Web Crawler

Python app to crawl websites.

Design decisions

I decided to implement a producer/consumer model:

• The producer has a task queue where only new urls are queued.
• Consumers fetch a task consisting of an url, and replies with a list of its inner urls.
• The producer is constantly waiting on new replies, it does not wait for all consumers to be done.
• When it receives a reply it filters known urls and adds the new ones to the task queue.
• The producer signals for the termination of all consumers when there are no more urls to process.

This system has a mix of cpu and i/o, for parsing pages and fetching urls. While i/o bound systems tend to do ok with python "threads" (GIL), I decided to use processes for true parallelism.

For further optimizations one could profile the observed workload and task execution time and potentially batch tasks to reduce inter-process communication and the queue lock acquisition time. Note that for a highly scalable system you would likely need to reduce synchronization, e.g., by implement bloom filters to detect already visited pages.

Compatibility and dependencies

This project was tested with Python 3.5.

To install the dependencies:

pip install tldextract # proper url suffix handling
pip install beautifulsoup4 # xml/html parsing
pip install robotexclusionrulesparser # Proper robot.txt support

How to run

The standard way to execute is to call:

python crawler.py -l <url>

By default the crawler will launch twice as many processes as your CPU count. If you want a different CPU count you can change it using the -w option. For limiting the crawl domain to the domain of the provided URL use the -l option.

For further commands check the --help option:

$ python3 crawler.py -h
usage: crawler.py [-h] [-w W] [-l] [-v] [-r] [-sd] [-s] url

Crawls webpages for URLs

positional arguments:
  url         URL to crawl

optional arguments:
  -h, --help  show this help message and exit.
  -w W        Number of processes (default: 2 * cpu_count()).
  -l          If set crawls only domain specific URLs.
  -v          Enable verbose.
  -r          Enable robots.txt url blocking and throttling. Superseedes -w.
              and forces workers to 1.
  -sd         Single depth url crawl.
  -s          Silent. Superseedes -v and disables logging.

Respecting robots.txt

To execute with robots.txt support use the -r parameter. Note that for now this forces single process processing to account for robots.txt based throttling.

E.g., how to run with robots.txt support:

python crawler.py <url> -r

Additionally, to allow only one depth of url crawling and print only the robot.txt allowed new urls:

python crawler.py <url> -r -sd -s

Testing

This project includes a set of tests to validate potential URLs, check test_crawler.py to get a feeling for the supported URL formats.

To run the tests:

python test_crawler.py

Happy crawling