To scrape a pool of URLs, I am paralell processing selenium with joblib. In this context, I am facing two challenges:
Challenge 1 is to speed up this process. In the moment, my code opens and closes a driver instance for every URL (ideally would be one for every process)
Challenge 2 is to get rid of the CPU-intensive while loop that I think I need to continue on empty results (I know that this is most likely wrong)
Pseudocode:
URL_list = [URL1, URL2, URL3, ..., URL100000] # List of URLs to be scraped
def scrape(URL):
while True: # Loop needed to use continue
try: # Try scraping
driver = webdriver.Firefox(executable_path=path) # Set up driver
website = driver.get(URL) # Get URL
results = do_something(website) # Get results from URL content
driver.close() # Close worker
if len(results) == 0: # If do_something() failed:
continue # THEN Worker to skip URL
else: # If do_something() worked:
safe_results("results.csv") # THEN Save results
break # Go to next worker/URL
except Exception as e: # If something weird happens:
save_exception(URL, e) # THEN Save error message
break # Go to next worker/URL
Parallel(n_jobs = 40)(delayed(scrape)(URL) for URL in URL_list))) # Run in 40 processes
My understanding is that in order to re-use a driver instance across iterations, the # Set up driver-line needs to be placed outside scrape(URL). However, everything outside scrape(URL) will not find its way to joblib's Parallel(n_jobs = 40). This would imply that you can't reuse driver instances while scraping with joblib which can't be true.
Q1: How to reuse driver instances during parallel processing in the above example?
Q2: How to get rid of the while-loop while maintaining functionality in the above-mentioned example?
Note: Flash and image loading is disabled in firefox_profile (code not shown)
1) You should first create a bunch of drivers: one for each process. And pass an instance to the worker. I don't know how to pass drivers to an Prallel object, but you could use threading.current_thread().name key to identify drivers. To do that, use backend="threading". So now each thread will has its own driver.
2) You don't need a loop at all. Parallel object itself iter all your urls (I hope I realy understend your intentions to use a loop)
import threading
from joblib import Parallel, delayed
from selenium import webdriver
def scrape(URL):
try:
driver = drivers[threading.current_thread().name]
except KeyError:
drivers[threading.current_thread().name] = webdriver.Firefox()
driver = drivers[threading.current_thread().name]
driver.get(URL)
results = do_something(driver)
if results:
safe_results("results.csv")
drivers = {}
Parallel(n_jobs=-1, backend="threading")(delayed(scrape)(URL) for URL in URL_list)
for driver in drivers.values():
driver.quit()
But I don't realy think you get profit in using n_job more than you have CPUs. So n_jobs=-1 is the best (of course I may be wrong, try it).
Related
I am trying to open a site multiple times in a loop to test if different credentials have expired so that I can notify our users. I'm achieving this by opening the database, getting the records, calling the chrome driver to open the site, and inputting the values into the site. The first loop works but when the next one initiates the driver hangs and eventually outputs the error:
"unknown error: cannot connect to chrome at 127.0.0.1:XXXX from chrome not reachable"
This error commonly occurs when there is already an instance running. I have tried to prevent this by using both driver.close() and driver.quit() when the first loop is done but to no avail. I have taken care of all other possibilities of detection such as using proxies, different user agents, and also using the undetected_chromedriver by https://github.com/ultrafunkamsterdam/undetected-chromedriver.
The core issue I am looking to solve is being able to open an instance of the chrome driver, close it and open it back again all in the same execution loop until all the credentials I am testing have finished. I have abstracted the code and provided an isolated version that replicates the issue:
# INSTALL CHROMDRIVER USING "pip install undetected-chromedriver"
import undetected_chromedriver.v2 as uc
# Python Libraries
import time
options = uc.ChromeOptions()
options.add_argument('--no-first-run')
driver = uc.Chrome(options=options)
length = 8
count = 0
if count < length:
print("Im outside the loop")
while count < length:
print("This is loop ",count)
time.sleep(2)
with driver:
print("Im inside the loop")
count =+ 1
driver.get("https://google.com")
time.sleep(5)
print("Im at the end of the loop")
driver.quit() # Used to exit the browser, and end the session
# driver.close() # Only closes the window in focus
I recommend using a python virtualenv to keep packages consistent. I am using python3.9 on a Linux machine. Any solutions, suggestions, or workarounds would be greatly appreciated.
You are quitting your driver in the loop and then trying to access the executor address, which no longer exists, hence your error. You need to reinitialize the driver by moving it down within the loop, before the while statement.
from multiprocessing import Process, freeze_support
import undetected_chromedriver as uc
# Python Libraries
import time
chroptions = uc.ChromeOptions()
chroptions.add_argument('--no-first-run enable_console_log = True')
# driver = uc.Chrome(options=chroptions)
length = 8
count = 0
if count < length:
print("Im outside the loop")
while count < length:
print("This is loop ",count)
driver = uc.Chrome(options=chroptions)
time.sleep(2)
with driver:
print("Im inside the loop")
count =+ 1
driver.get("https://google.com")
print("Session ID: ", end='') #added to show your session ID is changing
print(driver.session_id)
driver.quit()
I'm running a Python script which uses Selenium, on an EC2 instance (ubuntu).
