3

I have been getting a series of pesky Selenium/Chrome errors for a week or so, where everything works fine for a while and then suddenly doesn't. I have the right version of Chromedriver (89.0.4389.23) for my version of Google Chrome (89.0.4389.114) per Google Chrome official docs

Currently I'm getting the following error when I run driver = webdriver.Chrome(options=options) (at least on my production server, I don't get it for the same code repo on my local machine):

selenium.common.exceptions.SessionNotCreatedException: Message: session not created
from disconnected: unable to connect to renderer
  (Session info: headless chrome=89.0.4389.114)

Frustratingly, the only solution I've found for this error aside from "make sure your chrome and chromedriver match" is to remove options.add_argument("--remote-debugging-port=9222") from my webdriver options. I do have that line of code as well, and removing it appears to solve my problem for now, but my problem that line of code appears under the following comments in my codebase:

    # adding arg below fixed "DevToolsActivePort file doesn't exist."
    # ht https://github.com/heroku/heroku-buildpack-google-chrome/issues/46
    options.add_argument("--remote-debugging-port=9222")

So I've now reached a perfect circle where my Python Selenium code reliably works until it reliably breaks, with either of two different Selenium errors ("session not created from disconnected: unable to connect to renderer" or "DevToolsActivePort file doesn't exist."), where the solution to one is to add a specific line of code and the solution to the other is to remove it.

Below is my current launch_webdriver function, with comments reflecting the various transient Chrome/Selenium errors I've been trying to navigate:

from selenium import webdriver
import chromedriver_binary
from selenium.webdriver.chrome.options import Options

def launch_browser(headless=False):  # :, driverpath="./spinocchio/chromedriver"):
    options = Options()

    # if any other driver `options` are added, they must be added BELOW no-sandbox
    # per user parsecer's comment at https://stackoverflow.com/a/53073789/1870832
    # and my own excruciating experience.
    options.add_argument("--no-sandbox")

    # you may think the two lines below can be replaced with just options.headless=headless
    # please don't.
    if headless:
        options.add_argument("headless")

    # adding arg below fixed "DevToolsActivePort file doesn't exist."
    # ht https://github.com/heroku/heroku-buildpack-google-chrome/issues/46
    options.add_argument("--remote-debugging-port=9222")
    driver = webdriver.Chrome(options=options)
    return driver

How can I get to something more reliable/robust?

halfer
  • 19,824
  • 17
  • 99
  • 186
Max Power
  • 8,265
  • 13
  • 50
  • 91
  • 4
    This is why we need quantum computing. You can both have and not have the option set at the same time. It's Schrödinger's option. – MattDMo Apr 17 '21 at 23:44
  • posted bug report at chromium project (not sure if this is the right place vs selenium?): https://bugs.chromium.org/p/chromium/issues/detail?id=1206878 – Max Power May 08 '21 at 01:39
  • Similar issue, check it https://stackoverflow.com/questions/53808653/sessionnotcreatedexception-message-session-not-created-from-disconnected-unab – vitaliis May 08 '21 at 03:31
  • This might be a stupid question, but on your *product system* do you have only one *chromedriver* in the directory that contains the *chromedriver* used in your path? Also Chrome 89.0.4389.114 was release at the end of March 2021 and chromedriver 89.0.4389.23 was release at the end of January 2021, so there might be a compatibility issue. I noted that the release notes for 89.0.4389.23 mentioned a renderer issue that the release was solving. – Life is complex May 09 '21 at 20:27
  • @vitalis, thanks for the link. I have already confirmed the things suggested by the accepted answer, but Sebastian's other answer there seems like he also encountered both my errors one after the other like I have and suggests a promising approach I haven't tried yet. – Max Power May 09 '21 at 21:06
  • Can you also share your executable path of chromedriver ? – Keyur Shrimali May 14 '21 at 17:01
  • @keyur this is handled by chromedriver_binary library – Max Power May 14 '21 at 17:44

3 Answers3

1

Overview

  1. Prevalent issue - per @ss7777
  2. Albeit numerous potential causes for this/related

Blue-skies and 'silver-bullet' solution I'm afraid, however, here are some common causes and potential resolve / considerations etc.

