result, from_cache in results:
                    if self.cache_pipeline and not from_cache:
                        if result['ok']:
                            self.cache_pipeline.input_queue.put(
                                ('save', (result['task'], result['grab']))
                            )
                    self.log_network_result_stats(
                        result, from_cache=from_cache)

                    is_valid = False
                    if result['task'].get('raw'):
                        is_valid = True
                    elif result['ok']:
                        res_code = result['grab'].response.code
                        if self.is_valid_network_response_code(res_code, result['task']):
                            is_valid = True

                    if is_valid:
                        self.network_result_queue.put(result)
                    else:
                        self.log_failed_network_result(result)
                        # Try to do network request one more time
                        # TODO:
                        # Implement valid_try_limit
                        # Use it if request failed not because of network error
                        # But because of content integrity check
                        if self.network_try_limit > 0:
                            result['task'].refresh_cache = True
                            result['task'].setup_grab_config(
                                result['grab_config_backup'])
                            self.add_task(result['task'])
                    if from_cache:
                        self.stat.inc('spider:task-%s-cache' % result['task'].name)
                    self.stat.inc('spider:request')

                while True:
                    try:
                        p_res, p_task = self.parser_result_queue.get(block=False)
                    except queue.Empty:
                        break
                    else:
                        self.stat.inc('spider:parser-result')
                        self.process_handler_result(p_res, p_task)

                if not self.shutdown_event.is_set():
                    self.parser_pipeline.check_pool_health()

            logger_verbose.debug('Work done')
        except KeyboardInterrupt:
            logger.info('\nGot ^C signal in process %d. Stopping.'
                        % os.getpid())
            self.interrupted = True
            raise
        finally:
            # This code is executed when main cycles is breaked
            self.timer.stop('total')
            self.stat.print_progress_line()
            self.shutdown()

            # Stop HTTP API process
            if http_api_proc:
                http_api_proc.server.shutdown()
                http_api_proc.join()

            if self.task_queue:
                self.task_queue.clear()

            # Stop parser processes
            self.shutdown_event.set()
            self.parser_pipeline.shutdown()
            logger.debug('Main process [pid=%s]: work done' % os.getpid())

    def log_failed_network_result(self, res):
        # Log the error
        if res['ok']:
            msg = 'http-%s' % res['grab'].response.code
        else:
            msg = res['error_abbr']

        self.stat.inc('error:%s' % msg) 
        #logger.error(u'Network error: %s' % msg)#%
                     #make_unicode(msg, errors='ignore'))

    def log_rejected_task(self, task, reason):
        logger_verbose.debug('Task %s is rejected due to '
                             '%s limit'
                             % (task.name, 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)

    def process_handler_result(self, result, task):
        """
        Process result received from the task handler.

        Result could be:
        * None
        * Task instance
        * Data instance.
        * dict:
          * {type: "stat", counters: [], collections: []} 
        * ResponseNotValid-based exception
        * Arbitrary exception
        """

        if isinstance(result, Task):
            self.add_task(result)
        elif isinstance(result, Data):
            handler = self.find_data_handler(result)
            try:
                data_result = handler(**result.storage)
                if data_result is None:
                    pass
                else:
                    for something in data_result:
                        self.process_handler_result(something, task)

            except Exception as ex:
                self.process_handler_error('data_%s' % result.handler_key, ex,
                                           task)
        elif result is None:
            pass
        elif isinstance(result, ResponseNotValid):
            self.add_task(task.clone(refresh_cache=True))
            error_code = result.__class__.__name__.replace('_', '-')
            self.stat.inc('integrity:%s' % error_code)
        elif isinstance(result, Exception): 
            handler = self.find_task_handler(task)
            handler_name = getattr(handler, '__name__', 'NONE')
            self.process_handler_error(handler_name, result, task)
        elif isinstance(result, dict):
            if result.get('type') == 'stat':
                for name, count in result['counters'].items():
                    self.stat.inc(name, count)
                for name, items in result['collections'].items():
                    for item in items:
                        self.stat.collect(name, item)
            else:
                raise SpiderError('Unknown result type: %s' % result)
        else:
            raise SpiderError('Unknown result type: %s' % result)
Ejemplo n.º 5
0
class Spider(DeprecatedThingsSpiderMixin):
    """
    Asynchronous scraping framework.
    """

    # 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 = None

    class Meta:
        # 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 hasattr(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,
                 slave=None,
                 args=None,
                 # New options start here
                 taskq=None,
                 # MP:
                 network_result_queue=None,
                 parser_result_queue=None,
                 is_parser_idle=None,
                 shutdown_event=None,
                 mp_mode=False,
                 parser_pool_size=None,
                 parser_mode=False,
                 parser_requests_per_process=10000,
                 # http api
                 http_api_port=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
        * network_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
        New options:
        * taskq=None,
        * newtork_response_queue=None,
        """

        if slave is not None:
            raise SpiderConfigurtionError(
                'Slave mode is not supported anymore. '
                'Use `mp_mode=True` option to run multiple HTML'
                ' parser processes.')

