首批通过分布式安全可靠测评,为关键业务系统打造
Optimizer Hint
更新时间:2026-04-05 20:31:50
Hint 机制可以使优化器生成指定的执行计划。
一般情况下,优化器会为用户查询选择最佳的执行计划,不需要用户使用 Hint 指定。但在某些场景下,优化器生成的执行计划可能不满足用户的要求,这时就需要用户使用 Hint 来指定生成某种执行计划。
Hint 语法
Hint 从语法上看是一种特殊的 SQL 注释,所不同的是在注释的左标记后("/*"符号)增加了一个"+"。 既然是注释,如果服务器端无法识别 SQL 语句中的 Hint,优化器会选择忽略用户 Hint 而使用默认的计划生成逻辑结构。另外,Hint 只影响优化器生成计划的逻辑结构,而不影响 SQL 语句的语义。
{ DELETE | INSERT | SELECT | UPDATE | REPLACE } /*+ [hint_text][,hint_text]... */
注意
如果使用 MySQL 的 C 客户端执行带 Hint 的 SQL 语句,需要使用 -c 选项登录,否则 MySQL 客户端会将 Hint 作为注释从用户 SQL 语句中去除,导致系统无法收到用户 Hint。
Hint 参数
Hint 相关参数名称、语义和语法如下表所示。
| 名称 | 语法 | 语义 |
|---|---|---|
| NO_REWRITE | NO_REWRITE | 禁止 SQL 改写。 |
| READ_CONSISTENCY | READ_CONSISTENCY (WEAK[STRONGFROZEN]) | 读一致性设置(弱/强)。 |
| INDEX_HINT | /*+ INDEX(table_name index_name) */ | 设置表索引。 |
| QUERY_TIMEOUT | QUERY_TIMEOUT(INTNUM) | 设置超时时间。 |
| LOG_LEVEL | LOG_LEVEL([']log_level[']) | 设置日志级别,当设置模块级别语句时候,以第一个单引号(')作为开始,第二个单引号(')作为结束;例如'DEBUG'。 |
| LEADING | LEADING([qb_name] TBL_NAME_LIST) | 设置联接顺序。 |
| ORDERED | ORDERED | 设置按照 SQL 中的顺序进行联接。 |
| FULL | FULL([qb_name] TBL_NAME) | 设置表访问路径为主表等价于 INDEX(TBL_NAME PRIMARY)。 |
| USE_PLAN_CACHE | USE_PLAN_CACHE(NONE[DEFAULT]) | 设置是否使用计划缓存:
|
| USE_MERGE | USE_MERGE([qb_name] TBL_NAME_LIST) | 设置指定表在作为右表时使用 Merge Join。 |
| USE_HASH | USE_HASH([qb_name] TBL_NAME_LIST) | 设置指定表在作为右表时使用 Hash Join。 |
| NO_USE_HASH | NO_USE_HASH([qb_name] TBL_NAME_LIST) | 设置指定表在作为右表时不使用 Hash Join。 |
| USE_NL | USE_NL([qb_name] TBL_NAME_LIST) | 设置指定表在作为右表时使用 Nested Loop Join。 |
| USE_BNL | USE_BNL([qb_name] TBL_NAME_LIST) | 设置指定表在作为右表时使用 Block Nested Loop Join |
| USE_HASH_AGGREGATION | USE_HASH_AGGREGATION([qb_name]) | 设置聚合算法为 Hash。例如 Hash Group By 或者 Hash Distinct。 |
| NO_USE_HASH_AGGREGATION | NO_USE_HASH_AGGREGATION([qb_name]) | 设置 Aggregate 方法不使用 Hash Aggregate,使用 Merge Group By 或者 Merge Distinct。 |
| USE_LATE_MATERIALIZATION | USE_LATE_MATERIALIZATION | 设置使用晚期物化。 |
| NO_USE_LATE_MATERIALIZATION | NO_USE_LATE_MATERIALIZATION | 设置不使用晚期物化。 |
| TRACE_LOG | TRACE_LOG | 设置收集 Trace 记录用于 SHOW TRACE 展示。 |
| QB_NAME | QB_NAME( NAME ) | 设置 Query Block 的名称。 |
| PARALLEL | PARALLEL(INTNUM) | 设置分布式执行并行度。 |
| TOPK | TOPK(PRECISION MINIMUM_ROWS) | 设置模糊查询的精度和最小行数。 其中 PRECSION 为整型,取值范围[0,100],表示模糊查询的行数百分比;MINIMUM_ROWS 为最小返回行数。 |
| MAX_CONCURRENT | MAX_CONCURRENT(n) | 限制这个 SQL 文本的并发数。 |
UNION_MERGE
注意OceanBase 数据库从 V4.4.1 版本开始移除 |
UNION_MERGE(table_name index_name_list) | 设置对应的 Index Merge 计划。 |
| PUSH_SUBQ | PUSH_SUBQ[(@qb_name)] | 用于指示优化器尽早执行未改写成连接的子查询。 |
| NO_PUSH_SUBQ | NO_PUSH_SUBQ[(@qb_name)] | 是 PUSH_SUBQ 的反向操作,用于优化器将未改写成连接的子查询放在最后执行。 |
| OPT_PARAM | OPT_PARAM(parameter_name [,] parameter_value) | 指定在查询级别更新一些优化器相关的配置项/系统变量。 |
| INDEX_MERGE | INDEX_MERGE([@query_block_name] tbl_name [index_name [,index_name]...]) | 用于控制指定的表使用 index merge 计划。
注意
|
| NO_INDEX_MERGE | NO_INDEX_MERGE([@query_block_name] tbl_name [index_name [,index_name]...]) | 用于控制指定的表不使用指定的索引生成 index merge 计划。
注意
|
说明
QB_NAME语法是:@NAMETBL_NAME语法是:[db_name.]relation_name [qb_name]
QB_NAME 参数
在 DML 语句中,每一个 query_block 都会有一个 QB_NAME(Query Block Name),可以由用户指定,也可以由系统自动生成。在用户没有用 Hint 指定 QB_NAME 的时候,系统会按照 SEL$1、SEL$2、UPD$1、DEL$1 方式从左到右(实际也是 Resolver 的解析顺序)依次生成。