Under my AWS plan, I have 2 GB memory. I upgraded to this from the free version after some performance issues with my script. However, when I check the free memory when connected to the Ubuntu server, I'm only seeing 348MB of available memory, and 353MB of free memory!
As of now, I'm only running two Python scripts, once a day, using crontab. The scripts run through a fairly long array of URLs and grabs information from each of them.
base_url = 'https://www.bandsintown.com/en/c/san-francisco-ca?page='
events = []
eventContainerBucket = []
for i in range(1,25):
#cycle through pages in range
driver.get(base_url + str(i))
pageURL = base_url + str(i)
# get events links
event_list = driver.find_elements_by_css_selector('div[class^=_3buUBPWBhUz9KBQqgXm-gf] a[class^=_3UX9sLQPbNUbfbaigy35li]')
# collect href attribute of events in even_list
events.extend(list(event.get_attribute("href") for event in event_list))
allEvents = []
for event in events:
driver.get(event)
//do a bunch of other stuff
driver.quit()
Is there an inherent problem in this code that would be causing memory leak? I would think free memory should go up again when the script has stopped running, but it doesn't.
I tried to invoke driver.close() within the for-loop, so that after the information is retrieved from each URL, the window closes. I was thinking this would help with memory leak - unfortunately, this gave me an error selenium.common.exceptions.InvalidSessionIdException: Message: invalid session id.
Am I on the right path with driver.close() or is something else entirely the problem?
Instead of driver quit, have you tried driver.close()?
close() - It is used to close the browser
quite() - It is used to shut down the web driver instance.
for i in range(1,25):
#cycle through pages in range
driver.get(base_url + str(i))
pageURL = base_url + str(i)
# get events links
event_list = driver.find_elements_by_css_selector('div[class^=_3buUBPWBhUz9KBQqgXm-gf] a[class^=_3UX9sLQPbNUbfbaigy35li]')
# collect href attribute of events in even_list
events.extend(list(event.get_attribute("href") for event in event_list))
allEvents = []
for event in events:
driver.get(event)
driver.close()
I have checked your code its working fine without any issue. Please find below screenshot for more details
I have written functin based on selenium and I want it to parse simultaneously multiple webpages. I have list of urls that I pass to the function that I want scrape at the same time so as to save time.
I created scraper.py file where i put scraper function:
def parser_od(url):
price=[]
url_of = url
driver.get(url_of)
try:
price.append(browser.find_element_by_xpath("//*[#id='root']/article/header/div[2]/div[1]/div[2]").text.replace(" ","").replace("zł","").replace(",","."))
except NoSuchElementException:
price.append("")
Now I want to use the function to parse multiple urls from my urls at the same time using multiprocessing library:
from scraper import *
url_list=['https://www.otodom.pl/oferta/2-duze-pokoje-we-wrzeszczu-do-zamieszania-ID42f6s',
'https://www.otodom.pl/oferta/mieszkanie-na-zamknietym-osiedlu-z-ogrodkiem-ID40ZxM',
'https://www.otodom.pl/oferta/zaciszna-nowe-mieszkanie-3-pokoje-0-ID41UaX',
'https://www.otodom.pl/oferta/dwupoziomowe-dewel-mieszkanie-101-m2-lebork-i-p-ID3JEcQ']
driver = webdriver.Chrome(executable_path=r"C:\Users\Admin\chromedriver.exe")
from multiprocessing import Pool
with Pool(4) as p:
price = p.map(parser_od, url_list)
But I get following error:
NameError: name 'driver' is not defined
Which is weird because chrome is opened up.
Edit:
I need to have the browser(s) open while running this scraper, so that the driver is opened before not everytime this function is invoked.
Just should probably just split up the list of urls you want to process ino 4 equal parts, and have a driver for each process that processes one of the equal parts in the Pool.
def parser_od(urls, thread_index):
driver = webdriver.Chrome(executable_path=r"C:\Users\Admin\chromedriver.exe")
prices = []
for i in range(len(urls)):
url = urls[i]
if i % 4 == thread_index:
price=[]
url_of = url
driver.get(url_of)
try:
price.append(browser.find_element_by_xpath("//*[#id='root']/article/header/div[2]/div[1]/div[2]").text.replace(" ","").replace("zł","").replace(",","."))
except NoSuchElementException:
price.append("")
prices.append(price)
return prices
from multiprocessing import Pool
with Pool(4) as p:
price = p.map(lambda x: parser_od(x, url_list), list(range(len(url_list))))
I am trying to run selenium using ThreadsPoolExecutor. The website requires a login and I am trying to speed up a step in what I am trying to do in the website. But everytime a thread opens chrome, I need to relogin and it sometimes just hangs. I login once first without using threads to do some processing. And from here on, I like to open a few chome webdrivers without the need to relogin. Is there a way around this? PS: website has no id and password strings in the url.