        # API:
        self.http_api_port = http_api_port

        # MP:
        self.mp_mode = mp_mode
        if self.mp_mode:
            from multiprocessing import Process, Event, Queue
        else:
            from multiprocessing.dummy import Process, Event, Queue

        if network_result_queue is not None:
            self.network_result_queue = network_result_queue
        else:
            self.network_result_queue = Queue()
        self.parser_result_queue = parser_result_queue
        self.is_parser_idle = is_parser_idle
        if shutdown_event is not None:
            self.shutdown_event = shutdown_event
        else:
            self.shutdown_event = Event()
        if not self.mp_mode and parser_pool_size and parser_pool_size > 1:
            raise SpiderConfigurationError(
                'Parser pool size could be only 1 in '
                'non-multiprocess mode')
        self.parser_pool_size = parser_pool_size
        self.parser_mode = parser_mode
        self.parser_requests_per_process = parser_requests_per_process

        self.stat = Stat()
        self.timer = Timer()
        self.task_queue = taskq

        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.cache_pipeline = None
        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

    def setup_cache(self, backend='mongo', database=None, use_compression=True,
                    **kwargs):
        if database is None:
            raise SpiderMisuseError('setup_cache method requires database '
                                    'option')
        self.cache_enabled = True
        mod = __import__('grab.spider.cache_backend.%s' % backend,
                         globals(), locals(), ['foo'])
        cache = mod.CacheBackend(database=database,
                                 use_compression=use_compression,
                                 spider=self, **kwargs)
        self.cache_pipeline = CachePipeline(self, cache)

    def setup_queue(self, backend='memory', **kwargs):
        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, raise_error=False):
        """
        Add task to the task queue.
        """

        # MP:
        # ***
        if self.parser_mode:
            self.parser_result_queue.put((task, None))
            return

        if self.task_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_is_custom:
            task.priority = self.generate_task_priority()
            task.priority_is_custom = False
        else:
            task.priority_is_custom = True

        try:
            if not task.url.startswith(('http://', 'https://', 'ftp://',
                                        'file://', 'feed://')):
                if self.base_url is None:
                    msg = 'Could not resolve relative URL because base_url ' \
                          'is not specified. Task: %s, URL: %s'\
                          % (task.name, task.url)
                    raise SpiderError(msg)
                else:
                    warn('Class attribute `Spider::base_url` is deprecated. '
                         'Use Task objects with absolute URLs')
                    task.url = urljoin(self.base_url, task.url)
                    # If task has grab_config object then update it too
                    if task.grab_config:
                        task.grab_config['url'] = task.url
        except Exception as ex:
            self.stat.collect('task-with-invalid-url', task.url)
            if raise_error:
                raise
            else:
                logger.error('', exc_info=ex)
                return False

        # TODO: keep original task priority if it was set explicitly
        self.task_queue.put(task, 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.
        """

        logger_verbose.debug('Method `stop` was called')
        self.work_allowed = False

    def load_proxylist(self, source, source_type=None, proxy_type='http',
                       auto_init=True, auto_change=True,
                       **kwargs):
        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=True):
        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)
        elapsed = self.timer.timers['total']
        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'))

        if timing:
            out.append('')
            out.append(self.render_timing())
        return '\n'.join(out) + '\n'

    def render_timing(self):
        out = ['Timers:']
        out.append('  DOM: %.3f' % GLOBAL_STATE['dom_build_time'])
        time_items = [(x, y) for x, y in self.timer.timers.items()]
        time_items = sorted(time_items, key=lambda x: x[1])
        for time_item in time_items:
            out.append('  %s: %.03f' % time_item)
        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 prepare_parser(self):
        """
        You can do additional spider customization here
        before it has started working. Simply redefine
        this method in your Spider class.

        This method is called only from Spider working in parser mode
        that, in turn, is spawned automatically by main spider proces
        working in multiprocess mode.
        """

    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
        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:
            # 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 task_generator_thread_wrapper(self, task_generator):
        """
        Load new tasks from `self.task_generator_object`
        Create new tasks.

