Chromewhip - Google Chrome™ as a web service

Chrome browser as an HTTP service with an splash compatible HTTP API

Chromewhip is an easily deployable service that runs headless Chrome process wrapped with an HTTP API. Inspired by the splash project, we aim to provide a drop-in replacement for the splash service by adhering to their documented API.

It is currently in early alpha and still being heavily developed. Please use the issue tracker to track the progress towards beta. For now, the required milestone can be summarised as implementing the entire Splash API.

How to use as a service

One can simply deploy as a Docker container and use the API that is served on port 8080.

docker run --init -it --rm --shm-size=1024m -p=127.0.0.1:8080:8080 --cap-add=SYS_ADMIN \
  chuckus/chromewhip

Refer to the HTTP API reference at the bottom of the README for what features are available.

How to use the low-level driver

As part of the Chromewhip service, a Python 3.6 asyncio compatible driver for Chrome devtools protocol was developed and can be leveraged without having to run the HTTP server. The advantages of our devtools driver are:

Typed Python bindings for devtools protocol through templated generation - get autocomplete with your code editor.
Can bind events to concurrent commands, which is required for providing a robust HTTP service.

Prerequisites

Before executing the code below, please have the following:

Google Chrome Canary running with flag --remote-debugging-port=9222

Example driver code

import asyncio
import logging

from chromewhip import Chrome
from chromewhip.protocol import browser, page, dom

# see logging from chromewhip
logging.basicConfig(level=logging.DEBUG)

HOST = '127.0.0.1'
PORT = 9222

loop = asyncio.get_event_loop()
c = Chrome(host=HOST, port=PORT)

loop.run_until_complete(c.connect())

# use the startup tab or create a new one
tab = c.tabs[0]
tab = loop.run_until_complete(c.create_tab())

loop.run_until_complete(tab.enable_page_events())

def sync_cmd(*args, **kwargs):
    return loop.run_until_complete(tab.send_command(*args, **kwargs))

# send_command will return once the frameStoppedLoading event is received THAT matches
# the frameId that it is in the returned command payload.
result = sync_cmd(page.Page.navigate(url='http://nzherald.co.nz'), 
                  await_on_event_type=page.FrameStoppedLoadingEvent)

# send_command always returns a dict with keys `ack` and `event`
# `ack` contains the payload on response of a command
# `event` contains the payload of the awaited event if `await_on_event_type` is provided
ack = result['ack']['result']
event = result['event']
assert ack['frameId'] == event.frameId

sync_cmd(page.Page.setDeviceMetricsOverride(width=800,
                                            height=600,
                                            deviceScaleFactor=0.0,
                                            mobile=False))

result = sync_cmd(dom.DOM.getDocument())

dom_obj = result['ack']['result']['root']

# Python types are determined by the `types` fields in the JSON reference for the
# devtools protocol, and `send_command` will convert if possible.
assert isinstance(dom_obj, dom.Node)

print(dom_obj.nodeId)
print(dom_obj.nodeName)

# close the tab
loop.run_until_complete(c.close_tab(tab))

# or close the browser via Devtools API
tab = c.tabs[0]
sync_cmd(browser.Browser.close())

Implemented HTTP API

/render.html

Query params:

url : string : required
- The url to render (required)
js : string : optional Javascript profile name.
js_source : string : optional
- JavaScript code to be executed in page context
viewport : string : optional
- View width and height (in pixels) of the browser viewport to render the web page. Format is "x", e.g. 800x600. Default value is 1024x768.
'viewport' parameter is more important for PNG and JPEG rendering; it is supported for all rendering endpoints because javascript code execution can depend on viewport size.

/render.png

Query params (including render.html):

render_all : int : optional
- Possible values are 1 and 0. When render_all=1, extend the viewport to include the whole webpage (possibly very tall) before rendering.

Why not just use Selenium?

chromewhip uses the devtools protocol instead of the json wire protocol, where the devtools protocol has greater flexibility, especially when it comes to subscribing to granular events from the browser.

Bug reports and requests

Please simply file one using the Github tracker

Contributing

Please :)

How to regenerate the Python protocol files

In scripts, you can run regenerate_protocol.sh, which downloads HEAD of offical devtools specs, regenerates, runs some sanity tests and creates a commit with the message of official devtools specs HEAD.

From time to time, it will fail, due to desynchronization of the chromewhip patch with the json specs, or mistakes in the protocol.

Under data, there are *_patch files, which follow the RFC 6902 JSON Patch notation. You will see that there are some checks to see whether particular items in arrays exist before patching. If you get a jsonpatch.JsonPatchTestFailed exception, it's likely to desynchronization, so check the official spec and adjust the patch json file.

Implementation

Developed to run on Python 3.6, it leverages both aiohttp and asyncio for the implementation of the asynchronous HTTP server that wraps chrome.