X-Git-Url: https://git.mdrn.pl/wolnelektury.git/blobdiff_plain/f109b7f6e35d54281f1edf4b684c49db621d36c5..9e65086e2b3ab261138a95bfe8b2a17947c630dd:/apps/search/index.py?ds=sidebyside diff --git a/apps/search/index.py b/apps/search/index.py index 36f42478d..2ae909a84 100644 --- a/apps/search/index.py +++ b/apps/search/index.py @@ -1,4 +1,5 @@ # -*- coding: utf-8 -*- + from django.conf import settings from lucene import SimpleFSDirectory, IndexWriter, File, Field, \ NumericField, Version, Document, JavaError, IndexSearcher, \ @@ -8,8 +9,10 @@ from lucene import SimpleFSDirectory, IndexWriter, File, Field, \ BlockJoinQuery, BlockJoinCollector, TermsFilter, \ HashSet, BooleanClause, Term, CharTermAttribute, \ PhraseQuery, StringReader, TermQuery, BlockJoinQuery, \ - Sort, Integer + Sort, Integer, \ + initVM, CLASSPATH # KeywordAnalyzer +JVM = initVM(CLASSPATH) import sys import os import errno @@ -19,7 +22,7 @@ import catalogue.models from multiprocessing.pool import ThreadPool from threading import current_thread import atexit - +import traceback class WLAnalyzer(PerFieldAnalyzerWrapper): def __init__(self): @@ -85,7 +88,6 @@ class Index(IndexStore): if overwrite: self.remove_book(book) - doc = self.extract_metadata(book) parts = self.extract_content(book) block = ArrayList().of_(Document) @@ -245,6 +247,17 @@ class Index(IndexStore): self.close() +def log_exception_wrapper(f): + def _wrap(*a): + try: + f(*a) + except Exception, e: + print("Error in indexing thread: %s" % e) + traceback.print_exc() + raise e + return _wrap + + class ReusableIndex(Index): """ Works like index, but does not close/optimize Lucene index @@ -262,14 +275,14 @@ class ReusableIndex(Index): self.index = ReusableIndex.index else: print("opening index") - ReusableIndex.pool = ThreadPool(threads) + ReusableIndex.pool = ThreadPool(threads, initializer=lambda: JVM.attachCurrentThread() ) ReusableIndex.pool_jobs = [] Index.open(self, analyzer) ReusableIndex.index = self.index atexit.register(ReusableIndex.close_reusable) def index_book(self, *args, **kw): - job = ReusableIndex.pool.apply_async(Index.index_book, (self,)+ args, kw) + job = ReusableIndex.pool.apply_async(log_exception_wrapper(Index.index_book), (self,) + args, kw) ReusableIndex.pool_jobs.append(job) @staticmethod @@ -277,7 +290,9 @@ class ReusableIndex(Index): if ReusableIndex.index is not None: print("closing index") for job in ReusableIndex.pool_jobs: - job.wait() + job.get() + sys.stdout.write('.') + sys.stdout.flush() ReusableIndex.pool.close() ReusableIndex.index.optimize() @@ -370,6 +385,76 @@ class Search(IndexStore): # } +class SearchResult(object): + def __init__(self, searcher, scoreDocs, score=None): + if score: + self.score = score + else: + self.score = scoreDocs.score + + self.fragments = [] + self.scores = {} + self.sections = [] + + stored = searcher.doc(scoreDocs.doc) + self.book_id = int(stored.get("book_id")) + + fragment = stored.get("fragment_anchor") + if fragment: + self.fragments.append(fragment) + self.scores[fragment] = scoreDocs.score + + header_type = stored.get("header_type") + if header_type: + sec = (header_type, int(stored.get("header_index"))) + self.sections.append(sec) + self.scores[sec] = scoreDocs.score + + def get_book(self): + return catalogue.models.Book.objects.get(id=self.book_id) + + book = property(get_book) + + def get_parts(self): + book = self.book + parts = [{"header": s[0], "position": s[1], '_score_key': s} for s in self.sections] \ + + [{"fragment": book.fragments.get(anchor=f), '_score_key':f} for f in self.fragments] + + parts.sort(lambda a, b: cmp(self.scores[a['_score_key']], self.scores[b['_score_key']])) + print("bookid: %d parts: %s" % (self.book_id, parts)) + return parts + + parts = property(get_parts) + + def merge(self, other): + if self.book_id != other.book_id: + raise ValueError("this search result is or book %d; tried to merge with %d" % (self.book_id, other.book_id)) + self.fragments += other.fragments + self.sections += other.sections + self.scores.update(other.scores) + if other.score > self.score: + self.score = other.score + return self + + def __unicode__(self): + return u'SearchResult(book_id=%d, score=%d)' % (self.book_id, self.score) + + @staticmethod + def aggregate(*result_lists): + books = {} + for rl in result_lists: + for r in rl: + if r.book_id in books: + books[r.book_id].merge(r) + #print(u"already have one with score %f, and this one has score %f" % (books[book.id][0], found.score)) + else: + books[r.book_id] = r + return books.values() + + def __cmp__(self, other): + return cmp(self.score, other.score) + + class MultiSearch(Search): """Class capable of IMDb-like searching""" def get_tokens(self, queryreader): @@ -402,15 +487,25 @@ class MultiSearch(Search): return BlockJoinQuery(query, self.parent_filter, BlockJoinQuery.ScoreMode.Total) - def search_perfect(self, tokens, max_results=20): - qrys = [self.make_phrase(tokens, field=fld) for fld in ['author', 'title', 'content']] + def search_perfect_book(self, tokens, max_results=20): + qrys = [self.make_phrase(tokens, field=fld) for fld in ['author', 'title']] books = [] for q in qrys: top = self.searcher.search(q, max_results) for found in top.scoreDocs: - book_info = self.searcher.doc(found.doc) - books.append((found.score, catalogue.models.Book.objects.get(id=book_info.get("book_id")), [])) + books.append(SearchResult(self.searcher, found)) + return books + + def search_perfect_parts(self, tokens, max_results=20): + qrys = [self.make_phrase(tokens, field=fld) for fld in ['content']] + + books = [] + for q in qrys: + top = self.searcher.search(q, max_results) + for found in top.scoreDocs: + books.append(SearchResult(self.searcher, found)) + return books def search_everywhere(self, tokens, max_results=20): @@ -437,24 +532,10 @@ class MultiSearch(Search): top_groups = collector.getTopGroups(in_content_join, Sort.RELEVANCE, 0, max_results, 0, True) if top_groups: for grp in top_groups.groups: - doc_id = Integer.cast_(grp.groupValue).intValue() - book_data = self.searcher.doc(doc_id) - book = catalogue.models.Book.objects.get(id=book_data.get("book_id")) - parts = [] for part in grp.scoreDocs: - part_data = self.searcher.doc(part.doc) - header_type = part_data.get("header_type") - if header_type: - parts.append((part.score, {"header": header_type, "position": int(part_data.get("header_index"))})) - fragment = part_data.get("fragment_anchor") - if fragment: - fragment = book.fragments.get(anchor=fragment) - parts.append((part.score, {"fragment": fragment})) - books.append((grp.maxScore, book, parts)) - + books.append(SearchResult(self.searcher, part, score=grp.maxScore)) return books - def multisearch(self, query, max_results=50): """ Search strategy: