AIL-framework/bin/Crawler.py

454 lines
19 KiB
Python
Executable File

#!/usr/bin/env python3
# -*-coding:UTF-8 -*
import os
import sys
import re
import uuid
import json
import redis
import datetime
import time
import subprocess
import requests
from pyfaup.faup import Faup
sys.path.append(os.environ['AIL_BIN'])
from Helper import Process
from pubsublogger import publisher
# ======== FUNCTIONS ========
def load_blacklist(service_type):
try:
with open(os.environ['AIL_BIN']+'/torcrawler/blacklist_{}.txt'.format(service_type), 'r') as f:
redis_crawler.delete('blacklist_{}'.format(service_type))
lines = f.read().splitlines()
for line in lines:
redis_crawler.sadd('blacklist_{}'.format(service_type), line)
except Exception:
pass
def update_auto_crawler():
current_epoch = int(time.time())
list_to_crawl = redis_crawler.zrangebyscore('crawler_auto_queue', '-inf', current_epoch)
for elem_to_crawl in list_to_crawl:
mess, type = elem_to_crawl.rsplit(';', 1)
redis_crawler.sadd('{}_crawler_priority_queue'.format(type), mess)
redis_crawler.zrem('crawler_auto_queue', elem_to_crawl)
# Extract info form url (url, domain, domain url, ...)
def unpack_url(url):
to_crawl = {}
faup.decode(url)
url_unpack = faup.get()
# # FIXME: # TODO: remove me
try:
to_crawl['domain'] = url_unpack['domain'].decode()
except:
to_crawl['domain'] = url_unpack['domain']
to_crawl['domain'] = to_crawl['domain'].lower()
# force lower case domain/subdomain (rfc4343)
# # FIXME: # TODO: remove me
try:
url_host = url_unpack['host'].decode()
except:
url_host = url_unpack['host']
new_url_host = url_host.lower()
url_lower_case = url.replace(url_host, new_url_host, 1)
if url_unpack['scheme'] is None:
to_crawl['scheme'] = 'http'
url= 'http://{}'.format(url_lower_case)
else:
# # FIXME: # TODO: remove me
try:
scheme = url_unpack['scheme'].decode()
except Exception as e:
scheme = url_unpack['scheme']
if scheme in default_proto_map:
to_crawl['scheme'] = scheme
url = url_lower_case
else:
redis_crawler.sadd('new_proto', '{} {}'.format(scheme, url_lower_case))
to_crawl['scheme'] = 'http'
url= 'http://{}'.format(url_lower_case.replace(scheme, '', 1))
if url_unpack['port'] is None:
to_crawl['port'] = default_proto_map[to_crawl['scheme']]
else:
# # FIXME: # TODO: remove me
try:
port = url_unpack['port'].decode()
except:
port = url_unpack['port']
# Verify port number #################### make function to verify/correct port number
try:
int(port)
# Invalid port Number
except Exception as e:
port = default_proto_map[to_crawl['scheme']]
to_crawl['port'] = port
#if url_unpack['query_string'] is None:
# if to_crawl['port'] == 80:
# to_crawl['url']= '{}://{}'.format(to_crawl['scheme'], url_unpack['host'].decode())
# else:
# to_crawl['url']= '{}://{}:{}'.format(to_crawl['scheme'], url_unpack['host'].decode(), to_crawl['port'])
#else:
# to_crawl['url']= '{}://{}:{}{}'.format(to_crawl['scheme'], url_unpack['host'].decode(), to_crawl['port'], url_unpack['query_string'].decode())
to_crawl['url'] = url
if to_crawl['port'] == 80:
to_crawl['domain_url'] = '{}://{}'.format(to_crawl['scheme'], new_url_host)
else:
to_crawl['domain_url'] = '{}://{}:{}'.format(to_crawl['scheme'], new_url_host, to_crawl['port'])
# # FIXME: # TODO: remove me
try:
to_crawl['tld'] = url_unpack['tld'].decode()
except:
to_crawl['tld'] = url_unpack['tld']
return to_crawl
# get url, paste and service_type to crawl
def get_elem_to_crawl(rotation_mode):
message = None
domain_service_type = None
#load_priority_queue
for service_type in rotation_mode:
message = redis_crawler.spop('{}_crawler_priority_queue'.format(service_type))
if message is not None:
domain_service_type = service_type
break
#load_normal_queue
if message is None:
for service_type in rotation_mode:
message = redis_crawler.spop('{}_crawler_queue'.format(service_type))
if message is not None:
domain_service_type = service_type
break
if message:
splitted = message.rsplit(';', 1)
if len(splitted) == 2:
url, paste = splitted
