mirror of
https://github.com/searxng/searxng.git
synced 2024-11-27 13:21:04 +00:00
265 lines
10 KiB
Python
265 lines
10 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-or-later
|
|
|
|
from urllib.parse import urlparse
|
|
from time import time
|
|
import threading
|
|
|
|
import requests.exceptions
|
|
|
|
import searx.poolrequests as poolrequests
|
|
from searx.engines import settings
|
|
from searx import logger
|
|
from searx.utils import gen_useragent
|
|
from searx.exceptions import (SearxEngineAccessDeniedException, SearxEngineCaptchaException,
|
|
SearxEngineTooManyRequestsException,)
|
|
from searx.metrology.error_recorder import record_exception, record_error
|
|
|
|
from searx.search.processors.abstract import EngineProcessor
|
|
|
|
|
|
logger = logger.getChild('search.processor.online')
|
|
|
|
|
|
def default_request_params():
|
|
return {
|
|
'method': 'GET',
|
|
'headers': {},
|
|
'data': {},
|
|
'url': '',
|
|
'cookies': {},
|
|
'verify': True,
|
|
'auth': None
|
|
}
|
|
|
|
|
|
class OnlineProcessor(EngineProcessor):
|
|
|
|
engine_type = 'online'
|
|
|
|
def get_params(self, search_query, engine_category):
|
|
params = super().get_params(search_query, engine_category)
|
|
if params is None:
|
|
return None
|
|
|
|
# skip suspended engines
|
|
if self.engine.suspend_end_time >= time():
|
|
logger.debug('Engine currently suspended: %s', self.engine_name)
|
|
return None
|
|
|
|
# add default params
|
|
params.update(default_request_params())
|
|
|
|
# add an user agent
|
|
params['headers']['User-Agent'] = gen_useragent()
|
|
|
|
return params
|
|
|
|
def _send_http_request(self, params):
|
|
# create dictionary which contain all
|
|
# informations about the request
|
|
request_args = dict(
|
|
headers=params['headers'],
|
|
cookies=params['cookies'],
|
|
verify=params['verify'],
|
|
auth=params['auth']
|
|
)
|
|
|
|
# setting engine based proxies
|
|
if hasattr(self.engine, 'proxies'):
|
|
request_args['proxies'] = poolrequests.get_proxies(self.engine.proxies)
|
|
|
|
# max_redirects
|
|
max_redirects = params.get('max_redirects')
|
|
if max_redirects:
|
|
request_args['max_redirects'] = max_redirects
|
|
|
|
# allow_redirects
|
|
if 'allow_redirects' in params:
|
|
request_args['allow_redirects'] = params['allow_redirects']
|
|
|
|
# soft_max_redirects
|
|
soft_max_redirects = params.get('soft_max_redirects', max_redirects or 0)
|
|
|
|
# raise_for_status
|
|
request_args['raise_for_httperror'] = params.get('raise_for_httperror', True)
|
|
|
|
# specific type of request (GET or POST)
|
|
if params['method'] == 'GET':
|
|
req = poolrequests.get
|
|
else:
|
|
req = poolrequests.post
|
|
|
|
request_args['data'] = params['data']
|
|
|
|
# send the request
|
|
response = req(params['url'], **request_args)
|
|
|
|
# check soft limit of the redirect count
|
|
if len(response.history) > soft_max_redirects:
|
|
# unexpected redirect : record an error
|
|
# but the engine might still return valid results.
|
|
status_code = str(response.status_code or '')
|
|
reason = response.reason or ''
|
|
hostname = str(urlparse(response.url or '').netloc)
|
|
record_error(self.engine_name,
|
|
'{} redirects, maximum: {}'.format(len(response.history), soft_max_redirects),
|
|
(status_code, reason, hostname))
|
|
|
|
return response
|
|
|
|
def _search_basic(self, query, params):
|
|
# update request parameters dependent on
|
|
# search-engine (contained in engines folder)
|
|
self.engine.request(query, params)
|
|
|
|
# ignoring empty urls
|
|
if params['url'] is None:
|
|
return None
|
|
|
|
if not params['url']:
|
|
return None
|
|
|
|
# send request
|
|
response = self._send_http_request(params)
|
|
|
|
# parse the response
|
|
response.search_params = params
|
|
return self.engine.response(response)
|
|
|
|
def search(self, query, params, result_container, start_time, timeout_limit):
|
|
# set timeout for all HTTP requests
|
|
poolrequests.set_timeout_for_thread(timeout_limit, start_time=start_time)
|
|
# reset the HTTP total time
|
|
poolrequests.reset_time_for_thread()
|
|
# enable HTTP only if explicitly enabled
|
|
poolrequests.set_enable_http_protocol(self.engine.enable_http)
|
|
|
|
# suppose everything will be alright
|
|
requests_exception = False
|
|
suspended_time = None
|
|
|
|
try:
|
|
# send requests and parse the results
|
|
search_results = self._search_basic(query, params)
|
|
|
|
# check if the engine accepted the request
|
|
if search_results is not None:
|
|
# yes, so add results
|
|
result_container.extend(self.engine_name, search_results)
|
|
|
|
# update engine time when there is no exception
|
|
engine_time = time() - start_time
|
|
