[enh][mod] result handling refactor

Several changes has been made:
 - Parallel result merge
 - Scoring algorithm slightly changed (see result_score())
 - Proper Thread locking on global data manipulation
This commit is contained in:
Adam Tauber 2015-10-03 17:26:07 +02:00
parent 0ad272c5cb
commit b6c3cb0bdd
6 changed files with 321 additions and 292 deletions

239
searx/results.py Normal file
View file

@ -0,0 +1,239 @@
import re
from collections import defaultdict
from operator import itemgetter
from threading import RLock
from urlparse import urlparse, unquote
from searx.engines import engines
CONTENT_LEN_IGNORED_CHARS_REGEX = re.compile('[,;:!?\./\\\\ ()-_]', re.M | re.U)
WHITESPACE_REGEX = re.compile('( |\t|\n)+', re.M | re.U)
# return the meaningful length of the content for a result
def result_content_len(content):
if isinstance(content, basestring):
return len(CONTENT_LEN_IGNORED_CHARS_REGEX.sub('', content))
else:
return 0
def compare_urls(url_a, url_b):
if url_a.netloc != url_b.netloc or url_a.query != url_b.query:
return False
# remove / from the end of the url if required
path_a = url_a.path[:-1]\
if url_a.path.endswith('/')\
else url_a.path
path_b = url_b.path[:-1]\
if url_b.path.endswith('/')\
else url_b.path
return unquote(path_a) == unquote(path_b)
def merge_two_infoboxes(infobox1, infobox2):
if 'urls' in infobox2:
urls1 = infobox1.get('urls', None)
if urls1 is None:
urls1 = []
infobox1.set('urls', urls1)
urlSet = set()
for url in infobox1.get('urls', []):
urlSet.add(url.get('url', None))
for url in infobox2.get('urls', []):
if url.get('url', None) not in urlSet:
urls1.append(url)
if 'attributes' in infobox2:
attributes1 = infobox1.get('attributes', None)
if attributes1 is None:
attributes1 = []
infobox1.set('attributes', attributes1)
attributeSet = set()
for attribute in infobox1.get('attributes', []):
if attribute.get('label', None) not in attributeSet:
attributeSet.add(attribute.get('label', None))
for attribute in infobox2.get('attributes', []):
attributes1.append(attribute)
if 'content' in infobox2:
content1 = infobox1.get('content', None)
content2 = infobox2.get('content', '')
if content1 is not None:
if result_content_len(content2) > result_content_len(content1):
infobox1['content'] = content2
else:
infobox1.set('content', content2)
def result_score(result):
weight = 1.0
for result_engine in result['engines']:
if hasattr(engines[result_engine], 'weight'):
weight *= float(engines[result_engine].weight)
occurences = len(result['positions'])
return sum((occurences * weight) / position for position in result['positions'])
class ResultContainer(object):
"""docstring for ResultContainer"""
def __init__(self):
super(ResultContainer, self).__init__()
self.results = defaultdict(list)
self._merged_results = []
self.infoboxes = []
self._infobox_ids = {}
self.suggestions = set()
self.answers = set()
def extend(self, engine_name, results):
for result in list(results):
if 'suggestion' in result:
self.suggestions.add(result['suggestion'])
results.remove(result)
elif 'answer' in result:
self.answers.add(result['suggestion'])
results.remve(result)
elif 'infobox' in result:
self._merge_infobox(result)
results.remove(result)
with RLock():
engines[engine_name].stats['search_count'] += 1
engines[engine_name].stats['result_count'] += len(results)
if not results:
return
self.results[engine_name].extend(results)
for i, result in enumerate(results):
position = i + 1
