Scrapy & JavaScript integration through Splash

This library provides Scrapy and JavaScript integration using Splash. The license is BSD 3-clause.

Installation

Install scrapy-splash using pip:

$ pip install scrapy-splash

Scrapy-Splash uses Splash HTTP API, so you also need a Splash instance. Usually to install & run Splash, something like this is enough:

$ docker run -p 8050:8050 scrapinghub/splash

Check Splash install docs for more info.

Configuration

1. Add the Splash server address to settings.py of your Scrapy project like this:

   SPLASH_URL = 'http://192.168.59.103:8050'
   

2. Enable the Splash middleware by adding it to DOWNLOADER_MIDDLEWARES in your settings.py file and changing HttpCompressionMiddleware priority:

   DOWNLOADER_MIDDLEWARES = {
       'scrapy_splash.SplashCookiesMiddleware': 723,
       'scrapy_splash.SplashMiddleware': 725,
       'scrapy.downloadermiddlewares.httpcompression.HttpCompressionMiddleware': 810,
   }
   

   Order 723 is just before HttpProxyMiddleware (750) in default scrapy settings.

   HttpCompressionMiddleware priority should be changed in order to allow advanced response processing; see scrapy/scrapy#1895 for details.

3. Enable SplashDeduplicateArgsMiddleware by adding it to SPIDER_MIDDLEWARES in your settings.py:

   SPIDER_MIDDLEWARES = {
       'scrapy_splash.SplashDeduplicateArgsMiddleware': 100,
   }
   

   This middleware is needed to support cache_args feature; it allows to save disk space by not storing duplicate Splash arguments multiple times in a disk request queue. If Splash 2.1+ is used the middleware also allows to save network traffic by not sending these duplicate arguments to Splash server multiple times.

4. Set a custom DUPEFILTER_CLASS:

   DUPEFILTER_CLASS = 'scrapy_splash.SplashAwareDupeFilter'
   

5. If you use Scrapy HTTP cache then a custom cache storage backend is required. scrapy-splash provides a subclass of scrapy.contrib.httpcache.FilesystemCacheStorage:

   HTTPCACHE_STORAGE = 'scrapy_splash.SplashAwareFSCacheStorage'
   

   If you use other cache storage then it is necesary to subclass it and replace all scrapy.util.request.request_fingerprint calls with scrapy_splash.splash_request_fingerprint.

Note

Steps (4) and (5) are necessary because Scrapy doesn't provide a way to override request fingerprints calculation algorithm globally; this could change in future.

There are also some additional options available. Put them into your settings.py if you want to change the defaults:

• SPLASH_COOKIES_DEBUG is False by default. Set to True to enable debugging cookies in the SplashCookiesMiddleware. This option is similar to COOKIES_DEBUG for the built-in scarpy cookies middleware: it logs sent and received cookies for all requests.
• SPLASH_LOG_400 is True by default - it instructs to log all 400 errors from Splash. They are important because they show errors occurred when executing the Splash script. Set it to False to disable this logging.
• SPLASH_SLOT_POLICY is scrapy_splash.SlotPolicy.PER_DOMAIN (as object, not just a string) by default. It specifies how concurrency & politeness are maintained for Splash requests, and specify the default value for slot_policy argument for SplashRequest, which is described below.

Usage

Requests

The easiest way to render requests with Splash is to use scrapy_splash.SplashRequest:

yield SplashRequest(url, self.parse_result,
    args={
        # optional; parameters passed to Splash HTTP API
        'wait': 0.5,

        # 'url' is prefilled from request url
        # 'http_method' is set to 'POST' for POST requests
        # 'body' is set to request body for POST requests
    },
    endpoint='render.json', # optional; default is render.html
    splash_url='<url>',     # optional; overrides SPLASH_URL
    slot_policy=scrapy_splash.SlotPolicy.PER_DOMAIN,  # optional
)

Alternatively, you can use regular scrapy.Request and 'splash' Request meta key:

yield scrapy.Request(url, self.parse_result, meta={
    'splash': {
        'args': {
            # set rendering arguments here
            'html': 1,
            'png': 1,

            # 'url' is prefilled from request url
            # 'http_method' is set to 'POST' for POST requests
            # 'body' is set to request body for POST requests
        },