page_load_time = poolrequests.get_time_for_thread()
|
|
result_container.add_timing(self.engine_name, engine_time, page_load_time)
|
|
with threading.RLock():
|
|
self.engine.stats['engine_time'] += engine_time
|
|
self.engine.stats['engine_time_count'] += 1
|
|
# update stats with the total HTTP time
|
|
self.engine.stats['page_load_time'] += page_load_time
|
|
self.engine.stats['page_load_count'] += 1
|
|
except Exception as e:
|
|
record_exception(self.engine_name, e)
|
|
|
|
# Timing
|
|
engine_time = time() - start_time
|
|
page_load_time = poolrequests.get_time_for_thread()
|
|
result_container.add_timing(self.engine_name, engine_time, page_load_time)
|
|
|
|
# Record the errors
|
|
with threading.RLock():
|
|
self.engine.stats['errors'] += 1
|
|
|
|
if (issubclass(e.__class__, requests.exceptions.Timeout)):
|
|
result_container.add_unresponsive_engine(self.engine_name, 'HTTP timeout')
|
|
# requests timeout (connect or read)
|
|
logger.error("engine {0} : HTTP requests timeout"
|
|
"(search duration : {1} s, timeout: {2} s) : {3}"
|
|
.format(self.engine_name, engine_time, timeout_limit, e.__class__.__name__))
|
|
requests_exception = True
|
|
elif (issubclass(e.__class__, requests.exceptions.RequestException)):
|
|
result_container.add_unresponsive_engine(self.engine_name, 'HTTP error')
|
|
# other requests exception
|
|
logger.exception("engine {0} : requests exception"
|
|
"(search duration : {1} s, timeout: {2} s) : {3}"
|
|
.format(self.engine_name, engine_time, timeout_limit, e))
|
|
requests_exception = True
|
|
elif (issubclass(e.__class__, SearxEngineCaptchaException)):
|
|
result_container.add_unresponsive_engine(self.engine_name, 'CAPTCHA required')
|
|
logger.exception('engine {0} : CAPTCHA'.format(self.engine_name))
|
|
suspended_time = e.suspended_time # pylint: disable=no-member
|
|
elif (issubclass(e.__class__, SearxEngineTooManyRequestsException)):
|
|
result_container.add_unresponsive_engine(self.engine_name, 'too many requests')
|
|
logger.exception('engine {0} : Too many requests'.format(self.engine_name))
|
|
suspended_time = e.suspended_time # pylint: disable=no-member
|
|
elif (issubclass(e.__class__, SearxEngineAccessDeniedException)):
|
|
result_container.add_unresponsive_engine(self.engine_name, 'blocked')
|
|
logger.exception('engine {0} : Searx is blocked'.format(self.engine_name))
|
|
suspended_time = e.suspended_time # pylint: disable=no-member
|
|
else:
|
|
result_container.add_unresponsive_engine(self.engine_name, 'unexpected crash')
|
|
# others errors
|
|
logger.exception('engine {0} : exception : {1}'.format(self.engine_name, e))
|
|
else:
|
|
if getattr(threading.current_thread(), '_timeout', False):
|
|
record_error(self.engine_name, 'Timeout')
|
|
|
|
# suspend the engine if there is an HTTP error
|
|
# or suspended_time is defined
|
|
with threading.RLock():
|
|
if requests_exception or suspended_time:
|
|
# update continuous_errors / suspend_end_time
|
|
self.engine.continuous_errors += 1
|
|
if suspended_time is None:
|
|
suspended_time = min(settings['search']['max_ban_time_on_fail'],
|
|
self.engine.continuous_errors * settings['search']['ban_time_on_fail'])
|
|
self.engine.suspend_end_time = time() + suspended_time
|
|
else:
|
|
# reset the suspend variables
|
|
self.engine.continuous_errors = 0
|
|
self.engine.suspend_end_time = 0
|
|
|
|
def get_default_tests(self):
|
|
tests = {}
|
|
|
|
tests['simple'] = {
|
|
'matrix': {'query': ('life', 'computer')},
|
|
'result_container': ['not_empty'],
|
|
}
|
|
|
|
if getattr(self.engine, 'paging', False):
|
|
tests['paging'] = {
|
|
'matrix': {'query': 'time',
|
|
'pageno': (1, 2, 3)},
|
|
'result_container': ['not_empty'],
|
|
'test': ['unique_results']
|
|
}
|
|
if 'general' in self.engine.categories:
|
|
# avoid documentation about HTML tags (<time> and <input type="time">)
|
|
tests['paging']['matrix']['query'] = 'news'
|
|
|
|
if getattr(self.engine, 'time_range', False):
|
|
tests['time_range'] = {
|
|
'matrix': {'query': 'news',
|
|
'time_range': (None, 'day')},
|
|
'result_container': ['not_empty'],
|
|
'test': ['unique_results']
|
|
}
|
|
|
|
if getattr(self.engine, 'supported_languages', []):
|
|
tests['lang_fr'] = {
|
|
'matrix': {'query': 'paris', 'lang': 'fr'},
|
|
'result_container': ['not_empty', ('has_language', 'fr')],
|
|
}
|
|
tests['lang_en'] = {
|
|
'matrix': {'query': 'paris', 'lang': 'en'},
|
|
'result_container': ['not_empty', ('has_language', 'en')],
|
|
}
|
|
|
|
if getattr(self.engine, 'safesearch', False):
|
|
tests['safesearch'] = {
|
|
'matrix': {'query': 'porn',
|
|
'safesearch': (0, 2)},
|
|
'test': ['unique_results']
|
|
}
|
|
|
|
return tests
|