ant-nest 0.20

A simple and clear Web Crawler framework build on python3.6+ with async

Overview

AntNest is a simple, clear and fast Web Crawler framework build on python3.6+, powered by asyncio.

As a Scrapy user, I think scrapy provide many awesome features what I think AntNest should have too.This is some main difference:

• Scrapy use callback way to write code while AntNest use coroutines

• Scrapy is stable and widely usage while AntNest is in early development

• AntNest has only 600+ lines core code now(thanks powerful lib like aiohttp, lxml and other else), and it works

Features

• Things(request, response and item) can though pipelines(in async or not)

• Item and item extractor, it`s easy to define and extract(by xpath, jpath or regex) a validated(by field type) item

• Custom “ensure_future” and “as_completed” method provide concurrent limit and collection of completed coroutines

• Default coroutines concurrent limit, reduce memory usage

Install

pip install ant_nest

Usage

Let`s take a look, create book.py first:

from ant_nest.ant import Ant
from ant_nest.things import StringField, IntField ItemExtractor
from ant_nest.pipelines import *

# define a item structure we want to crawl
class BookItem(Item):
    name = StringField()
    author = StringField()
    content = StringField()
    origin_url = StringField()
    date = IntField(null=True)  # The filed is optional


# define our ant
class BookAnt(Ant):
    # the things(request, response, item) will pass through pipelines in order, pipelines can change or drop them
    item_pipelines = [ItemValidatePipeline(),
                      ItemMysqlInsertPipeline(settings.MYSQL_HOST, settings.MYSQL_PORT, settings.MYSQL_USER,
                                              settings.MYSQL_PASSWORD, settings.MYSQL_DATABASE, 'book'),
                      ReportPipeline()]
    request_pipelines = [RequestDuplicateFilterPipeline(), RequestUserAgentPipeline(), ReportPipeline()]
    response_pipelines = [ResponseRetryPipeline(), ResponseFilterErrorPipeline(), ReportPipeline()]


    # define ItemExtractor to extract item field by xpath from response(html source code)
    self.item_extractor = ItemExtractor(BookItem)
    self.item_extractor.add_regex('name', 'name=(\w+);')
    self.item_extractor.add_xpath('author', '/html/body/div[1]/div[@class="author"]/text()')
    self.item_extractor.add_xpath('content', '/html/body/div[2]/div[2]/div[2]//text()',
                                  ItemExtractor.join_all)

    # crawl book information
    async def crawl_book(self, url):
        # send request and wait for response
        response = await self.request(url)
        # extract item from response
        item = self.item_extractor.extract(response)
        item.origin_url = str(response.url)  # or item['origin_url'] = str(response.url)
        # wait "collect" coroutine, it will let item pass through "item_pipelines"
        await self.collect(item)

    # app entrance
    async def run(self):
        response = await self.request('https://fake_bookstore.com')
        # extract all book links by xpath ("html_element" is a HtmlElement object from lxml lib)
        urls = response.html_element.xpath('//a[@class="single_book"]/@href')
        # run "crawl_book" coroutines in concurrent
        for url in urls:
            # "self.ensure_future" is a method like "ensure_future" in "asyncio", but it provide something else
            self.ensure_future(self.crawl_book(url))

Create a settings.py:

import logging


logging.basicConfig(level=logging.DEBUG)
ANT_PACKAGES = ['book']

Then in a console:

$ant_nest -a book.BookAnt

Defect

• Complex exception handle

one coroutine`s exception will break await chain especially in a loop unless we handle it by hand. eg:

for cor in self.as_completed((self.crawl(url) for url in self.urls)):
    try:
        await cor
    except Exception:  # may raise many exception in a await chain
        pass

• High memory usage

It`s a “feature” that asyncio eat large memory especially with high concurrent IO, one simple solution is set a concurrent limit, but it`s complex to get the balance between performance and limit.

Todo

• Memory leaks?

• Log system