if paste:
paste = paste.replace(PASTES_FOLDER+'/', '')
message = {'url': url, 'paste': paste, 'type_service': domain_service_type, 'original_message': message}
return message
def get_crawler_config(redis_server, mode, service_type, domain, url=None):
crawler_options = {}
if mode=='auto':
config = redis_server.get('crawler_config:{}:{}:{}:{}'.format(mode, service_type, domain, url))
else:
config = redis_server.get('crawler_config:{}:{}:{}'.format(mode, service_type, domain))
if config is None:
config = {}
else:
config = json.loads(config)
for option in default_crawler_config:
if option in config:
crawler_options[option] = config[option]
else:
crawler_options[option] = default_crawler_config[option]
if mode == 'auto':
crawler_options['time'] = int(config['time'])
elif mode == 'manual':
redis_server.delete('crawler_config:{}:{}:{}'.format(mode, service_type, domain))
return crawler_options
def load_crawler_config(service_type, domain, paste, url, date):
crawler_config = {}
crawler_config['splash_url'] = splash_url
crawler_config['item'] = paste
crawler_config['service_type'] = service_type
crawler_config['domain'] = domain
crawler_config['date'] = date
# Auto and Manual Crawling
# Auto ################################################# create new entry, next crawling => here or when ended ?
if paste == 'auto':
crawler_config['crawler_options'] = get_crawler_config(redis_crawler, 'auto', service_type, domain, url=url)
crawler_config['requested'] = True
# Manual
elif paste == 'manual':
crawler_config['crawler_options'] = get_crawler_config(r_cache, 'manual', service_type, domain)
crawler_config['requested'] = True
# default crawler
else:
crawler_config['crawler_options'] = get_crawler_config(redis_crawler, 'default', service_type, domain)
crawler_config['requested'] = False
return crawler_config
def is_domain_up_day(domain, type_service, date_day):
if redis_crawler.sismember('{}_up:{}'.format(type_service, date_day), domain):
return True
else:
return False
def set_crawled_domain_metadata(type_service, date, domain, father_item):
# first seen
if not redis_crawler.hexists('{}_metadata:{}'.format(type_service, domain), 'first_seen'):
redis_crawler.hset('{}_metadata:{}'.format(type_service, domain), 'first_seen', date['date_day'])
redis_crawler.hset('{}_metadata:{}'.format(type_service, domain), 'paste_parent', father_item)
# last check
redis_crawler.hset('{}_metadata:{}'.format(type_service, domain), 'last_check', date['date_day'])
# Put message back on queue
def on_error_send_message_back_in_queue(type_service, domain, message):
if not redis_crawler.sismember('{}_domain_crawler_queue'.format(type_service), domain):
redis_crawler.sadd('{}_domain_crawler_queue'.format(type_service), domain)
redis_crawler.sadd('{}_crawler_priority_queue'.format(type_service), message)
def crawl_onion(url, domain, port, type_service, message, crawler_config):
crawler_config['url'] = url
crawler_config['port'] = port
print('Launching Crawler: {}'.format(url))
r_cache.hset('metadata_crawler:{}'.format(splash_port), 'crawling_domain', domain)
r_cache.hset('metadata_crawler:{}'.format(splash_port), 'started_time', datetime.datetime.now().strftime("%Y/%m/%d - %H:%M.%S"))
retry = True
nb_retry = 0
while retry:
try:
r = requests.get(splash_url , timeout=30.0)
retry = False
except Exception:
# TODO: relaunch docker or send error message
nb_retry += 1
if nb_retry == 6:
on_error_send_message_back_in_queue(type_service, domain, message)
publisher.error('{} SPASH DOWN'.format(splash_url))
print('--------------------------------------')
print(' \033[91m DOCKER SPLASH DOWN\033[0m')
print(' {} DOWN'.format(splash_url))
r_cache.hset('metadata_crawler:{}'.format(splash_port), 'status', 'SPLASH DOWN')
nb_retry == 0
print(' \033[91m DOCKER SPLASH NOT AVAILABLE\033[0m')
print(' Retry({}) in 10 seconds'.format(nb_retry))
time.sleep(10)
if r.status_code == 200:
r_cache.hset('metadata_crawler:{}'.format(splash_port), 'status', 'Crawling')
