crawlera-session 1.2.8

Class that provides decorators and functions for easy handling of crawlera sessions in a scrapy spider.

README

Class that provides decorators and functions for easy handling of crawlera sessions in a scrapy spider.

Installation

pip install crawlera-session

Usage

Ensure COOKIES_ENABLED is set to True (otherwise there is no point in using this class) Subclass your spider from CrawleraSessionMixinSpider.

Provided decorator must be applied for every callback that yields requests that must conserve session. For starting requests, use init_start_request. This decorator will instruct requests provided by start requests, to initiate each one a new session.

The callback that yields requests that must follow the session initiated previously within same requests chain must be decorated with follow_session.

Example:

from crawlera_session import CrawleraSessionMixinSpider, RequestSession

crawlera_session = RequestSession()


class MySpider(CrawleraSessionMixinSpider, Spider):

    @crawlera_session.init_start_requests
    def start_requests(self):
        ...
        yield Request(...)


    @crawlera_session.follow_session
    def parse(self, response):
        ...
        yield Request(...)

Some times you need to initialize a session for a single request generated in a spider method. In that case, you can use init_request() method:

    def parse(self, response):
        ...
        # this request will not initiate session
        yield Request(...)
        ...
        # this request will initiate session
        yield crawlera_session.init_request(Request(...))

Alternatively, and probably more elegant with better code separation, you could do:

    def parse(self, response):
        ...
        # this request will not initiate session
        yield Request(...)

        # this request will initiate session
        yield from self.generate_init_requests(response)

    @crawlera_session.init_requests
    def generate_init_requests(self, response):
        ...
        yield Request(...)

The decorator init_requests (don't confuse with init_request() method just described) is similar to init_start_request but decorates a callback instead of start_requests().

If on the contrary, you want to send a normal (not session) request from a callback that was decorated with follow_session, you can use the no_crawlera_session meta tag:

    @crawlera_session.follow_session
    def parse(self, response):
        ...
        # this request will follow same session coming from response
        yield Request(...)
        ...
        # this request will not follow session
        yield Request(..., meta={'no_crawlera_session': True})

or, alternatively:

    def parse(self, response):
        # this request will follow same session coming from response
        yield from generate_session_reqs(response)
        ...
        # this request will not follow session
        yield Request(...)

    @crawlera_session.follow_session
    def generate_session_reqs(self, response)
        ...
        yield Request(...)

In short, the logic init_request->follow_session makes a chain of requests to use the same session. So requests issued by callbacks decorated by follow_session reuse the session created by the request which initiated it, in the same request chain as defined by the spider logic.

However, there are use cases where you want to reuse a session initiated in another chain, instead of initiating a new one. For that purpose, you can defer the session assignation of the requests until a previously created session is available for reusage (when the chain that created it is completed). There are two other decorators that implements this logic: defer_assign_session and unlock_session. Their logic must be used in combination of spider attribute MAX_PARALLEL_SESSIONS.

defer_assign_session makes requests yielded by the decorated callback:

• either to initiate a new request if number of initiated sessions is below MAX_PARALLEL_SESSIONS or MAX_PARALLEL_SESSIONS is None.
• or wait for an available session for reusage in other case.

In order to indicate the end of a request chain for unlocking its session for reusage, the last callback of the chain must be decorated with unlock_session.

Example:

from crawlera_session import CrawleraSessionMixinSpider, RequestSession

crawlera_session = RequestSession()


class MySpider(CrawleraSessionMixinSpider, Spider):

    MAX_PARALLEL_SESSIONS = 4

    def start_requests(self):
        ...
        yield Request(...)

    @crawlera_session.defer_assign_session
    def parse(self, response):
        ...
        yield Request(..., callback=callback2)

    @crawlera_session.follow_session
    def callback2(self, response):
        yield Request(..., callback=callback3)

    ...

    @crawlera_session.unlock_session
    def callbackN(self, response):
        yield Item(...)

