python - How do I set up Scrapy to deal with a captcha

Question

I'm trying to scrape a site that requires the user to enter the search value and a captcha. I've got an optical character recognition (OCR) routine for the captcha that succeeds about 33% of the time. Since the captchas are always alphabetic text, I want to reload the captcha if the OCR function returns non-alphabetic characters. Once I have a text "word", I want to submit the search form.

The results come back in the same page, with the form ready for a new search and a new captcha. So I need to rinse and repeat until I've exhausted my search terms.

Here's the top-level algorithm:

1. Load page initially
2. Download the captcha image, run it through the OCR
3. If the OCR doesn't come back with a text-only result, refresh the captcha and repeat this step
4. Submit the query form in the page with search term and captcha
5. Check the response to see whether the captcha was correct
6. If it was correct, scrape the data
7. Go to 2

I've tried using a pipeline for getting the captcha, but then I don't have the value for the form submission. If I just fetch the image without going through the framework, using urllib or something, then the cookie with the session is not submitted, so the captcha validation on the server fails.

What's the ideal Scrapy way of doing this?

Answer 1

It's a really deep topic with a bunch of solutions. But if you want to apply the logic you've defined in your post you can use scrapy Downloader Middlewares.

Something like:

class CaptchaMiddleware(object):
    max_retries = 5
    def process_response(request, response, spider):
        if not request.meta.get('solve_captcha', False):
            return response  # only solve requests that are marked with meta key
        catpcha = find_catpcha(response)
        if not captcha:  # it might not have captcha at all!
            return response
        solved = solve_captcha(captcha)
        if solved:
            response.meta['catpcha'] = captcha
            response.meta['solved_catpcha'] = solved
            return response
        else:
            # retry page for new captcha
            # prevent endless loop
            if request.meta.get('catpcha_retries', 0) == max_retries:
                logging.warning('max retries for captcha reached for {}'.format(request.url))
                raise IgnoreRequest 
            request.meta['dont_filter'] = True
            request.meta['captcha_retries'] = request.meta.get('captcha_retries', 0) + 1
            return request
    

This example will intercept every response and try to solve the captcha. If failed it will retry the page for new captcha, if successful it will add some meta keys to response with solved captcha values.
In your spider you would use it like this:

class MySpider(scrapy.Spider):
    def parse(self, response):
        url = ''# url that requires captcha
        yield Request(url, callback=self.parse_captchad, meta={'solve_captcha': True},
                      errback=self.parse_fail)
    
    def parse_captchad(self, response):
        solved = response['solved']
        # do stuff
    
    def parse_fail(self, response):
        # failed to retrieve captcha in 5 tries :(
        # do stuff