self._merge_result(result, position)
def _merge_infobox(self, infobox):
add_infobox = True
infobox_id = infobox.get('id', None)
if infobox_id is not None:
existingIndex = self._infobox_ids.get(infobox_id, None)
if existingIndex is not None:
merge_two_infoboxes(self.infoboxes[existingIndex], infobox)
add_infobox = False
if add_infobox:
self.infoboxes.append(infobox)
self._infobox_ids[infobox_id] = len(self.infoboxes) - 1
def _merge_result(self, result, position):
result['parsed_url'] = urlparse(result['url'])
# if the result has no scheme, use http as default
if not result['parsed_url'].scheme:
result['parsed_url'] = result['parsed_url']._replace(scheme="http")
result['host'] = result['parsed_url'].netloc
if result['host'].startswith('www.'):
result['host'] = result['host'].replace('www.', '', 1)
result['engines'] = [result['engine']]
# strip multiple spaces and cariage returns from content
if result.get('content'):
result['content'] = WHITESPACE_REGEX.sub(' ', result['content'])
# check for duplicates
duplicated = False
for merged_result in self._merged_results:
if compare_urls(result['parsed_url'], merged_result['parsed_url'])\
and result.get('template') == merged_result.get('template'):
duplicated = merged_result
break
# merge duplicates together
if duplicated:
# using content with more text
if result_content_len(result.get('content', '')) >\
result_content_len(duplicated.get('content', '')):
duplicated['content'] = result['content']
# add the new position
duplicated['positions'].append(position)
# add engine to list of result-engines
duplicated['engines'].append(result['engine'])
# using https if possible
if duplicated['parsed_url'].scheme != 'https' and result['parsed_url'].scheme == 'https':
duplicated['url'] = result['parsed_url'].geturl()
duplicated['parsed_url'] = result['parsed_url']
# if there is no duplicate found, append result
else:
result['positions'] = [position]
with RLock():
self._merged_results.append(result)
def get_ordered_results(self):
for result in self._merged_results:
score = result_score(result)
result['score'] = score
with RLock():
for result_engine in result['engines']:
engines[result_engine].stats['score_count'] += score
results = sorted(self._merged_results, key=itemgetter('score'), reverse=True)
# pass 2 : group results by category and template
gresults = []
categoryPositions = {}
for i, res in enumerate(results):
# FIXME : handle more than one category per engine
category = engines[res['engine']].categories[0] + ':' + ''\
if 'template' not in res\
else res['template']
current = None if category not in categoryPositions\
else categoryPositions[category]
# group with previous results using the same category
# if the group can accept more result and is not too far
# from the current position
if current is not None and (current['count'] > 0)\
and (len(gresults) - current['index'] < 20):
# group with the previous results using
# the same category with this one
index = current['index']
gresults.insert(index, res)
# update every index after the current one
# (including the current one)
for k in categoryPositions:
v = categoryPositions[k]['index']
if v >= index:
categoryPositions[k]['index'] = v + 1
# update this category
current['count'] -= 1
else:
# same category
gresults.append(res)
# update categoryIndex
categoryPositions[category] = {'index': len(gresults), 'count': 8}
# return gresults
return gresults
def results_length(self):
return len(self._merged_results)