        # optional parameters
        'endpoint': 'render.json',  # optional; default is render.json
        'splash_url': '<url>',      # optional; overrides SPLASH_URL
        'slot_policy': scrapy_splash.SlotPolicy.PER_DOMAIN,
        'splash_headers': {},       # optional; a dict with headers sent to Splash
        'dont_process_response': True, # optional, default is False
        'dont_send_headers': True,  # optional, default is False
        'magic_response': False,    # optional, default is True
    }
})

Use request.meta['splash'] API in middlewares or when scrapy.Request subclasses are used (there is also SplashFormRequest described below). For example, meta['splash'] allows to create a middleware which enables Splash for all outgoing requests by default.

SplashRequest is a convenient utility to fill request.meta['splash']; it should be easier to use in most cases. For each request.meta['splash'] key there is a corresponding SplashRequest keyword argument: for example, to set meta['splash']['args'] use SplashRequest(..., args=myargs).

• meta['splash']['args'] contains arguments sent to Splash. scrapy-splash adds some default keys/values to args:

  • 'url' is set to request.url;
  • 'http_method' is set to 'POST' for POST requests;
  • 'body' is set to to request.body for POST requests.

  You can override default values by setting them explicitly.

  Note that by default Scrapy escapes URL fragments using AJAX escaping scheme. If you want to pass a URL with a fragment to Splash then set url in args dict manually. This is handled automatically if you use SplashRequest, but you need to keep that in mind if you use raw meta['splash'] API.

  Splash 1.8+ is required to handle POST requests; in earlier Splash versions 'http_method' and 'body' arguments are ignored. If you work with /execute endpoint and want to support POST requests you have to handle http_method and body arguments in your Lua script manually.

• meta['splash']['cache_args'] is a list of argument names to cache on Splash side. These arguments are sent to Splash only once, then cached values are used; it allows to save network traffic and decreases request queue disk memory usage. Use cache_args only for large arguments which don't change with each request; lua_source is a good candidate (if you don't use string formatting to build it). Splash 2.1+ is required for this feature to work.

• meta['splash']['endpoint'] is the Splash endpoint to use. In case of SplashRequest render.html is used by default. If you're using raw scrapy.Request then render.json is a default (for historical reasons). It is better to always pass endpoint explicitly.

  See Splash HTTP API docs for a full list of available endpoints and parameters.

• meta['splash']['splash_url'] overrides the Splash URL set in settings.py.

• meta['splash']['splash_headers'] allows to add or change headers which are sent to Splash server. Note that this option is not for setting headers which are sent to the remote website.

• meta['splash']['slot_policy'] customize how concurrency & politeness are maintained for Splash requests.

  Currently there are 3 policies available:

  1. scrapy_splash.SlotPolicy.PER_DOMAIN (default) - send Splash requests to downloader slots based on URL being rendered. It is useful if you want to maintain per-domain politeness & concurrency settings.
  2. scrapy_splash.SlotPolicy.SINGLE_SLOT - send all Splash requests to a single downloader slot. It is useful if you want to throttle requests to Splash.
  3. scrapy_splash.SlotPolicy.SCRAPY_DEFAULT - don't do anything with slots. It is similar to SINGLE_SLOT policy, but can be different if you access other services on the same address as Splash.

• meta['splash']['dont_process_response'] - when set to True, SplashMiddleware won't change the response to a custom scrapy.Response subclass. By default for Splash requests one of SplashResponse, SplashTextResponse or SplashJsonResponse is passed to the callback.

• meta['splash']['dont_send_headers']: by default scrapy-splash passes request headers to Splash in 'headers' JSON POST field. For all render.xxx endpoints it means Scrapy header options are respected by default (http://splash.readthedocs.org/en/stable/api.html#arg-headers). In Lua scripts you can use headers argument of splash:go to apply the passed headers: splash:go{url, headers=splash.args.headers}.

  Set 'dont_send_headers' to True if you don't want to pass headers to Splash.

• meta['splash']['http_status_from_error_code'] - set response.status to HTTP error code when assert(splash:go(..)) fails; it requires meta['splash']['magic_response']=True. http_status_from_error_code option is False by default if you use raw meta API; SplashRequest sets it to True by default.

