2011-12-14 80 views
3

我正在使用Lucene.net來索引產品目錄。我使用ANTS Profiler來分析我的搜索結果,我注意到使用MultiFieldQueryParser創建和解析查詢的行爲幾乎與實際搜索(約100ms)一樣長。然後我嘗試手動創建查詢,這發生得非常快(約1ms)。我寧願不必手動解析,雖然它確實給了我相同的結果集,但我擔心我可能無法處理某些用例或輸入(儘管輸入來自網站上的文本搜索,用戶將不知道關於Lucene搜索語法的任何內容)。我的代碼(這兩種方法)如下:爲什麼MultiFieldQueryParser比手工創建查詢慢得多?

 IApplicationSettings settings = new ApplicationSettingService(); 
     FSDirectory directory = FSDirectory.Open(new DirectoryInfo(settings.GetSetting<string>("LuceneMainSearchDirectory"))); 
     RAMDirectory ramDir = new RAMDirectory(directory); 
     _Searcher = new IndexSearcher(ramDir, true);   
     string[] searchFields = new string[] { "ProductName", "ProductLongDescription", "BrandName", "CategoryName" }; 

     //Add a wildcard character to end of search to give broader results 
     if (!searchTerm.EndsWith(" ")) { searchTerm = searchTerm + "*"; } 


     //Use query parser...this block typically takes about 100ms on my machine, roughly 40% on the constructor and 60% on the call to Parse 
     MultiFieldQueryParser multiParser = new MultiFieldQueryParser(Lucene.Net.Util.Version.LUCENE_29, searchFields, _analyzer); 
     multiParser.SetDefaultOperator(QueryParser.AND_OPERATOR); 
     Query query = multiParser.Parse(searchTerm); 



     //Manually create query....this block doesn't even take 1ms on my machine 
     BooleanQuery booleanQuery = new BooleanQuery(true); 
     var terms = searchTerm.Split(' '); 
     foreach (string s in terms) 
     { 
      BooleanQuery subQuery = new BooleanQuery(true); 
      if (!s.EndsWith("*")) 
      { 
       Query query1 = new TermQuery(new Term("ProductName", s)); 
       Query query2 = new TermQuery(new Term("ProductLongDescription", s)); 
       Query query3 = new TermQuery(new Term("BrandName", s)); 
       Query query4 = new TermQuery(new Term("CategoryName", s)); 
       subQuery.Add(query1, BooleanClause.Occur.SHOULD); 
       subQuery.Add(query2, BooleanClause.Occur.SHOULD); 
       subQuery.Add(query3, BooleanClause.Occur.SHOULD); 
       subQuery.Add(query4, BooleanClause.Occur.SHOULD); 
      } 
      else 
      { 
       Query query1 = new WildcardQuery(new Term("ProductName", s)); 
       Query query2 = new WildcardQuery(new Term("ProductLongDescription", s)); 
       Query query3 = new WildcardQuery(new Term("BrandName", s)); 
       Query query4 = new WildcardQuery(new Term("CategoryName", s)); 
       subQuery.Add(query1, BooleanClause.Occur.SHOULD); 
       subQuery.Add(query2, BooleanClause.Occur.SHOULD); 
       subQuery.Add(query3, BooleanClause.Occur.SHOULD); 
       subQuery.Add(query4, BooleanClause.Occur.SHOULD); 
      } 
      booleanQuery.Add(subQuery, BooleanClause.Occur.MUST); 
     } 


    //Run the search....results are the same for simple multiword text queries 
     var result2 = _Searcher.Search(booleanQuery, null, maxResults); 
     var result = _Searcher.Search(query, null, maxResults); 

一種選擇使用手動查詢構建可能是共享MultiFieldQueryParser救了我,但我猜想它的解析方法不是線程安全的(雖然我只看過那關於Java版本...請糾正我,如果我在這個假設是錯誤的)。

我做錯了什麼或者這只是野獸的性質?

回答

5

MultiFieldQueryParser只是在場景下使用多個常規QueryParser,它會爲每個要查詢的字段創建一個。

創建QueryParser的成本比手動創建Query的成本更高是正常的。

它可以處理記錄在這裏的複雜查詢synthax:Apache Lucene - Query Parser Syntax

它也將處理使用Analyzer您指定的搜索查詢。如果在索引時使用Analyzer,則必須在搜索代碼中使用相同的Analyzer /邏輯。如果你不這樣做,你最終會失去結果。

如果您使用空白分析器進行索引,那麼您的代碼來手動構建布爾查詢是好的。

相關問題