# save config in cash
UUID = str(uuid.uuid4())
r_cache.set('crawler_request:{}'.format(UUID), json.dumps(crawler_config))
process = subprocess.Popen(["python", './torcrawler/tor_crawler.py', UUID],
stdout=subprocess.PIPE)
while process.poll() is None:
time.sleep(1)
if process.returncode == 0:
output = process.stdout.read().decode()
print(output)
# error: splash:Connection to proxy refused
if 'Connection to proxy refused' in output:
on_error_send_message_back_in_queue(type_service, domain, message)
publisher.error('{} SPASH, PROXY DOWN OR BAD CONFIGURATION'.format(splash_url))
print('------------------------------------------------------------------------')
print(' \033[91m SPLASH: Connection to proxy refused')
print('')
print(' PROXY DOWN OR BAD CONFIGURATION\033[0m'.format(splash_url))
print('------------------------------------------------------------------------')
r_cache.hset('metadata_crawler:{}'.format(splash_port), 'status', 'Error')
exit(-2)
else:
print(process.stdout.read())
exit(-1)
else:
on_error_send_message_back_in_queue(type_service, domain, message)
print('--------------------------------------')
print(' \033[91m DOCKER SPLASH DOWN\033[0m')
print(' {} DOWN'.format(splash_url))
r_cache.hset('metadata_crawler:{}'.format(splash_port), 'status', 'Crawling')
exit(1)
# check external links (full_crawl)
def search_potential_source_domain(type_service, domain):
external_domains = set()
for link in redis_crawler.smembers('domain_{}_external_links:{}'.format(type_service, domain)):
# unpack url
url_data = unpack_url(link)
if url_data['domain'] != domain:
if url_data['tld'] == 'onion' or url_data['tld'] == 'i2p':
external_domains.add(url_data['domain'])
# # TODO: add special tag ?
if len(external_domains) >= 20:
redis_crawler.sadd('{}_potential_source'.format(type_service), domain)
print('New potential source found: domain')
redis_crawler.delete('domain_{}_external_links:{}'.format(type_service, domain))
if __name__ == '__main__':
if len(sys.argv) != 2:
print('usage:', 'Crawler.py', 'splash_port')
exit(1)
##################################################
#mode = sys.argv[1]
splash_port = sys.argv[1]
rotation_mode = ['onion', 'regular']
default_proto_map = {'http': 80, 'https': 443}
######################################################## add ftp ???
publisher.port = 6380
publisher.channel = "Script"
publisher.info("Script Crawler started")
config_section = 'Crawler'
# Setup the I/O queues
p = Process(config_section)
splash_url = '{}:{}'.format( p.config.get("Crawler", "splash_url_onion"), splash_port)
print('splash url: {}'.format(splash_url))
PASTES_FOLDER = os.path.join(os.environ['AIL_HOME'], p.config.get("Directories", "pastes"))
r_serv_metadata = redis.StrictRedis(
host=p.config.get("ARDB_Metadata", "host"),
port=p.config.getint("ARDB_Metadata", "port"),
db=p.config.getint("ARDB_Metadata", "db"),
decode_responses=True)
r_cache = redis.StrictRedis(
host=p.config.get("Redis_Cache", "host"),
port=p.config.getint("Redis_Cache", "port"),
db=p.config.getint("Redis_Cache", "db"),
decode_responses=True)
redis_crawler = redis.StrictRedis(
host=p.config.get("ARDB_Onion", "host"),
port=p.config.getint("ARDB_Onion", "port"),
db=p.config.getint("ARDB_Onion", "db"),
decode_responses=True)
faup = Faup()
# Default crawler options
default_crawler_config = {'html': 1,
'har': 1,
'png': 1,
'depth_limit': p.config.getint("Crawler", "crawler_depth_limit"),
'closespider_pagecount': 50,
'user_agent': 'Mozilla/5.0 (Windows NT 6.1; rv:24.0) Gecko/20100101 Firefox/24.0'}
# Track launched crawler
r_cache.sadd('all_crawler', splash_port)
r_cache.hset('metadata_crawler:{}'.format(splash_port), 'status', 'Waiting')
r_cache.hset('metadata_crawler:{}'.format(splash_port), 'started_time', datetime.datetime.now().strftime("%Y/%m/%d - %H:%M.%S"))
# update hardcoded blacklist
load_blacklist('onion')
load_blacklist('regular')
while True:
update_auto_crawler()
to_crawl = get_elem_to_crawl(rotation_mode)