View file

@ -16,13 +16,8 @@ along with searx. If not, see < http://www.gnu.org/licenses/ >.
'''
import threading
import re
import searx.poolrequests as requests_lib
from itertools import izip_longest, chain
from operator import itemgetter
from Queue import Queue
from time import time
from urlparse import urlparse, unquote
from searx import settings
from searx.engines import (
categories, engines
@ -30,6 +25,7 @@ from searx.engines import (
from searx.languages import language_codes
from searx.utils import gen_useragent, get_blocked_engines
from searx.query import Query
from searx.results import ResultContainer
from searx import logger
logger = logger.getChild('search')
@ -42,6 +38,7 @@ def search_request_wrapper(fn, url, engine_name, **kwargs):
return fn(url, **kwargs)
except:
# increase errors stats
with threading.RLock():
engines[engine_name].stats['errors'] += 1
# print engine name and specific error message
@ -84,7 +81,7 @@ def default_request_params():
# create a callback wrapper for the search engine results
def make_callback(engine_name, results_queue, callback, params):
def make_callback(engine_name, callback, params, result_container):
# creating a callback wrapper for the search engine results
def process_callback(response, **kwargs):
@ -96,11 +93,16 @@ def make_callback(engine_name, results_queue, callback, params):
response.search_params = params
timeout_overhead = 0.2 # seconds
search_duration = time() - params['started']
# update stats with current page-load-time
with threading.RLock():
engines[engine_name].stats['page_load_time'] += search_duration
timeout_overhead = 0.2 # seconds
timeout_limit = engines[engine_name].timeout + timeout_overhead
if search_duration > timeout_limit:
engines[engine_name].stats['page_load_time'] += timeout_limit
with threading.RLock():
engines[engine_name].stats['errors'] += 1
return
@ -111,212 +113,11 @@ def make_callback(engine_name, results_queue, callback, params):
for result in search_results:
result['engine'] = engine_name
results_queue.put_nowait((engine_name, search_results))
# update stats with current page-load-time
engines[engine_name].stats['page_load_time'] += search_duration
result_container.extend(engine_name, search_results)
return process_callback
# return the meaningful length of the content for a result
def content_result_len(content):
if isinstance(content, basestring):
content = re.sub('[,;:!?\./\\\\ ()-_]', '', content)
return len(content)
else:
return 0
# score results and remove duplications
def score_results(results):
# calculate scoring parameters
flat_res = filter(
None, chain.from_iterable(izip_longest(*results.values())))
flat_len = len(flat_res)
engines_len = len(results)
results = []
# pass 1: deduplication + scoring
for i, res in enumerate(flat_res):
res['parsed_url'] = urlparse(res['url'])
# if the result has no scheme, use http as default
if not res['parsed_url'].scheme:
res['parsed_url'] = res['parsed_url']._replace(scheme="http")
res['host'] = res['parsed_url'].netloc
if res['host'].startswith('www.'):
res['host'] = res['host'].replace('www.', '', 1)
res['engines'] = [res['engine']]
weight = 1.0
# strip multiple spaces and cariage returns from content
if res.get('content'):
res['content'] = re.sub(' +', ' ',
res['content'].strip().replace('\n', ''))
# get weight of this engine if possible
if hasattr(engines[res['engine']], 'weight'):
weight = float(engines[res['engine']].weight)
# calculate score for that engine
score = int((flat_len - i) / engines_len) * weight + 1
# check for duplicates
duplicated = False
for new_res in results:
# remove / from the end of the url if required
p1 = res['parsed_url'].path[:-1]\
if res['parsed_url'].path.endswith('/')\
else res['parsed_url'].path
p2 = new_res['parsed_url'].path[:-1]\
if new_res['parsed_url'].path.endswith('/')\
else new_res['parsed_url'].path
# check if that result is a duplicate
if res['host'] == new_res['host'] and\
unquote(p1) == unquote(p2) and\
res['parsed_url'].query == new_res['parsed_url'].query and\
res.get('template') == new_res.get('template'):
duplicated = new_res
break
# merge duplicates together
if duplicated:
# using content with more text
if content_result_len(res.get('content', '')) >\
content_result_len(duplicated.get('content', '')):
duplicated['content'] = res['content']
# increase result-score
duplicated['score'] += score
# add engine to list of result-engines
duplicated['engines'].append(res['engine'])
# using https if possible
if duplicated['parsed_url'].scheme == 'https':
continue
elif res['parsed_url'].scheme == 'https':
duplicated['url'] = res['parsed_url'].geturl()
duplicated['parsed_url'] = res['parsed_url']
# if there is no duplicate found, append result
else:
