如何优化对 Oracle 数据库的 SQL 查询

Question

我有一个对 Oracle 数据库的 SQL 查询 （查询通过与集群的每个元素具有最大平均连接将元素链接到集群相关的集群。两个元素的关系由它们共有的链接数定义。我们计算给定元素的关系到簇的每个元素，然后将元素与簇的关系作为该元素与簇中所有元素的关系的平均值，然后选择我们的元素与该元素具有最大关系的簇并将该元素添加到簇中)

INSERT INTO t_clusters (transaction_id,
                    CLUSTER_ID,
                    word_id,
                    connection)
SELECT 72,
      CLUSTER_ID,
      word_id,
      avg_cnt
 FROM (  SELECT word_id, CLUSTER_ID, AVG (cnt) avg_cnt
           FROM (  SELECT a.word_id,
                          b.word_id bword,
                          c.CLUSTER_ID,
                          COUNT (*) cnt
                     FROM t_semantic a,
                          t_words a1,
                          t_semantic b,
                          t_clusters c
                    WHERE     a.transaction_id = 72
                          AND a.word_id = a1.word_id
                          AND a1.checked = 0
                          AND CLUSTER_ID <> 0
                          AND a.link_id = b.link_id
                          AND b.word_id = c.word_id
                 GROUP BY a.word_id, b.word_id, c.CLUSTER_ID)
       GROUP BY word_id, CLUSTER_ID
         HAVING AVG (cnt) >= 3) zz
WHERE avg_cnt = (SELECT MAX (avg_cnt)
                   FROM (  SELECT AVG (cnt) avg_cnt, word_id
                             FROM (  SELECT a.word_id,
                                            b.word_id bword,
                                            c.CLUSTER_ID,
                                            COUNT (*) cnt
                                       FROM t_semantic a,
                                            t_words a1,
                                            t_semantic b,
                                            t_clusters c
                                      WHERE     a.transaction_id = 72
                                            AND a.word_id = a1.word_id
                                            AND a1.checked = 0
                                            AND CLUSTER_ID <> 0
                                            AND a.link_id = B.link_id
                                            AND b.word_id = c.word_id
                                   GROUP BY a.word_id,
                                            b.word_id,
                                            c.CLUSTER_ID)
                         GROUP BY word_id, CLUSTER_ID
                           HAVING AVG (cnt) > 3) zzy
                  WHERE zzy.word_id = zz.word_id)

它已经执行了 23 小时

关系表（t_semantic）的记录数为532125 集群表 (t_clusters) 中存储元素的记录数，已经连接到集群是 23014 表中元素尚未连接到簇的记录数（t_words）为 36111

在所有表中索引的字段 word_id、cluster_id、link_id。 有什么办法可以提高这个查询的性能，为什么这么慢？

Plan
INSERT STATEMENT  ALL_ROWSCost: 9,234                                                   
    28 FILTER                                               
        10 HASH GROUP BY  Cost: 9,234  Bytes: 46,000  Cardinality: 2,000                                            
            9 VIEW  Cost: 9,234  Bytes: 8,599,493  Cardinality: 373,891                                         
                8 SORT GROUP BY  Cost: 9,234  Bytes: 15,329,531  Cardinality: 373,891                                   
                    7 HASH JOIN  Cost: 5,215  Bytes: 15,329,531  Cardinality: 373,891                               
                        1 TABLE ACCESS FULL TABLE DKIM.T_WORDS Cost: 86  Bytes: 286,568  Cardinality: 35,821                            
                        6 HASH JOIN  Cost: 5,117  Bytes: 20,274,408  Cardinality: 614,376                           
                            4 HASH JOIN  Cost: 2,245  Bytes: 4,108,540  Cardinality: 205,427                        
                                2 INDEX FAST FULL SCAN INDEX (UNIQUE) DKIM.UK_CLUSTER Cost: 333  Bytes: 229,140  Cardinality: 22,914                    
                                3 TABLE ACCESS FULL TABLE DKIM.T_SEMANTIC Cost: 1,903  Bytes: 5,277,070  Cardinality: 527,707                   
                            5 TABLE ACCESS FULL TABLE DKIM.T_SEMANTIC Cost: 1,909  Bytes: 6,859,632  Cardinality: 527,664                       
        27 SORT AGGREGATE  Bytes: 26  Cardinality: 1                                            
            26 VIEW  Cost: 18  Bytes: 26  Cardinality: 1                                        
                25 FILTER                                   
                    24 SORT GROUP BY  Cost: 18  Bytes: 23  Cardinality: 1                               
                        23 VIEW  Cost: 18  Bytes: 23  Cardinality: 1                            
                            22 SORT GROUP BY  Cost: 18  Bytes: 41  Cardinality: 1                       
                                21 TABLE ACCESS BY INDEX ROWID TABLE DKIM.T_CLUSTERS Cost: 4  Bytes: 10  Cardinality: 1                     
                                    20 NESTED LOOPS  Cost: 17  Bytes: 41  Cardinality: 1                
                                        18 NESTED LOOPS  Cost: 9  Bytes: 93  Cardinality: 3             
                                            15 NESTED LOOPS  Cost: 5  Bytes: 21  Cardinality: 1         
                                                12 TABLE ACCESS BY INDEX ROWID TABLE DKIM.T_WORDS Cost: 2  Bytes: 8  Cardinality: 1     
                                                    11 INDEX UNIQUE SCAN INDEX (UNIQUE) DKIM.T_WORDS_PK Cost: 1  Cardinality: 1  
                                                14 TABLE ACCESS BY INDEX ROWID TABLE DKIM.T_SEMANTIC Cost: 3  Bytes: 13  Cardinality: 1     
                                                    13 INDEX RANGE SCAN INDEX DKIM.IND_SEMANTIC_WORD_ID Cost: 2  Cardinality: 1  
                                            17 TABLE ACCESS BY INDEX ROWID TABLE DKIM.T_SEMANTIC Cost: 4  Bytes: 30  Cardinality: 3         
                                                16 INDEX RANGE SCAN INDEX DKIM.IND_SEMANTIC_LINK_ID Cost: 2  Cardinality: 3     
                                        19 INDEX RANGE SCAN INDEX DKIM.CLUSTERS_WORD_ID Cost: 2  Cardinality: 1     

Answer 1

使用分析函数来确定最大 avg_cnt 可能有助于加快速度，例如：

insert into t_clusters (transaction_id,
                        cluster_id,
                        word_id,
                        connection)
select 72,
       cluster_id,
       word_id,
       avg_cnt
 from  (select   word_id,
                 cluster_id,
                 avg(cnt) avg_cnt,
                 max(avg(cnt)) over (partition by word_id) max_avg_cnt
        from     (select   a.word_id,
                           b.word_id bword,
                           c.cluster_id,
                           count (*) cnt
                  from     t_semantic a,
                           t_words a1,
                           t_semantic b,
                           t_clusters c
                  where    a.transaction_id = 72
                  and      a.word_id = a1.word_id
                  and      a1.checked = 0
                  and      cluster_id <> 0
                  and      a.link_id = b.link_id
                  and      b.word_id = c.word_id
                  group by a.word_id, b.word_id, c.cluster_id)
        group by word_id, cluster_id
        having   avg(cnt) >= 3)
where  avg_cnt = max_avg_cnt;