if to_crawl:
url_data = unpack_url(to_crawl['url'])
# remove domain from queue
redis_crawler.srem('{}_domain_crawler_queue'.format(to_crawl['type_service']), url_data['domain'])
print()
print()
print('\033[92m------------------START CRAWLER------------------\033[0m')
print('crawler type: {}'.format(to_crawl['type_service']))
print('\033[92m-------------------------------------------------\033[0m')
print('url: {}'.format(url_data['url']))
print('domain: {}'.format(url_data['domain']))
print('domain_url: {}'.format(url_data['domain_url']))
print()
# Check blacklist
if not redis_crawler.sismember('blacklist_{}'.format(to_crawl['type_service']), url_data['domain']):
date = {'date_day': datetime.datetime.now().strftime("%Y%m%d"),
'date_month': datetime.datetime.now().strftime("%Y%m"),
'epoch': int(time.time())}
# Update crawler status type
r_cache.sadd('{}_crawlers'.format(to_crawl['type_service']), splash_port)
crawler_config = load_crawler_config(to_crawl['type_service'], url_data['domain'], to_crawl['paste'], to_crawl['url'], date)
# check if default crawler
if not crawler_config['requested']:
# Auto crawl only if service not up this month
if redis_crawler.sismember('month_{}_up:{}'.format(to_crawl['type_service'], date['date_month']), url_data['domain']):
continue
set_crawled_domain_metadata(to_crawl['type_service'], date, url_data['domain'], to_crawl['paste'])
#### CRAWLER ####
# Manual and Auto Crawler
if crawler_config['requested']:
######################################################crawler strategy
# CRAWL domain
crawl_onion(url_data['url'], url_data['domain'], url_data['port'], to_crawl['type_service'], to_crawl['original_message'], crawler_config)
# Default Crawler
else:
# CRAWL domain
crawl_onion(url_data['domain_url'], url_data['domain'], url_data['port'], to_crawl['type_service'], to_crawl['original_message'], crawler_config)
#if url != domain_url and not is_domain_up_day(url_data['domain'], to_crawl['type_service'], date['date_day']):
# crawl_onion(url_data['url'], url_data['domain'], to_crawl['original_message'])
# Save last_status day (DOWN)
if not is_domain_up_day(url_data['domain'], to_crawl['type_service'], date['date_day']):
redis_crawler.sadd('{}_down:{}'.format(to_crawl['type_service'], date['date_day']), url_data['domain'])
# if domain was UP at least one time
if redis_crawler.exists('crawler_history_{}:{}:{}'.format(to_crawl['type_service'], url_data['domain'], url_data['port'])):
# add crawler history (if domain is down)
if not redis_crawler.zrangebyscore('crawler_history_{}:{}:{}'.format(to_crawl['type_service'], url_data['domain'], url_data['port']), date['epoch'], date['epoch']):
# Domain is down
redis_crawler.zadd('crawler_history_{}:{}:{}'.format(to_crawl['type_service'], url_data['domain'], url_data['port']), int(date['epoch']), int(date['epoch']))
############################
# extract page content
############################
# update list, last crawled domains
redis_crawler.lpush('last_{}'.format(to_crawl['type_service']), '{}:{};{}'.format(url_data['domain'], url_data['port'], date['epoch']))
redis_crawler.ltrim('last_{}'.format(to_crawl['type_service']), 0, 15)
#update crawler status
r_cache.hset('metadata_crawler:{}'.format(splash_port), 'status', 'Waiting')
r_cache.hdel('metadata_crawler:{}'.format(splash_port), 'crawling_domain')
# Update crawler status type
r_cache.srem('{}_crawlers'.format(to_crawl['type_service']), splash_port)
# add next auto Crawling in queue:
if to_crawl['paste'] == 'auto':
redis_crawler.zadd('crawler_auto_queue', int(time.time()+crawler_config['crawler_options']['time']) , '{};{}'.format(to_crawl['original_message'], to_crawl['type_service']))
# update list, last auto crawled domains
redis_crawler.lpush('last_auto_crawled', '{}:{};{}'.format(url_data['domain'], url_data['port'], date['epoch']))
redis_crawler.ltrim('last_auto_crawled', 0, 9)
else:
print(' Blacklisted Domain')
print()
print()
else:
time.sleep(1)