Oceanbase目前只实现了一个基于代价的查询改写(or-expansion)。其实数据库中很多高级的改写规则(比如complex view merge和窗口函数改写)都需要基于代价进行改写,Oceanbase后续会实现这些复杂的改写规则。

    或展开(Or-Expansion)

    Or-expansion把一个查询改写成若干个用union组成的子查询,这个改写可能会给每个子查询提供更优的优化空间,但是也会导致多个子查询的执行,所以这个改写需要基于代价去判断。通常来说,Or-expansion的改写主要有如下三个作用:

    • 允许每个分支使用不同的索引来加速查询

    如下图所示,Q1会被改写成Q2的形式,其中Q2中的谓词lnnvl(t1.a = 1)保证了这两个子查询不会生成重复的结果。如果不进 行改写,Q1一般来说会选择主表作为访问路径,对于Q2来说,如果t1上存在索引(a)和索引(b),那么该改写可能会让Q2中的每一个子查询选择索引作为访问路径。

    1. Q1: select * from t1 where t1.a = 1 or t1.b = 1;
    2. Q2: select * from t1 where t1.a = 1 union all select * from t1.b = 1 and lnnvl(t1.a = 1);
    1. OceanBase (root@test)> create table t1(a int, b int, c int, d int, e int, index idx_a(a), index idx_b(b));
    2. --- 如果不进行or-expansion的改写,该查询只能使用主表访问路径
    3. OceanBase (root@test)> explain select/*+NO_REWRITE()*/ * from t1 where t1.a = 1 or t1.b = 1;
    4. +-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
    5. | Query Plan |
    6. +-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
    7. | ===================================
    8. |ID|OPERATOR |NAME|EST. ROWS|COST|
    9. -----------------------------------
    10. |0 |TABLE SCAN|t1 |4 |649 |
    11. ===================================
    12. Outputs & filters:
    13. -------------------------------------
    14. 0 - output([t1.a], [t1.b], [t1.c], [t1.d], [t1.e]), filter([t1.a = 1 OR t1.b = 1]),
    15. access([t1.a], [t1.b], [t1.c], [t1.d], [t1.e]), partitions(p0)
    16. --- 改写之后,每个子查询能使用不同的索引访问路径
    17. OceanBase (root@test)> explain select * from t1 where t1.a = 1 or t1.b = 1;
    18. +------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
    19. | Query Plan |
    20. +------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
    21. | =========================================
    22. |ID|OPERATOR |NAME |EST. ROWS|COST|
    23. -----------------------------------------
    24. |0 |UNION ALL | |3 |190 |
    25. |1 | TABLE SCAN|t1(idx_a)|2 |94 |
    26. |2 | TABLE SCAN|t1(idx_b)|1 |95 |
    27. =========================================
    28. Outputs & filters:
    29. -------------------------------------
    30. 0 - output([UNION(t1.a, t1.a)], [UNION(t1.b, t1.b)], [UNION(t1.c, t1.c)], [UNION(t1.d, t1.d)], [UNION(t1.e, t1.e)]), filter(nil)
    31. 1 - output([t1.a], [t1.b], [t1.c], [t1.d], [t1.e]), filter(nil),
    32. access([t1.a], [t1.b], [t1.c], [t1.d], [t1.e]), partitions(p0)
    33. 2 - output([t1.a], [t1.b], [t1.c], [t1.d], [t1.e]), filter([lnnvl(t1.a = 1)]),
    34. access([t1.a], [t1.b], [t1.c], [t1.d], [t1.e]), partitions(p02
    • 允许每个分支使用不同的连接算法来加速查询,避免使用笛卡尔连接

    如下图所示,Q1会被改写成Q2的形式。对于Q1来说,它的连接方式只能是nested loop join(笛卡尔乘积), 但是被改写之后,每个子查询都可以选择nested loop join,hash join或者merge join,这样会有更多的优化空间。

