Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
Sometimes sites make crawling hard. Selenium-crawler uses selenium automation to fix that.
Python
branch: master

Fixed #1.

latest commit e49df4c1f4
Cory Walker authored
Failed to load latest commit information.
docs First commit.
seleniumcrawler Fixed #1.
.gitignore Organized tests.
CHANGES.txt Distribution changes.
LICENSE.txt Wrong license filename.
MANIFEST.in Distribution changes.
README.md Organized tests.
requirements.txt Requirements.txt
setup.py Organized tests.

README.md

selenium-crawler 0.1.0

Have you ever needed to crawl a list of urls that may or may not directly contain the content you so desperately crave? The web is full of links that do not behave for a number of reasons, and here is a list of just some of them:

  1. The page is actually a landing page that links to the page you want:
    • Hacker News
    • Hack-a-day
    • Reddit
  2. The page content is only made available after closing an ad:
    • Forbes
  3. The content is behind some sort of login or paywall:
    • Boston Globe
  4. One must click through some sort of pagination to find the content:
    • Web forums

You might be asking, why use Selenium when you can use a combination of PhantomJS and BeautifulSoup to extract the needed data? This is a great way to accomplish some of the listed tasks above, but it has a number of limitations:

  • Business teams would sometimes rather work with a visual tool rather than writing lines of code.
  • Selenium has most of the code that would be needed already built in.

Depending on Selenium DOES NOT mean that your crawling servers will need to also run a GUI. Selenium can run in a headless environment. See below for more information.

Quickstart

pip install -e git+https://github.com/cmwslw/selenium-crawler.git#egg=selenium-crawler
from seleniumcrawler import handle_url
print handle_url('https://news.ycombinator.com/item?id=5626377')

This will open up a browser window, 'click' on the main link, and load the article. It will print the following:

{
    'url': 'http://googleblog.blogspot.com/2013/04/google-now-on-your-iphone-and-ipad-with.html',
    'source': {{HTMLSOURCE}},
    'handler': 'hnews'
}

Where {{HTMLSOURCE}} is the actual HTML of the article.

Creating test cases

Create test cases as you normally would in Selenium, but be sure to take the following things into account:

Never make a test case article-specific

By default, when one records Selenium test cases, the very first instruction will be for the browser to load up a specific URL. This url will be the current article that you are viewing. Since you will be designing scripts that should work for ANY given article on the site. Make sure you always remove this line. I might make selenium-crawler take care of this step in the future. That is TBD.

Make your locators as robust as possible

On the one hand, make the locators specific enough to be confident that the will only match exactly what needs to be matched. On the other hand, make sure that your locator will continue to match even in the case of a website theme change or even redesign. It is impossible to account for every possible change in site structure, but you get the idea. By default, Selenium tends to create very specific locators while recording. Make sure to fix these up a bit.

Save test cases in a standard way

While it really doesn't matter where saved test case files even go, for reference and debugging purposes it is useful to save them in a standard location. For the time being, I have established the form sites/{name}/{name}.tc. Hopefully this will prove to be a decent convention.

Exporting test cases

Selenium-crawler takes in exported selenium Python scripts as inputs. Follow these instructions to obtain the exported scripts. Be sure to use the WebDriver backend.

ScreenShot

Next we need to save the script in a place where selenium-crawler can find it. Find the sites directory. Next, create a directory for the site. Choose a useful, concise name. Save the actual exported script in the format of {name}_raw.py:

ScreenShot

Parsing selenium cases

The test cases that Selenium exports are not even valid Python code. Here is an example:

self.accept_next_alert = true

Once you fix the syntax errors, they are useful for writing test suites, but not for writing crawlers. Running python makeparsed.py takes all these Selenium test cases and converts them to Python code usable for crawling. Here is an example that clicks through an annoying Javascript ad on Forbes pages:

from selenium import webdriver

def handle_link(link):
    driver = webdriver.Firefox()
    driver.implicitly_wait(30)
    driver.get(link)

    driver.find_element_by_xpath("//div[@class='header']/div[@class='continue']/a").click()

    results = {
        'url': driver.current_url,
        'source': driver.page_source
    }
    driver.quit()

    return results

So go ahead. Run python makeparsed.py. You should see output similar to the following:

Parsed ./sites/forbes/forbes_raw.py.
Parsed ./sites/hnews/hnews_raw.py.
Parsed ./sites/reddit/reddit_raw.py.

Don't worry if the paths are different for your installation. Keep in mind that makeparsed.py only has to be run when site scripts have either been changed or added.

Headless configuration

Running headless means that no actual GUI will be running on a monitor during use. Put simply, it means that no browser window will pop up when handling a URL. One way to run headless is through the use of xvfb, a tool used to set up virtual framebuffers. Run this before using selenium-crawler:

sh -e /etc/init.d/xvfb start
export DISPLAY=:99.0

This is the method that CI systems like Travis-CI and CircleCI recommend. There are other methods of running Selenium in a headless environment. Do a quick Google search for more information.

Contributing

Contributing is easy. If you write any new site handling scripts, just be sure to follow the guide above and write a quick test for it in test_all.py. Just send in a pull request and I'll get a CONTRIBUTORS.txt file going.

Something went wrong with that request. Please try again.