Update: if using versions since Chrome 67/chromedriver 2.39, my alternative answer above provides a simpler solution
The core issue here is Chromedriver also uses the remote debugging port connection to communicate with Chrome. This uses the websocket protocol, which only supports a single-client connected at a time. Normally, when chromedriver starts the chromedriver process, it will choose a random free TCP port number and use this to access the remote debug port. If you specify --remote-debuggging-port=9222 however, Chrome will be opened with the debug port you have requested, but chromedriver will silently continue to try and open a connection using this random port number.
The solution I ended up with was heavily inspired by comment 20 in this chromedriver issue. It required quite a bit of code to get it working, but works solidly. It makes uses of a nodejs library, crmux - which allows multiple clients to connect to the remote debug port of chrome at the same time.
- Get nodejs installed first: Nodejs v9.7.0 works fine
- Install crmux by running
npm install crmux -g
- Before you start chromedriver (
Capybara::Selenium::Driver.new), you need to spawn a separate thread that will do a few things: look for the remote debug port chromedriver is trying to use to connect to chrome, and then use this to fire up crmux. Once this has happened Capybara etc will work as normal.
- My separate thread runs a ruby script that first executes a netstat command repeatedly until it finds the relevant entry for chromedriver (TCP status is SYN_SENT). This separate thread must continue to run in the background while chrome is up and running.
The code for this is:
$chrdrv_wait_timeout = 60
$chrdrv_exe = "chromedriver.exe"
def get_netstat_output
stdout = `netstat -a -b -n`
stat_lines = stdout.split("\n")
stat_lines
end
def try_get_requested_port
socket_state = "SYN_SENT" # i.e. sent with no reply
statout = get_netstat_output
n = statout.length
i = 0
loop do
i += 1
# find lines relating to chromedriver
exe_match = /^ +\[#{$chrdrv_exe}\]$/.match statout[i]
if exe_match != nil
# check preceeding lines which should contain port info
port_match = /TCP.*:([0-9]+)\W+#{socket_state}/.match statout[i-1]
if port_match != nil
return port_match[1].to_i
end
end
break unless i < n
end
return nil
end
def get_tcp_port_requested_by_chromedriver
i = 1
loop do
puts "Waiting for #{$chrdrv_exe}: #{i}"
port = try_get_requested_port
if port != nil
return port
end
break unless i < $chrdrv_wait_timeout
sleep 1
i += 1
end
raise Exception, "Failed to get TCP port requested by #{$chrdrv_exe} (gave up after #{$chrdrv_wait_timeout} seconds)"
end
(I'm working in Windows: for Mac/Linux the netstat syntax/output is probably different so the code will need adjustment; the key thing is you need it to output the executable owner of each connection entry - and parse the bit relating to chromedriver to get the port in question).
Once this has found the random port (I'll use 12225 as an example), the background ruby script can then execute a crmux process, which will reunite chromedriver with chrome itself via the port you specified in Capybara (4444):
crmux --port=4444 --listen=12225
Finally, this separate script saves the discovered listen port to a text file. This allows the main script/thread that is running capybara to know the port number it can use to get access to chrome (via crmux's multiplexed connection) by reading in the port from that file. So you can then use chrome_remote to access chrome using port 12225, for example, while capybara is doing its thing.