• meta['splash']['magic_response'] - when set to True and a JSON response is received from Splash, several attributes of the response (headers, body, url, status code) are filled using data returned in JSON:

  • response.headers are filled from 'headers' keys;
  • response.url is set to the value of 'url' key;
  • response.body is set to the value of 'html' key, or to base64-decoded value of 'body' key;
  • response.status is set to the value of 'http_status' key. When meta['splash']['http_status_from_error_code'] is True and assert(splash:go(..)) fails with an HTTP error response.status is also set to HTTP error code.

  Original URL, status and headers are available as response.real_url, response.splash_response_status and response.splash_response_headers.

  This option is set to True by default if you use SplashRequest. render.json and execute endpoints may not have all the necessary keys/values in the response. For non-JSON endpoints, only url is filled, regardless of the magic_response setting.

Use scrapy_splash.SplashFormRequest if you want to make a FormRequest via splash. It accepts the same arguments as SplashRequest, and also formdata, like FormRequest from scrapy:

>>> SplashFormRequest('http://example.com', formdata={'foo': 'bar'})
<POST http://example.com>

SplashFormRequest.from_response is also supported, and works as described in scrapy documentation.

Responses

scrapy-splash returns Response subclasses for Splash requests:

• SplashResponse is returned for binary Splash responses - e.g. for /render.png responses;
• SplashTextResponse is returned when the result is text - e.g. for /render.html responses;
• SplashJsonResponse is returned when the result is a JSON object - e.g. for /render.json responses or /execute responses when script returns a Lua table.

To use standard Response classes set meta['splash']['dont_process_response']=True or pass dont_process_response=True argument to SplashRequest.

All these responses set response.url to the URL of the original request (i.e. to the URL of a website you want to render), not to the URL of the requested Splash endpoint. "True" URL is still available as response.real_url.

SplashJsonResponse provide extra features:

• response.data attribute contains response data decoded from JSON; you can access it like response.data['html'].
• If Splash session handling is configured, you can access current cookies as response.cookiejar; it is a CookieJar instance.
• If Scrapy-Splash response magic is enabled in request (default), several response attributes (headers, body, url, status code) are set automatically from original response body:
  • response.headers are filled from 'headers' keys;
  • response.url is set to the value of 'url' key;
  • response.body is set to the value of 'html' key, or to base64-decoded value of 'body' key;
  • response.status is set from the value of 'http_status' key.

When response.body is updated in SplashJsonResponse (either from 'html' or from 'body' keys) familiar response.css and response.xpath methods are available.

To turn off special handling of JSON result keys either set meta['splash']['magic_response']=False or pass magic_response=False argument to SplashRequest.

Session Handling

Splash itself is stateless - each request starts from a clean state. In order to support sessions the following is required:

1. client (Scrapy) must send current cookies to Splash;
2. Splash script should make requests using these cookies and update them from HTTP response headers or JavaScript code;
3. updated cookies should be sent back to the client;
4. client should merge current cookies wiht the updated cookies.

For (2) and (3) Splash provides splash:get_cookies() and splash:init_cookies() methods which can be used in Splash Lua scripts.

scrapy-splash provides helpers for (1) and (4): to send current cookies in 'cookies' field and merge cookies back from 'cookies' response field set request.meta['splash']['session_id'] to the session identifier. If you only want a single session use the same session_id for all request; any value like '1' or 'foo' is fine.

For scrapy-splash session handling to work you must use /execute endpoint and a Lua script which accepts 'cookies' argument and returns 'cookies' field in the result:

function main(splash)
    splash:init_cookies(splash.args.cookies)

    -- ... your script

    return {
        cookies = splash:get_cookies(),
        -- ... other results, e.g. html
    }
end

SplashRequest sets session_id automatically for /execute endpoint, i.e. cookie handling is enabled by default if you use SplashRequest, /execute endpoint and a compatible Lua rendering script.

If you want to start from the same set of cookies, but then 'fork' sessions set request.meta['splash']['new_session_id'] in addition to session_id. Request cookies will be fetched from cookiejar session_id, but response cookies will be merged back to the new_session_id cookiejar.

Standard Scrapy cookies argument can be used with SplashRequest to add cookies to the current Splash cookiejar.