    1. Q1: select * from t1, t2 where t1.a = t2.a or t1.b = t2.b;
    2. Q2: select * from t1, t2 where t1.a = t2.a union all
    3. select * from t1, t2 where t1.b = t2.b and lnnvl(t1.a = t2.a)
    1. OceanBase (root@test)> create table t1(a int, b int);
    2. Query OK, 0 rows affected (0.17 sec)
    3. OceanBase (root@test)> create table t2(a int, b int);
    4. Query OK, 0 rows affected (0.13 sec)
    5. ---如果不进行改写,只能使用nested loop join
    6. OceanBase (root@test)> explain select/*+NO_REWRITE()*/ * from t1, t2 where t1.a = t2.a or t1.b = t2.b;
    7. | ===========================================
    8. |ID|OPERATOR |NAME|EST. ROWS|COST |
    9. -------------------------------------------
    10. |0 |NESTED-LOOP JOIN| |3957 |585457|
    11. |1 | TABLE SCAN |t1 |1000 |499 |
    12. |2 | TABLE SCAN |t2 |4 |583 |
    13. ===========================================
    14. Outputs & filters:
    15. -------------------------------------
    16. 0 - output([t1.a], [t1.b], [t2.a], [t2.b]), filter(nil),
    17. conds(nil), nl_params_([t1.a], [t1.b])
    18. 1 - output([t1.a], [t1.b]), filter(nil),
    19. access([t1.a], [t1.b]), partitions(p0)
    20. 2 - output([t2.a], [t2.b]), filter([? = t2.a OR ? = t2.b]),
    21. access([t2.a], [t2.b]), partitions(p0)
    22. ---被改写之后,每个子查询都使用了hash join
    23. OceanBase (root@test)> explain select * from t1, t2 where t1.a = t2.a or t1.b = t2.b;
    24. |ID|OPERATOR |NAME|EST. ROWS|COST|
    25. -------------------------------------
    26. |0 |UNION ALL | |2970 |9105|
    27. |1 | HASH JOIN | |1980 |3997|
    28. |2 | TABLE SCAN|t1 |1000 |499 |
    29. |3 | TABLE SCAN|t2 |1000 |499 |
    30. |4 | HASH JOIN | |990 |3659|
    31. |5 | TABLE SCAN|t1 |1000 |499 |
    32. |6 | TABLE SCAN|t2 |1000 |499 |
    33. =====================================
    34. Outputs & filters:
    35. -------------------------------------
    36. 0 - output([UNION(t1.a, t1.a)], [UNION(t1.b, t1.b)], [UNION(t2.a, t2.a)], [UNION(t2.b, t2.b)]), filter(nil)
    37. 1 - output([t1.a], [t1.b], [t2.a], [t2.b]), filter(nil),
    38. equal_conds([t1.a = t2.a]), other_conds(nil)
    39. 2 - output([t1.a], [t1.b]), filter(nil),
    40. access([t1.a], [t1.b]), partitions(p0)
    41. 3 - output([t2.a], [t2.b]), filter(nil),
    42. access([t2.a], [t2.b]), partitions(p0)
    43. 4 - output([t1.a], [t1.b], [t2.a], [t2.b]), filter(nil),
    44. equal_conds([t1.b = t2.b]), other_conds([lnnvl(t1.a = t2.a)])
    45. 5 - output([t1.a], [t1.b]), filter(nil),
    46. access([t1.a], [t1.b]), partitions(p0)
    47. 6 - output([t2.a], [t2.b]), filter(nil),
    48. access([t2.a], [t2.b]), partitions(p0)
    • 允许每个分支分别消除排序,更加快速的获取top-k结果

    如下图所示,Q1会被改写成Q2。对于Q1来说,执行方式是只能把满足条件的行数找出来,然后进行排序,最终取top-10结果。对于Q2来说,如果存在索引(a,b), 那么Q2中的两个子查询都可以使用索引把排序消除,每个子查询取top-10结果,然后最终对这20行数据排序一下取出最终的top-10行。

    1. Q1: select * from t1 where t1.a = 1 or t1.a = 2 order by b limit 10;
    2. Q2: select * from
    3. (select * from t1 where t1.a = 1 order by b limit 10 union all
    4. select * from t1 where t1.a = 2 order by b limit 10) as temp
    5. order by temp.b limit 10;
    1. OceanBase (root@test)> create table t1(a int, b int, index idx_a(a, b));
    2. Query OK, 0 rows affected (0.20 sec)
    3. ---不改写的话,需要排序最终获取top-k结果
    4. OceanBase (root@test)> explain select/*+NO_REWRITE()*/ * from t1 where t1.a = 1 or t1.a = 2 order by b limit 10;
    5. | ==========================================
    6. |ID|OPERATOR |NAME |EST. ROWS|COST|
    7. ------------------------------------------
    8. |0 |LIMIT | |4 |77 |
    9. |1 | TOP-N SORT | |4 |76 |
    10. |2 | TABLE SCAN|t1(idx_a)|4 |73 |
    11. ==========================================
    12. Outputs & filters:
    13. -------------------------------------
    14. 0 - output([t1.a], [t1.b]), filter(nil), limit(10), offset(nil)
    15. 1 - output([t1.a], [t1.b]), filter(nil), sort_keys([t1.b, ASC]), topn(10)
    16. 2 - output([t1.a], [t1.b]), filter(nil),
    17. access([t1.a], [t1.b]), partitions(p0)
    18. --- 进行改写的话,排序算子可以被消除,最终获取top-k结果
    19. OceanBase (root@test)> explain select * from t1 where t1.a = 1 or t1.a = 2 order by b limit 10;
    20. | ===========================================
    21. |ID|OPERATOR |NAME |EST. ROWS|COST|
    22. -------------------------------------------
    23. |0 |LIMIT | |3 |76 |
    24. |1 | TOP-N SORT | |3 |76 |
    25. |2 | UNION ALL | |3 |74 |
    26. |3 | TABLE SCAN|t1(idx_a)|2 |37 |
    27. |4 | TABLE SCAN|t1(idx_a)|1 |37 |
    28. ===========================================
    29. Outputs & filters:
    30. -------------------------------------
    31. 0 - output([UNION(t1.a, t1.a)], [UNION(t1.b, t1.b)]), filter(nil), limit(10), offset(nil)
    32. 1 - output([UNION(t1.a, t1.a)], [UNION(t1.b, t1.b)]), filter(nil), sort_keys([UNION(t1.b, t1.b), ASC]), topn(10)
    33. 2 - output([UNION(t1.a, t1.a)], [UNION(t1.b, t1.b)]), filter(nil)
    34. 3 - output([t1.a], [t1.b]), filter(nil),
    35. access([t1.a], [t1.b]), partitions(p0),
    36. limit(10), offset(nil)
    37. 4 - output([t1.a], [t1.b]), filter([lnnvl(t1.a = 1)]),
    38. access([t1.a], [t1.b]), partitions(p0),
    39. limit(10), offset(nil)