        If task queue size is less than some value
        then load new tasks from tasks file.
        """

        while True:
            with self.timer.log_time('task_generator'):
                queue_size = self.task_queue.size()
                min_limit = self.thread_number * 10
            if queue_size < min_limit:
                with self.timer.log_time('task_generator'):
                    logger_verbose.debug(
                        'Task queue contains less tasks (%d) than '
                        'allowed limit (%d). Trying to add '
                        'new tasks.' % (queue_size, min_limit))
                    try:
                        for x in six.moves.range(min_limit - queue_size):
                            item = next(task_generator)
                            logger_verbose.debug('Got new item from generator. '
                                                 'Processing it.')
                            self.process_handler_result(item)
                    except StopIteration:
                        # If generator have no values to yield
                        # then disable it
                        logger_verbose.debug('Task generator has no more tasks. '
                                             'Disabling it')
                        break
            else:
                time.sleep(0.1)

    def start_task_generators(self):
        """
        Process `self.initial_urls` list and `self.task_generator`
        method.
        """

        logger_verbose.debug('Processing initial urls')
        if self.initial_urls:
            for url in self.initial_urls:
                self.add_task(Task('initial', url=url))

        self._task_generator_list = []
        th = Thread(target=self.task_generator_thread_wrapper,
                    args=[self.task_generator()])
        th.daemon = True
        th.start()
        self._task_generator_list.append(th)

    def get_task_from_queue(self):
        start = time.time()
        try:
            with self.timer.log_time('task_queue'):
                return self.task_queue.get()
        except queue.Empty:
            size = self.task_queue.size()
            if size:
                logger_verbose.debug(
                    'No ready-to-go tasks, Waiting for '
                    'scheduled tasks (%d)' % size)
                return True
            else:
                logger_verbose.debug('Task queue is empty.')
                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)
        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 faield and should be processed as error.
        """

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

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

        if hasattr(ex, 'tb'):
            logger.error('Error in %s function' % func_name)
            logger.error(ex.tb)
        else:
            logger.error('Error in %s function' % func_name, exc_info=ex)

        # 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 is not None else None
        self.stat.collect('fatal', '%s|%s|%s|%s' % (
            func_name, ex.__class__.__name__, ex_str, task_url))
        if isinstance(ex, FatalError):
            #raise FatalError()
            #six.reraise(FatalError, ex)
            #logger.error(ex.tb)
            raise ex

    def find_data_handler(self, data):
        try:
            return getattr(data, 'handler')
        except AttributeError:
            try:
                handler = getattr(self, 'data_%s' % data.handler_key)
            except AttributeError:
                raise NoDataHandler('No handler defined for Data %s'
                                    % data.handler_key)
            else:
                return handler

    def is_valid_network_result(self, res):
        if res['task'].get('raw'):
            return True
        if res['ok']:
            res_code = res['grab'].response.code
            if self.is_valid_network_response_code(res_code, res['task']):
                return True
        return False

    def run_parser(self):
        """
        Main work cycle of spider process working in parser-mode.
        """
        self.is_parser_idle.clear()
        # Use Stat instance that does not print any logging messages
        if self.parser_mode:
            self.stat = Stat(logging_period=None)
        self.prepare_parser()
        process_request_count = 0
        try:
            recent_task_time = time.time()
            while True:
                try:
                    result = self.network_result_queue.get(block=False)
                except queue.Empty:
                    self.is_parser_idle.set()
                    time.sleep(0.1)
                    self.is_parser_idle.clear()
                    logger_verbose.debug('Network result queue is empty')
                    # Set `waiting_shutdown_event` only after 1 seconds
                    # of waiting for tasks to avoid
                    # race-condition issues
                    #if time.time() - recent_task_time > 1:
                    #    self.waiting_shutdown_event.set()
                    if self.shutdown_event.is_set():
                        logger_verbose.debug('Got shutdown event')
                        return
                else:
                    process_request_count += 1
                    recent_task_time = time.time()
                    if self.parser_mode:
                        self.stat.reset()
                    #if self.waiting_shutdown_event.is_set():
                    #    self.waiting_shutdown_event.clear()
                    try:
                        handler = self.find_task_handler(result['task'])
                    except NoTaskHandler as ex:
                        ex.tb = format_exc()
                        self.parser_result_queue.put((ex, result['task']))
                        self.stat.inc('parser:handler-not-found')
                    else:
                        self.process_network_result_with_handler(
                            result, handler)
                        self.stat.inc('parser:handler-processed')
                    finally:
                        if self.parser_mode:
                            data = {
                                'type': 'stat',
                                'counters': self.stat.counters,
                                'collections': self.stat.collections,
                            }
                            self.parser_result_queue.put((data,
                                                          result['task']))
                        if self.parser_mode:
                            if self.parser_requests_per_process:
                                if (process_request_count >=
                                        self.parser_requests_per_process):
                                    break
        except Exception as ex:
            logging.error('', exc_info=ex)
            raise
        #finally:
        #    self.waiting_shutdown_event.set()