def startup(dirPath):
# Start the WebDriver, load options
options = webdriver.ChromeOptions()
options.add_argument("--disable-infobars")
options.add_argument("--enable-file-cookies")
params = {'behavior': 'allow', 'downloadPath': dirPath}
wd = webdriver.Chrome(options=options, executable_path=r"C:\Chrome\chromedriver.exe")
wd.execute_cdp_cmd('Page.setDownloadBehavior', params)
# wd.delete_all_cookies()
wd.set_page_load_timeout(30)
wd.implicitly_wait(10)
return wd
def webLogin(dID, pw, wd):
wd.get('some url')
# Login, clear any outstanding login in id
wd.find_element_by_id('username').clear()
wd.find_element_by_id('username').send_keys(dID)
wd.find_element_by_id('password').clear()
wd.find_element_by_id('password').send_keys(pw)
wd.find_element_by_css_selector('.button').click()
if __name__ == '__main__':
dirPath, styleList = firstProcessing()
loginAndClearLB(dID, dPw, dirPath) # calls startup & webLogin, this is also my 1st login
# many webdrivers spawned here
with concurrent.futures.ThreadPoolExecutor(max_workers=5) as executor:
results = {executor.submit(addArtsToLB, dID, dPw, dirPath, style): style for style in styleList}
#Do other stuff
wd2 = startup(dirPath)
webLogin(dID, dPw, wd2)
startDL(wd2)
logOut(wd2, dirPath)
Any help would be greatly appreciated. Thanks!!
Like mentioned above, you could obtain the authentication token from the first login and than include it in all the subsequent requests.
However, another option (if you're using basic auth) is to just add the username and password into the URL, like:
https://username:password#your.domain.com
ok it looks like there is no solution yet for more complicated websites that do not use basic authentication. My modified Solution:
def webOpenWithCookie(wd, cookies):
wd.get('https://some website url/404')
for cookie in cookies:
wd.add_cookie(cookie)
wd.get('https://some website url/home')
return wd
def myThreadedFunc(dirPath, style, cookies): # this is the function that gets threaded
wd = startup(dirPath) # just starts chrome
wd = webOpenWithCookie(wd, cookies) # opens a page in the site and adds cookies to wd and then open your real target page. No login required now.
doSomethingHere(wd, style)
wd.quit() # close all the threads here better I think
if __name__ == '__main__':
dirPath, styleList = firstProcessing()
wd1 = startup(dirPath)
wd1 = webLogin(dID, dPw, wd1) # here i login once
cookies = wd1.get_cookies() # get the cookie from here
with concurrent.futures.ThreadPoolExecutor(max_workers=2) as executor:
results = {executor.submit(myThreadedFunc, dirPath, style, cookies): style for style in styleList} # this spawns threads, but each thread will not need login although the compromise is it needs to go to 404 page first.
I have built a webscraper using python and selenium with geckodriver, it is currently running in an EC2 instance on a crontab schedule.
My issue is it takes more than 5 minutes to finish downloading and I want to use lamda functions to run my scraper but they only allow for 5 minutes of runtime.
So I have a code similar to this.
from selenium import webdriver
def start_browser(url):
browser = webdriver.Firefox( executable_path="./geckodriver")
executable_path="./geckodriver")
browser.get(url)
return browser
def log_in(user, pass, user_elem, pass_elem, login_elem, browser):
user_elem.click().send_keys(user)
pass_elem.click().send_keys(pass)
login_elem.click()
return browser
def nav_to_data(browser, data_elem)
data_elem.click()
return browser
def find_data(browser, data_table)
data_links = data_table.find_elements_by_tag_name("tr")
return data_links, browser
I'm thinking these functions could be ran on lambda functions passing the browser/webdriver instance to each other?
The part I'm struggling with is looping through the data and waiting for all downloads to finish, this would take longer than 5 mins.
Is there anyway around this?
def download_data(browser, link)
link.click()
time.sleep(2)
download_elem = browser.find_element_by_id("download_xls_file")
download_path = download_elem.click()
return download_path
# THIS TAKES LONGER THAN 5 mins
download_paths = []
for link in data_links:
download = download_data(browser, link) # clicks a link to a new page wdownload button and returns path to the .xls file
download_paths.append(download)
upload_data()
You can partition your data and use a recursive lambda to process chunks of your list.
Taking an example from my blog
def invoke_self_async(data_list, context):
this_data_list = data_list[0:20] # increase number as needed
new_event = {
'data': data_list[20:] # needs to match above number
}
boto3.client('lambda').invoke_async(
FunctionName=context.invoked_function_arn,
InvokeArgs=json.dumps(new_event)
)
my_data = []
for data in data_list:
download = download_data(browser, data) # returns path to .xls file
my_data.append(download)
return my_data