Examples

Get HTML contents:

import scrapy
from scrapy_splash import SplashRequest

class MySpider(scrapy.Spider):
    start_urls = ["http://example.com", "http://example.com/foo"]

    def start_requests(self):
        for url in self.start_urls:
            yield SplashRequest(url, self.parse, args={'wait': 0.5})

    def parse(self, response):
        # response.body is a result of render.html call; it
        # contains HTML processed by a browser.
        # ...

Get HTML contents and a screenshot:

import json
import base64
import scrapy
from scrapy_splash import SplashRequest

class MySpider(scrapy.Spider):

    # ...
        splash_args = {
            'html': 1,
            'png': 1,
            'width': 600,
            'render_all': 1,
        }
        yield SplashRequest(url, self.parse_result, endpoint='render.json',
                            args=splash_args)

    # ...
    def parse_result(self, response):
        # magic responses are turned ON by default,
        # so the result under 'html' key is available as response.body
        html = response.body

        # you can also query the html result as usual
        title = response.css('title').extract_first()

        # full decoded JSON data is available as response.data:
        png_bytes = base64.b64decode(response.data['png'])

        # ...

Run a simple Splash Lua Script:

import json
import base64
from scrapy_splash import SplashRequest


class MySpider(scrapy.Spider):

    # ...
        script = """
        function main(splash)
            assert(splash:go(splash.args.url))
            return splash:evaljs("document.title")
        end
        """
        yield SplashRequest(url, self.parse_result, endpoint='execute',
                            args={'lua_source': script})

    # ...
    def parse_result(self, response):
        doc_title = response.text
        # ...

More complex Splash Lua Script example - get a screenshot of an HTML element by its CSS selector (it requires Splash 2.1+). Note how are arguments passed to the script:

import json
import base64
from scrapy_splash import SplashRequest

script = """
-- Arguments:
-- * url - URL to render;
-- * css - CSS selector to render;
-- * pad - screenshot padding size.

-- this function adds padding around region
function pad(r, pad)
  return {r[1]-pad, r[2]-pad, r[3]+pad, r[4]+pad}
end

-- main script
function main(splash)

  -- this function returns element bounding box
  local get_bbox = splash:jsfunc([[
    function(css) {
      var el = document.querySelector(css);
      var r = el.getBoundingClientRect();
      return [r.left, r.top, r.right, r.bottom];
    }
  ]])

  assert(splash:go(splash.args.url))
  assert(splash:wait(0.5))

  -- don't crop image by a viewport
  splash:set_viewport_full()

  local region = pad(get_bbox(splash.args.css), splash.args.pad)
  return splash:png{region=region}
end
"""

class MySpider(scrapy.Spider):


    # ...
        yield SplashRequest(url, self.parse_element_screenshot,
            endpoint='execute',
            args={
                'lua_source': script,
                'pad': 32,
                'css': 'a.title'
            }
         )

    # ...
    def parse_element_screenshot(self, response):
        image_data = response.body  # binary image data in PNG format
        # ...

Use a Lua script to get an HTML response with cookies, headers, body and method set to correct values; lua_source argument value is cached on Splash server and is not sent with each request (it requires Splash 2.1+):

import scrapy
from scrapy_splash import SplashRequest

script = """
function main(splash)
  splash:init_cookies(splash.args.cookies)
  assert(splash:go{
    splash.args.url,
    headers=splash.args.headers,
    http_method=splash.args.http_method,
    body=splash.args.body,
    })
  assert(splash:wait(0.5))

  local entries = splash:history()
  local last_response = entries[#entries].response
  return {
    url = splash:url(),
    headers = last_response.headers,
    http_status = last_response.status,
    cookies = splash:get_cookies(),
    html = splash:html(),
  }
end
"""

class MySpider(scrapy.Spider):


    # ...
        yield SplashRequest(url, self.parse_result,
            endpoint='execute',
            cache_args=['lua_source'],
            args={'lua_source': script},
            headers={'X-My-Header': 'value'},
        )

    def parse_result(self, response):
        # here response.body contains result HTML;
        # response.headers are filled with headers from last
        # web page loaded to Splash;
        # cookies from all responses and from JavaScript are collected
        # and put into Set-Cookie response header, so that Scrapy
        # can remember them.