res['score'] = score
results.append(res)
results = sorted(results, key=itemgetter('score'), reverse=True)
# pass 2 : group results by category and template
gresults = []
categoryPositions = {}
for i, res in enumerate(results):
# FIXME : handle more than one category per engine
category = engines[res['engine']].categories[0] + ':' + ''\
if 'template' not in res\
else res['template']
current = None if category not in categoryPositions\
else categoryPositions[category]
# group with previous results using the same category
# if the group can accept more result and is not too far
# from the current position
if current is not None and (current['count'] > 0)\
and (len(gresults) - current['index'] < 20):
# group with the previous results using
# the same category with this one
index = current['index']
gresults.insert(index, res)
# update every index after the current one
# (including the current one)
for k in categoryPositions:
v = categoryPositions[k]['index']
if v >= index:
categoryPositions[k]['index'] = v + 1
# update this category
current['count'] -= 1
else:
# same category
gresults.append(res)
# update categoryIndex
categoryPositions[category] = {'index': len(gresults), 'count': 8}
# return gresults
return gresults
def merge_two_infoboxes(infobox1, infobox2):
if 'urls' in infobox2:
urls1 = infobox1.get('urls', None)
if urls1 is None:
urls1 = []
infobox1.set('urls', urls1)
urlSet = set()
for url in infobox1.get('urls', []):
urlSet.add(url.get('url', None))
for url in infobox2.get('urls', []):
if url.get('url', None) not in urlSet:
urls1.append(url)
if 'attributes' in infobox2:
attributes1 = infobox1.get('attributes', None)
if attributes1 is None:
attributes1 = []
infobox1.set('attributes', attributes1)
attributeSet = set()
for attribute in infobox1.get('attributes', []):
if attribute.get('label', None) not in attributeSet:
attributeSet.add(attribute.get('label', None))
for attribute in infobox2.get('attributes', []):
attributes1.append(attribute)
if 'content' in infobox2:
content1 = infobox1.get('content', None)
content2 = infobox2.get('content', '')
if content1 is not None:
if content_result_len(content2) > content_result_len(content1):
infobox1['content'] = content2
else:
infobox1.set('content', content2)
def merge_infoboxes(infoboxes):
results = []
infoboxes_id = {}
for infobox in infoboxes:
add_infobox = True
infobox_id = infobox.get('id', None)
if infobox_id is not None:
existingIndex = infoboxes_id.get(infobox_id, None)
if existingIndex is not None:
merge_two_infoboxes(results[existingIndex], infobox)
add_infobox = False
if add_infobox:
results.append(infobox)
infoboxes_id[infobox_id] = len(results) - 1
return results
class Search(object):
"""Search information container"""
@ -334,10 +135,7 @@ class Search(object):
# set blocked engines
self.blocked_engines = get_blocked_engines(engines, request.cookies)
self.results = []
self.suggestions = set()
self.answers = set()
self.infoboxes = []
self.result_container = ResultContainer()
self.request_data = {}
# set specific language if set
@ -449,8 +247,6 @@ class Search(object):
# init vars
requests = []
results_queue = Queue()
results = {}
# increase number of searches
number_of_searches += 1
@ -504,9 +300,9 @@ class Search(object):
# create a callback wrapper for the search engine results
callback = make_callback(
selected_engine['name'],
results_queue,
engine.response,
request_params)
request_params,
self.result_container)
# create dictionary which contain all
# informations about the request
@ -539,42 +335,5 @@ class Search(object):
# send all search-request
threaded_requests(requests)
while not results_queue.empty():
engine_name, engine_results = results_queue.get_nowait()
# TODO type checks
[self.suggestions.add(x['suggestion'])
for x in list(engine_results)
if 'suggestion' in x
and engine_results.remove(x) is None]
[self.answers.add(x['answer'])
for x in list(engine_results)
if 'answer' in x
and engine_results.remove(x) is None]
self.infoboxes.extend(x for x in list(engine_results)
if 'infobox' in x
and engine_results.remove(x) is None)
results[engine_name] = engine_results
# update engine-specific stats
for engine_name, engine_results in results.items():
engines[engine_name].stats['search_count'] += 1
engines[engine_name].stats['result_count'] += len(engine_results)
# score results and remove duplications
self.results = score_results(results)
# merge infoboxes according to their ids
self.infoboxes = merge_infoboxes(self.infoboxes)
# update engine stats, using calculated score
for result in self.results:
for res_engine in result['engines']:
engines[result['engine']]\
.stats['score_count'] += result['score']
# return results, suggestions, answers and infoboxes
return self

