全域性索引與分割槽索引對於SQL效能影響的比較

阿新 • • 發佈：2022-06-01

KingbaseES 提供了對於分割槽表 global index 的支援。global index 不僅提供了對於唯一索引功能的改進（無需包含分割槽鍵），而且在效能上相比非global index （local index）有很大的提升（無法提供分割槽條件情況下）。以下舉例說明二者在效能方面的差異。

1、準備資料

create table t1(id1 integer,id2 integer,name text) partition by hash(id1) partitions 200;

insert into t1 select generate_series(1,10000000),generate_series(1,10000000),repeat('a',500);

2、本地索引的效能

沒有提供分割槽條件時

create index ind_t1_id2 on t1(id2);

test=# \di+ ind_t1_id2
                                List of relations
 Schema |    Name    |       Type        | Owner  | Table |  Size   | Description
--------+------------+-------------------+--------+-------+---------+-------------
 public 
 | ind_t1_id2 | partitioned index | system | t1    | 0 bytes |
(1 row)

test=# explain analyze select * from t1 where id2=10004;
                                                           QUERY PLAN
---------------------------------------------------------------------------------------------------------------------------------
 Append  (cost 
=0.29..1662.50 rows=200 width=512) (actual time=1.324..3.249 rows=1 loops=1)
   ->  Index Scan using t1_p0_id2_idx on t1_p0  (cost=0.29..8.31 rows=1 width=512) (actual time=0.054..0.055 rows=0 loops=1)
         Index Cond: (id2 = 10004)
   ->  Index Scan using t1_p1_id2_idx on t1_p1  (cost=0.29..8.31 rows=1 width=512) (actual time=0.065..0.065 rows=0 loops=1)
         Index Cond: (id2 = 10004)
   ......
   ->  Index Scan using t1_p198_id2_idx on t1_p198  (cost=0.29..8.31 rows=1 width=512) (actual time=0.031..0.031 rows=0 loops=1)
         Index Cond: (id2 = 10004)
   ->  Index Scan using t1_p199_id2_idx on t1_p199  (cost=0.29..8.31 rows=1 width=512) (actual time=0.025..0.025 rows=0 loops=1)
         Index Cond: (id2 = 10004)
 Planning Time: 39.262 ms
 Execution Time: 5.673 ms
(403 rows)

使用非全域性索引，並且沒有提供分割槽條件情況下，優化器需要讀取所有索引分割槽及表分割槽的統計資料，才能確定最優的執行計劃。對於資料訪問，同樣需要訪問所有分割槽的索引（即使該分割槽沒有所需要的資料）。

提供分割槽條件時

test=# explain analyze select * from t1 where id2=10004 and id1=10004;
                                                       QUERY PLAN
-------------------------------------------------------------------------------------------------------------------------
 Index Scan using t1_p71_id2_idx on t1_p71  (cost=0.29..8.31 rows=1 width=512) (actual time=0.045..0.046 rows=1 loops=1)
   Index Cond: (id2 = 10004)
   Filter: (id1 = 10004)
 Planning Time: 0.346 ms
 Execution Time: 0.064 ms
(5 rows)

在提供分割槽條件情況下，只需要訪問單個索引分割槽及表分割槽的統計資料，因此，所需的語句的解析時間更少。

3、全域性索引的效能

create unique index ind_t1_id2 on t1(id2) global;

test=# \di+ ind_t1_id2
                             List of relations
 Schema |    Name    |     Type     | Owner  | Table |  Size  | Description
--------+------------+--------------+--------+-------+--------+-------------
 public | ind_t1_id2 | global index | system | t1    | 215 MB |
(1 row)


test=# explain analyze select * from t1 where id2=10004;
                                                        QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------
 Global Index Scan using ind_t1_id2 on t1  (cost=0.38..8.39 rows=200 width=512) (actual time=0.136..0.137 rows=1 loops=1)
   Index Cond: (id2 = 10004)
 Planning Time: 9.896 ms
 Execution Time: 0.264 ms
(4 rows)