HTTP Basic Auth

If you need to use HTTP Basic Authentication to access Splash, use the SPLASH_USER and SPLASH_PASS optional settings:

SPLASH_USER = 'user'
SPLASH_PASS = 'userpass'

Another option is meta['splash']['splash_headers']: it allows to set custom headers which are sent to Splash server; add Authorization header to splash_headers if you want to change credentials per-request:

import scrapy
from w3lib.http import basic_auth_header

class MySpider(scrapy.Spider):
    # ...
    def start_requests(self):
        auth = basic_auth_header('user', 'userpass')
        yield SplashRequest(url, self.parse,
                            splash_headers={'Authorization': auth})

WARNING: Don't use HttpAuthMiddleware (i.e. http_user / http_pass spider attributes) for Splash authentication: if you occasionally send a non-Splash request from your spider, you may expose Splash credentials to a remote website, as HttpAuthMiddleware sets credentials for all requests unconditionally.

Why not use the Splash HTTP API directly?

The obvious alternative to scrapy-splash would be to send requests directly to the Splash HTTP API. Take a look at the example below and make sure to read the observations after it:

import json

import scrapy
from scrapy.http.headers import Headers

RENDER_HTML_URL = "http://127.0.0.1:8050/render.html"

class MySpider(scrapy.Spider):
    start_urls = ["http://example.com", "http://example.com/foo"]

    def start_requests(self):
        for url in self.start_urls:
            body = json.dumps({"url": url, "wait": 0.5}, sort_keys=True)
            headers = Headers({'Content-Type': 'application/json'})
            yield scrapy.Request(RENDER_HTML_URL, self.parse, method="POST",
                                 body=body, headers=headers)

    def parse(self, response):
        # response.body is a result of render.html call; it
        # contains HTML processed by a browser.
        # ...

It works and is easy enough, but there are some issues that you should be aware of:

1. There is a bit of boilerplate.
2. As seen by Scrapy, we're sending requests to RENDER_HTML_URL instead of the target URLs. It affects concurrency and politeness settings: CONCURRENT_REQUESTS_PER_DOMAIN, DOWNLOAD_DELAY, etc could behave in unexpected ways since delays and concurrency settings are no longer per-domain.
3. As seen by Scrapy, response.url is an URL of the Splash server. scrapy-splash fixes it to be an URL of a requested page. "Real" URL is still available as response.real_url. scrapy-splash also allows to handle response.status and response.headers transparently on Scrapy side.
4. Some options depend on each other - for example, if you use timeout Splash option then you may want to set download_timeout scrapy.Request meta key as well.
5. It is easy to get it subtly wrong - e.g. if you won't use sort_keys=True argument when preparing JSON body then binary POST body content could vary even if all keys and values are the same, and it means dupefilter and cache will work incorrectly.
6. Default Scrapy duplication filter doesn't take Splash specifics in account. For example, if an URL is sent in a JSON POST request body Scrapy will compute request fingerprint without canonicalizing this URL.
7. Splash Bad Request (HTTP 400) errors are hard to debug because by default response content is not displayed by Scrapy. SplashMiddleware logs content of HTTP 400 Splash responses by default (it can be turned off by setting SPLASH_LOG_400 = False option).
8. Cookie handling is tedious to implement, and you can't use Scrapy built-in Cookie middleware to handle cookies when working with Splash.
9. Large Splash arguments which don't change with every request (e.g. lua_source) may take a lot of space when saved to Scrapy disk request queues. scrapy-splash provides a way to store such static parameters only once.
10. Splash 2.1+ provides a way to save network traffic by caching large static arguments on server, but it requires client support: client should send proper save_args and load_args values and handle HTTP 498 responses.

scrapy-splash utlities allow to handle such edge cases and reduce the boilerplate.

Getting help

• for problems with rendering pages read "Splash FAQ" page
• for Scrapy-related bugs take a look at "reporting Scrapy bugs" page

Best approach to get any other help is to ask a question on Stack Overflow

Contributing

Source code and bug tracker are on github: https://github.com/scrapy-plugins/scrapy-splash

To run tests, install "tox" Python package and then run tox command from the source checkout.

To run integration tests, start Splash and set SPLASH_URL env variable to Splash address before running tox command:

docker run -d --rm -p8050:8050 scrapinghub/splash:3.0
SPLASH_URL=http://127.0.0.1:8050 tox -e py36