Fixes/resolve

(accompanied by subj. view of 'commonality' | frequency)

  1. Driver: ChromeDriver version/file location (~50%) - incorrect version + ensure location to executable correct location

"This issue is pretty evident when there is a incompatibility between the version of the binaries you are using." (adapted from this source, with useful summary of historical versions v/ binaries)

  1. Chrome/driver still 'active' 5-20% - despite closing down, process tree elements can remain 'alive' or 'spinning' for some time - do you notice this right after restart of PC (here)?

(Have suffered and even evidenced this previously - e.g. running code in parallel, or simply executing through .bat at same time as PyCharm console).

  1. Updates: plugin/other incompatibly (~5-15%) - noting you're using headless - does it work with head?)
  2. System: clean up your system/restore-point or worst-face (~10-20%). (restore point worked for me)
  3. Other: spawn separate thread (~5% - see here)/alternative ports (as req/feasible)

Workarounds

  • Chrome dependency (e.g. see here - albeit potential red-herring..)
  • Mode of execution (through editor main program, through CMD interface hosted via editor (e.g. PyCharm console), or through .bat file (as mentioned previously), Standalone Selenium here
  • Configuration (technical/advanced - e.g. here)for advanced users)

Unresolved (albeit with promise)

Several cases of where such a Q remains unresolved/limited contribution:

(although 2nd point here points to remedy in item 2 above).

Extreme solns:

  • Have seen intense investigation and many too-fro re Q&A - with ultimate conclusion being complete reinstallation (Chrome/python, editor) etc.

Links / related

JB-007
  • 2,156
  • 1
  • 6
  • 22
  • what has helped me is to make sure there is a gap of time between quitting one driver and starting another. Obviously if you are multi-threading this can be difficult to handle. Without Chrome dropping that file with sessionID you can't create a session, and without it cleaning up properly, the next file may not drop or the next driver session could be using an old sessionID. I think the port is somewhat related there? It'll pick a random port if not set? – pcalkins May 11 '21 at 20:53
  • Yes- interesting, this makes a difference too which I suspected (initially) was simply fickle / volatile results. This is a far bigger issue than you might think based upon my research. – JB-007 May 17 '21 at 21:08
0

I faced the same error once and I couldn't find a feasible solution anywhere. So I degraded my chrome version to v86.0
Even when you install the older version of chrome, you need to disable the chrome update to continue with your intended version. You can find the the steps here: https://www.webnots.com/7-ways-to-disable-automatic-chrome-update-in-windows-and-mac/
Along with the lower bowser version, I use the following code:

def test_setup(self):
    chrome_options=Options()
    chrome_options.add_argument('--start-maximized')
    chrome_options.add_experimental_option("excludeSwitches", ["enable-logging"])
    chrome_options.add_argument("--remote-debugging-port=9222")
    chrome_options.add_argument('--no-sandbox')
    driver=webdriver.Chrome(executable_path="//your folder path//")

Try this once and let me know if it worked for you

Libin Thomas
  • 829
  • 9
  • 18
  • Thanks Libin, do you know if you still need to disable chrome auto updates in Linux as well? I'd guess not, right? – Max Power May 10 '21 at 14:47
  • I guess not. But when you install chrome browser on Linux computers, the Google repository is also installed. The repository ensures that your system automatically keeps chrome browser up to date. Need to check if its true – Libin Thomas May 11 '21 at 06:25
  • 1
    hm thanks that makes sense. so after apt installing chrome, I think you're right that the next apt update would update chrome (among other packages). hm. – Max Power May 11 '21 at 14:07
  • Sorry - I really tired that (various different versions of Chrome in fact) - and found same issues referred to in various other related threads re: topic. – JB-007 May 17 '21 at 21:06
0

If you get this error.
Try this way:

from webdriver_manager.chrome import ChromeDriverManager
options.add_argument("--restore-last-session")
options.add_argument("--disable-gaia-services")
driver = webdriver.Chrome(ChromeDriverManager().install(),options=options)

https://peter.sh/experiments/chromium-command-line-switches/

top talent
  • 615
  • 4
  • 17