It is important to consider that isage of unlock_session is not mandatory, as it is not always possible to determine whether a session can be unlocked. For example, consider the case where you need to extract multiple items from a search, and you can't unlock the session until the last request is handled. In this case you don't know which will be the last request, so you don't have a callback where to apply unlock_session decorator. In this situation, locked sessions will be automatically unlocked only when no more requests are pending in the scheduler. But notice that this is a suboptimal approach, as many locked sessions should be really available, but not used because we cannot know when they are really available until scheduler is idle. (There is a circumvented way to know in advance if there is some request in process belonging to a given session, by scanning all scrapy scheduler and downloader queues, a TODO is to check viability of this approach, For now it has not been implemented for simplicity.)

For better performance, normally it is better to set the number of concurrent requests to the same or more than as MAX_PARALLEL_SESSIONS. If you are combining crawlera session requests with no crawlera session requests, concurrent requests need to be bigger than MAX_PARALLEL_SESSIONS. Otherwise they can be equal. If smaller, session handling will be suboptimal in terms of speed. If too much bigger, many requests may be retained on memory and its use may increase significantly, so handling may be suboptimal in terms of memory usage.

Notice that if you don't set MAX_PARALLEL_SESSIONS, the behavior of the callback decorated by defer_assign_session will be that all requests yielded by it will initiate a new session. So the lock/unlock logic doesn't have much sense. In this case, you can just use init_requests decorator:

from crawlera_session import CrawleraSessionMixinSpider, RequestSession

crawlera_session = RequestSession()


class MySpider(CrawleraSessionMixinSpider, Spider):

    MAX_PARALLEL_SESSIONS = 4

    def start_requests(self):
        ...
        yield Request(...)

    @crawlera_session.init_requests
    def parse(self, response):
        ...
        yield Request(..., callback=callback2)

    @crawlera_session.follow_session
    def callback2(self, response):
        yield Request(..., callback=callback3)

    ...

    @crawlera_session.discard_session
    def callbackN(self, response):
        yield Item(...)

Notice that in the last callback we replaced the decorator unlock_session by discard_session. This decorator is optional, but in case your spider generates large amounts of requests, the memory usage can increase significantly if you don't drop unused sessions. Regardless you need to use it or not for saving memory, it is still a good practice when you can apply it (in situations where there are multiple final chain requests, it is not possible to drop a session when reaching any one of them).

By default, retrying a request uses the same session. However, frequently you need to reinitialize a session instead. The RequestSession() object provides a decorator for indicating special retry treatment on specific requests. For example, if the requests yielded by callback3() in the example above need to reinitialize session starting from the first request in the chain, you would write something like this:

class MySpider(CrawleraSessionMixinSpider, Spider):

    (...)

    @crawlera_session.init_requests
    def parse(self, response):
        ...
        yield Request(..., callback=callback2)


    (...)

    @crawlera_session.new_session_on_retry
    @crawlera_session.follow_session
    def callback3(self, response):
        yield Request(..., callback=self.callback4)

    (...)

    def callback4_errback(self, failure) -> Request:
        return Request(..., callback=callback2)

Notice the introduction of a new decorator, new_session_on_retry, and the definition (in this case) of callback4_errback() method:

• Every response yielded by a callback decorated by new_session_on_retry, will be assigned an errback with the same name as its callback, but with the _errback suffix appended.
• If this method is not implemented in the spider, an assertion error will be raised.
• Request returned by callback_errback() will be instructed internaly to initialize a new session, and the same errback will be assigned to it for further retries. So you don't need explicit action for it.

By default, the number of session retries for a given request will be 3. This default value can be overriden in the RequestSession() object initialization, i.e.:

crawlera_session = RequestSession(new_session_retries=5)

Finally, if you want to adjust priority on each successive request of a chain of requests, the RequestSession() instantiator admits the parameter priority_adjust. For example, if you want to ensure that requests more advanced in the chain are sent before new initial chain requests when you have multiple of them, you would rather use:

crawlera_session = RequestSession(priority_adjust=1)