    def process_network_result_with_handler(self, result, handler):
        handler_name = getattr(handler, '__name__', 'NONE')
        try:
            with self.timer.log_time('response_handler'):
                with self.timer.log_time('response_handler.%s' % handler_name):
                    handler_result = handler(result['grab'], result['task'])
                    if handler_result is None:
                        pass
                    else:
                        for something in handler_result:
                            self.parser_result_queue.put((something,
                                                          result['task']))
        except NoDataHandler as ex:
            ex.tb = format_exc()
            self.parser_result_queue.put((ex, result['task']))
        except Exception as ex:
            ex.tb = format_exc()
            self.parser_result_queue.put((ex, result['task']))

    def find_task_handler(self, task):
        if task.origin_task_generator is not None:
            return self.handler_for_inline_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 handler_for_inline_task(self, grab, task):
        # It can be subroutine for the first call,
        # So we should check it
        if isinstance(task, types.GeneratorType):
            coroutines_stack = []
            sendval = None
            origin_task_generator = task
            target = origin_task_generator
        else:
            coroutines_stack = task.coroutines_stack
            sendval = grab
            origin_task_generator = task.origin_task_generator
            target = origin_task_generator

        while True:
            try:
                result = target.send(sendval)
                # If it is subroutine we have to initialize it and
                # save coroutine in the coroutines stack
                if isinstance(result, types.GeneratorType):
                    coroutines_stack.append(target)
                    sendval = None
                    target = result
                    origin_task_generator = target
                else:
                    new_task = result
                    new_task.origin_task_generator = origin_task_generator
                    new_task.coroutines_stack = coroutines_stack
                    self.add_task(new_task)
                    return
            except StopIteration:
                # If coroutine is over we should check coroutines stack,
                # may be it is subroutine
                if coroutines_stack:
                    target = coroutines_stack.pop()
                    origin_task_generator = target
                else:
                    return

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

        # Update traffic statistics
        if res['grab'] and res['grab'].response:
            resp = res['grab'].response
            self.timer.inc_timer('network-name-lookup', resp.name_lookup_time)
            self.timer.inc_timer('network-connect', resp.connect_time)
            self.timer.inc_timer('network-total', resp.total_time)
            if from_cache:
                self.stat.inc('spider:download-size-with-cache',
                              resp.download_size)
                self.stat.inc('spider:upload-size-with-cache',
                              resp.upload_size)
            else:
                self.stat.inc('spider:download-size', resp.download_size)
                self.stat.inc('spider:upload-size', resp.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:
                # Need this to work around
                # pycurl feature/bug: 
                # pycurl instance uses previously connected proxy server
                # even if `proxy` options is set with another proxy server
                grab.setup(connection_reuse=False)
                if self.proxy_auto_change:
                    self.proxy = self.change_proxy(task, grab)

    def change_proxy(self, task, grab):
        proxy = self.proxylist.get_random_proxy()
        grab.setup(proxy=proxy.get_address(),
                   proxy_userpwd=proxy.get_userpwd(),
                   proxy_type=proxy.proxy_type)
        return proxy

    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)
            with self.timer.log_time('network_transport'):
                logger_verbose.debug('Submitting task to the transport '
                                     'layer')
                try:
                    self.transport.start_task_processing(
                        task, grab, grab_config_backup)
                except GrabInvalidUrl:
                    logger.debug('Task %s has invalid URL: %s' % (
                        task.name, task.url))
                    self.stat.collect('invalid-url', task.url)

    def start_api_thread(self):
        from grab.spider.http_api import HttpApiThread