View file

@ -0,0 +1,41 @@
# -*- coding: utf-8 -*-
from searx.results import ResultContainer
from searx.testing import SearxTestCase
def fake_result(url='https://aa.bb/cc?dd=ee#ff',
title='aaa',
content='bbb',
engine='wikipedia', **kwargs):
result = {'url': url,
'title': title,
'content': content,
'engine': engine}
result.update(kwargs)
return result
# TODO
class ResultContainerTestCase(SearxTestCase):
def test_empty(self):
c = ResultContainer()
self.assertEqual(c.get_ordered_results(), [])
def test_one_result(self):
c = ResultContainer()
c.extend('wikipedia', [fake_result()])
self.assertEqual(c.results_length(), 1)
def test_one_suggestion(self):
c = ResultContainer()
c.extend('wikipedia', [fake_result(suggestion=True)])
self.assertEqual(len(c.suggestions), 1)
self.assertEqual(c.results_length(), 0)
def test_result_merge(self):
c = ResultContainer()
c.extend('wikipedia', [fake_result()])
c.extend('wikidata', [fake_result(), fake_result(url='https://example.com/')])
self.assertEqual(c.results_length(), 2)

View file

@ -1,25 +1,10 @@
# -*- coding: utf-8 -*-
from searx.search import score_results
from searx.testing import SearxTestCase
def fake_result(url='https://aa.bb/cc?dd=ee#ff',
title='aaa',
content='bbb',
engine='wikipedia'):
return {'url': url,
'title': title,
'content': content,
'engine': engine}
# TODO
class SearchTestCase(SearxTestCase):
class ScoreResultsTestCase(SearxTestCase):
def test_empty(self):
self.assertEqual(score_results(dict()), [])
def test_urlparse(self):
results = score_results(dict(a=[fake_result(url='https://aa.bb/cc?dd=ee#ff')]))
parsed_url = results[0]['parsed_url']
self.assertEqual(parsed_url.query, 'dd=ee')
def test_(self):
pass

View file

@ -1,6 +1,7 @@
# -*- coding: utf-8 -*-
import json
from mock import Mock
from urlparse import ParseResult
from searx import webapp
from searx.testing import SearxTestCase
@ -33,7 +34,12 @@ class ViewsTestCase(SearxTestCase):
]
def search_mock(search_self, *args):
search_self.results = self.test_results
search_self.result_container = Mock(get_ordered_results=lambda: self.test_results,
answers=set(),
suggestions=set(),
infoboxes=[],
results=self.test_results,
results_length=lambda: len(self.test_results))
webapp.Search.search = search_mock

View file

@ -383,7 +383,7 @@ def index():
plugins.call('post_search', request, locals())
for result in search.results:
for result in search.result_container.get_ordered_results():
plugins.call('on_result', request, locals())
if not search.paging and engines[result['engine']].paging:
@ -411,7 +411,7 @@ def index():
minutes = int((timedifference.seconds / 60) % 60)
hours = int(timedifference.seconds / 60 / 60)
if hours == 0:
result['publishedDate'] = gettext(u'{minutes} minute(s) ago').format(minutes=minutes) # noqa
result['publishedDate'] = gettext(u'{minutes} minute(s) ago').format(minutes=minutes)
else:
result['publishedDate'] = gettext(u'{hours} hour(s), {minutes} minute(s) ago').format(hours=hours, minutes=minutes) # noqa
else:
@ -419,14 +419,13 @@ def index():
if search.request_data.get('format') == 'json':
return Response(json.dumps({'query': search.query,
'results': search.results}),
'results': search.result_container.get_ordered_results()}),
mimetype='application/json')
elif search.request_data.get('format') == 'csv':
csv = UnicodeWriter(cStringIO.StringIO())
keys = ('title', 'url', 'content', 'host', 'engine', 'score')
if search.results:
csv.writerow(keys)
for row in search.results:
for row in search.result_container.get_ordered_results():
row['host'] = row['parsed_url'].netloc
csv.writerow([row.get(key, '') for key in keys])
csv.stream.seek(0)
@ -437,24 +436,24 @@ def index():
elif search.request_data.get('format') == 'rss':
response_rss = render(
'opensearch_response_rss.xml',
results=search.results,
results=search.result_container.get_ordered_results(),
q=search.request_data['q'],
number_of_results=len(search.results),
number_of_results=search.result_container.results_length(),
base_url=get_base_url()
)
return Response(response_rss, mimetype='text/xml')
return render(
'results.html',
results=search.results,
results=search.result_container.get_ordered_results(),
q=search.request_data['q'],
selected_categories=search.categories,
paging=search.paging,
pageno=search.pageno,
base_url=get_base_url(),
suggestions=search.suggestions,
answers=search.answers,
infoboxes=search.infoboxes,
suggestions=search.result_container.suggestions,
answers=search.result_container.answers,
infoboxes=search.result_container.infoboxes,
theme=get_current_theme_name(),
favicons=global_favicons[themes.index(get_current_theme_name())]
)