lookyloo/bin/background_indexer.py

110 lines
4.3 KiB
Python
Executable File

#!/usr/bin/env python3
from __future__ import annotations
import logging
import logging.config
from redis import Redis
from typing import Generator
from lookyloo import Lookyloo, Indexing
from lookyloo.capturecache import get_pickle_path
from lookyloo.default import AbstractManager, get_config, get_socket_path
from lookyloo.exceptions import NoValidHarFile
logging.config.dictConfig(get_config('logging'))
class BackgroundIndexer(AbstractManager):
def __init__(self, full: bool=False, loglevel: int | None=None):
super().__init__(loglevel)
self.lookyloo = Lookyloo(cache_max_size=1)
self.is_public_instance = get_config('generic', 'public_instance')
self.full_indexer = full
self.indexing = Indexing(full_index=self.full_indexer)
if self.full_indexer:
self.script_name = 'background_full_indexer'
else:
self.script_name = 'background_indexer'
# Redis connector so we don't use the one from Lookyloo
self.redis = Redis(unix_socket_path=get_socket_path('cache'), decode_responses=True)
def _to_run_forever(self) -> None:
self._check_indexes()
# Don't need the cache in this class.
self.lookyloo.clear_tree_cache()
def _to_index_no_cache(self) -> Generator[tuple[tuple[bool, bool, bool, bool, bool], str], None, None]:
# NOTE: only get the non-archived captures for now.
for uuid, directory in self.redis.hscan_iter('lookup_dirs'):
if not self.full_indexer:
# If we're not running the full indexer, check if the capture should be indexed.
if self.is_public_instance and self.redis.hexists(directory, 'no_index'):
# Capture unindexed
continue
if get_pickle_path(directory) is None:
# pickle isn't ready, we can't index.
continue
indexed = self.indexing.capture_indexed(uuid)
if all(indexed):
continue
yield indexed, uuid
def _check_indexes(self) -> None:
if not self.indexing.can_index:
# There is no reason to run this method in multiple scripts.
self.logger.info('Indexing already ongoing in another process.')
return None
self.logger.info(f'Check {self.script_name}...')
for indexed, uuid_to_index in self._to_index_no_cache():
try:
ct = self.lookyloo.get_crawled_tree(uuid_to_index)
except NoValidHarFile:
self.logger.warning(f'Broken pickle for {uuid_to_index}')
self.lookyloo.remove_pickle(uuid_to_index)
continue
if not indexed[0]:
self.logger.info(f'Indexing urls for {uuid_to_index}')
self.indexing.index_url_capture(ct)
if not indexed[1]:
self.logger.info(f'Indexing resources for {uuid_to_index}')
self.indexing.index_body_hashes_capture(ct)
if not indexed[2]:
self.logger.info(f'Indexing cookies for {uuid_to_index}')
self.indexing.index_cookies_capture(ct)
if not indexed[3]:
self.logger.info(f'Indexing HH Hashes for {uuid_to_index}')
self.indexing.index_http_headers_hashes_capture(ct)
if not indexed[4]:
self.logger.info(f'Indexing favicons for {uuid_to_index}')
favicons = self.lookyloo.get_potential_favicons(uuid_to_index, all_favicons=True, for_datauri=False)
self.indexing.index_favicons_capture(uuid_to_index, favicons)
# NOTE: categories aren't taken in account here, should be fixed(?)
# see indexing.index_categories_capture(capture_uuid, categories)
self.indexing.indexing_done()
self.logger.info('... done.')
def main() -> None:
i = BackgroundIndexer()
i.run(sleep_in_sec=60)
def main_full_indexer() -> None:
if not get_config('generic', 'index_everything'):
raise Exception('Full indexer is disabled.')
# NOTE: for now, it only indexes the captures that aren't archived.
# we will change that later, but for now, it's a good start.
i = BackgroundIndexer(full=True)
i.run(sleep_in_sec=60)
if __name__ == '__main__':
main()