        proc = HttpApiThread(self)
        proc.start()
        return proc

    def is_ready_to_shutdown(self):
        # Things should be true to shutdown spider
        # 1) No active task handlers (task_* functions)
        # 2) All task generators has completed work
        # 3) No active network threads
        # 4) Task queue is empty
        # 5) Network result queue is empty
        # 6) Cache is disabled or is in idle mode
        return (
            not self.parser_result_queue.qsize()
            and all(x['is_parser_idle'].is_set()
                    for x in self.parser_pipeline.parser_pool)
            and not any(x.isAlive() for x in self._task_generator_list) # (2)
            and not self.transport.get_active_threads_number() # (3)
            and not self.task_queue.size() # (4)
            and not self.network_result_queue.qsize() # (5)
            and (self.cache_pipeline is None
                 or (self.cache_pipeline.is_idle()
                     and self.cache_pipeline.input_queue.qsize() == 0
                     and self.cache_pipeline.result_queue.qsize() == 0))
        )

    def run(self):
        """
        Main method. All work is done here.
        """
        if self.mp_mode:
            from multiprocessing import Process, Event, Queue
        else:
            from multiprocessing.dummy import Process, Event, Queue

        self.timer.start('total')
        self.transport = MulticurlTransport(self.thread_number)

        if self.http_api_port:
            http_api_proc = self.start_api_thread()
        else:
            http_api_proc = None

        self.parser_result_queue = Queue()
        self.parser_pipeline = ParserPipeline(
            bot=self,
            mp_mode=self.mp_mode,
            pool_size=self.parser_pool_size,
            shutdown_event=self.shutdown_event,
            network_result_queue=self.network_result_queue,
            parser_result_queue=self.parser_result_queue,
            requests_per_process=self.parser_requests_per_process,
        )
        network_result_queue_limit = max(10, self.thread_number * 2)
        
        try:
            # Run custom things defined by this specific spider
            # By defaut it does nothing
            self.prepare()

            # Setup task queue if it has not been configured yet
            if self.task_queue is None:
                self.setup_queue()

            # Initiate task generator. Only in main process!
            with self.timer.log_time('task_generator'):
                self.start_task_generators()

            # Work in infinite cycle untill
            # `self.work_allowed` flag is True
            #shutdown_countdown = 0 # !!!
            pending_tasks = deque()
            while self.work_allowed:
                free_threads = self.transport.get_free_threads_number()
                # Load new task only if:
                # 1) network transport has free threads
                # 2) network result queue is not full
                # 3) cache is disabled OR cache has free resources
                if (self.transport.get_free_threads_number()
                        and (self.network_result_queue.qsize()
                             < network_result_queue_limit)
                        and (self.cache_pipeline is None
                             or self.cache_pipeline.has_free_resources())):
                    if pending_tasks:
                        task = pending_tasks.popleft()
                    else:
                        task = self.get_task_from_queue()
                    if task is None:
                        # If received task is None then
                        # check if spider is ready to be shut down
                        if not pending_tasks and self.is_ready_to_shutdown():
                            #shutdown_countdown -= 1
                            #time.sleep(0.02)
                            #if shutdown_countdown <= 0:
                            self.shutdown_event.set()
                            #print('STOP!!!!!!!!!')
                            self.stop()
                            break # Break from `while self.work_allowed` cycle
                    elif isinstance(task, bool) and (task is True):
                        # If received task is True
                        # and there is no active network threads then
                        # take some sleep
                        if not self.transport.get_active_threads_number():
                            time.sleep(0.01)
                    else:
                        logger_verbose.debug('Got new task from task queue: %s'
                                             % task)
                        task.network_try_count += 1
                        is_valid, reason = self.check_task_limits(task)
                        if is_valid:
                            task_grab = self.setup_grab_for_task(task)
                            if self.cache_pipeline:
                                self.cache_pipeline.input_queue.put(
                                    ('load', (task, task_grab)),
                                )
                            else:
                                self.submit_task_to_transport(task, task_grab)
                        else:
                            self.log_rejected_task(task, reason)
                            handler = task.get_fallback_handler(self)
                            if handler:
                                handler(task)

                with self.timer.log_time('network_transport'):
                    logger_verbose.debug('Asking transport layer to do '
                                         'something')
                    self.transport.process_handlers()

                logger_verbose.debug('Processing network results (if any).')

                # Collect completed network results
                # Each result could be valid or failed
                # Result is dict {ok, grab, grab_config_backup, task, emsg}
                results = [(x, False) for x in
                           self.transport.iterate_results()]
                if self.cache_pipeline:
                    while True:
                        try:
                            action, result = self.cache_pipeline\
                                                 .result_queue.get(False)
                        except queue.Empty:
                            break
                        else:
                            assert action in ('network_result', 'task')
                            if action == 'network_result':
                                results.append((result, True))
                            elif action == 'task':
                                task = result
                                task_grab = self.setup_grab_for_task(task)
                                if (self.transport.get_free_threads_number()
                                        and (self.network_result_queue.qsize()
                                             < network_result_queue_limit)):
                                    self.submit_task_to_transport(task, task_grab)
                                else:
                                    pending_tasks.append(task)

                # Take sleep to avoid millions of iterations per second.
                # 1) If no results from network transport
                # 2) If task queue is empty (or if there are only delayed tasks)
                # 3) If no network activity
                # 4) If parser result queue is empty
                if (not results
                    and (task is None or bool(task) == True)
                    and not self.transport.get_active_threads_number()
                    and not self.parser_result_queue.qsize()
                    and (self.cache_pipeline is None
                         or (self.cache_pipeline.input_queue.qsize() == 0
                             and self.cache_pipeline.is_idle()
                             and self.cache_pipeline.result_queue.qsize() == 0))
                    ):
                        time.sleep(0.001)

                for result, from_cache in results:
                    if self.cache_pipeline and not from_cache:
                        if result['ok']:
                            self.cache_pipeline.input_queue.put(
                                ('save', (result['task'], result['grab']))
                            )
                    self.log_network_result_stats(
                        result, from_cache=from_cache)
                    if self.is_valid_network_result(result):
                        #print('!! PUT NETWORK RESULT INTO QUEUE (base.py)')
                        self.network_result_queue.put(result)
                    else:
                        self.log_failed_network_result(result)
                        # Try to do network request one more time
                        if self.network_try_limit > 0:
                            result['task'].refresh_cache = True
                            result['task'].setup_grab_config(
                                result['grab_config_backup'])
                            self.add_task(result['task'])
                    if from_cache:
                        self.stat.inc('spider:task-%s-cache' % result['task'].name)
                    self.stat.inc('spider:request')

                while True:
                    try:
                        p_res, p_task = self.parser_result_queue.get(block=False)
                    except queue.Empty:
                        break
                    else:
                        self.stat.inc('spider:parser-result')
                        self.process_handler_result(p_res, p_task)

                if not self.shutdown_event.is_set():
                    self.parser_pipeline.check_pool_health()

            logger_verbose.debug('Work done')
        except KeyboardInterrupt:
            logger.info('\nGot ^C signal in process %d. Stopping.'
                        % os.getpid())
            self.interrupted = True
            raise
        finally:
            # This code is executed when main cycles is breaked
            self.timer.stop('total')
            self.stat.print_progress_line()
            self.shutdown()

            # Stop HTTP API process
            if http_api_proc:
                http_api_proc.server.shutdown()
                http_api_proc.join()

            if self.task_queue:
                self.task_queue.clear()

            # Stop parser processes
            self.shutdown_event.set()
            self.parser_pipeline.shutdown()
            logger.debug('Main process [pid=%s]: work done' % os.getpid())

    def log_failed_network_result(self, res):
        # Log the error
        if res['ok']:
            msg = 'http-%s' % res['grab'].response.code
        else:
            msg = res['error_abbr']

        self.stat.inc('error:%s' % msg) 
        #logger.error(u'Network error: %s' % msg)#%
                     #make_unicode(msg, errors='ignore'))

    def log_rejected_task(self, task, reason):
        logger_verbose.debug('Task %s is rejected due to '
                             '%s limit'
                             % (task.name, 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)

    def process_handler_result(self, result, task=None):
        """
        Process result received from the task handler.

        Result could be:
        * None
        * Task instance
        * Data instance.
        """

        if isinstance(result, Task):
            self.add_task(result)
        elif isinstance(result, Data):
            handler = self.find_data_handler(result)
            try:
                data_result = handler(**result.storage)
                if data_result is None:
                    pass
                else:
                    for something in data_result:
                        self.process_handler_result(something, task)

            except Exception as ex:
                self.process_handler_error('data_%s' % result.handler_key, ex,
                                           task)
        elif result is None:
            pass
        elif isinstance(result, Exception): 
            handler = self.find_task_handler(task)
            handler_name = getattr(handler, '__name__', 'NONE')
            self.process_handler_error(handler_name, result, task)
        elif isinstance(result, dict):
            if result.get('type') == 'stat':
                for name, count in result['counters'].items():
                    self.stat.inc(name, count)
                for name, items in result['collections'].items():
                    for item in items:
                        self.stat.collect(name, item)
            else:
                raise SpiderError('Unknown result type: %s' % result)
        else:
            raise SpiderError('Unknown result type: %s' % result)