通过 QB_NAME 可以精确定位到每一个表,也可以在某处指定任意 Query Block 的行为。TBL_NAME 中的 QB_NAME 用于定位表,在 Hint 中最前面的 QB_NAME 用于定位 Hint 作用于哪一个 query_block。
如下例所示,按照默认规则,会为 SEL$1 中的 t1 表选择 t1_c1 路径,为 SEL$2 中的 t2 表选择主表(Primary)访问。
obclient> CREATE TABLE t1(c1 INT, c2 INT, KEY t1_c1(c1));
Query OK, 0 rows affected
obclient> CREATE TABLE t2(c1 INT, c2 INT, KEY t2_c1(c1));
Query OK, 0 rows affected
obclient> EXPLAIN SELECT * FROM t1, (SELECT * FROM t2 WHERE c2 = 1 LIMIT 5)
WHERE t1.c1 = 1;
+-----------------------------------------------------------------------------------------------------------+
| Query Plan |
+-----------------------------------------------------------------------------------------------------------+
| ====================================================================== |
| |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| |
| ---------------------------------------------------------------------- |
| |0 |NESTED-LOOP JOIN CARTESIAN | |1 |7 | |
| |1 |├─TABLE RANGE SCAN |t1(t1_c1) |1 |7 | |
| |2 |└─MATERIAL | |1 |3 | |
| |3 | └─SUBPLAN SCAN |ANONYMOUS_VIEW1|1 |3 | |
| |4 | └─TABLE FULL SCAN |t2 |1 |3 | |
| ====================================================================== |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([t1.c1], [t1.c2], [ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]), filter(nil), rowset=16 |
| conds(nil), nl_params_(nil), use_batch=false |
| 1 - output([t1.c1], [t1.c2]), filter(nil), rowset=16 |
| access([t1.__pk_increment], [t1.c1], [t1.c2]), partitions(p0) |
| is_index_back=true, is_global_index=false, |
| range_key([t1.c1], [t1.__pk_increment]), range(1,MIN ; 1,MAX), |
| range_cond([t1.c1 = 1]) |
| 2 - output([ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]), filter(nil), rowset=16 |
| 3 - output([ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]), filter(nil), rowset=16 |
| access([ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]) |
| 4 - output([t2.c1], [t2.c2]), filter([t2.c2 = 1]), rowset=16 |
| access([t2.c2], [t2.c1]), partitions(p0) |
| limit(5), offset(nil), is_index_back=false, is_global_index=false, filter_before_indexback[false], |
| range_key([t2.__pk_increment]), range(MIN ; MAX)always true |
+-----------------------------------------------------------------------------------------------------------+
25 rows in set
如果 SQL 通过 Hint 来指定 SEL$1 的 t1 表走主表,SEL$2 的 t2 表走索引,示例如下:
obclient> EXPLAIN SELECT /*+INDEX(t1 PRIMARY) INDEX(@SEL$2 t2 t2_c1)*/ *
FROM t1 , (SELECT * FROM t2 WHERE c2 = 1 LIMIT 5)
WHERE t1.c1 = 1;
+----------------------------------------------------------------------------------------------------------+
| Query Plan |
+----------------------------------------------------------------------------------------------------------+
| ====================================================================== |
| |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| |
| ---------------------------------------------------------------------- |
| |0 |NESTED-LOOP JOIN CARTESIAN | |1 |3 | |
| |1 |├─TABLE FULL SCAN |t1 |1 |3 | |
| |2 |└─MATERIAL | |1 |7 | |
| |3 | └─SUBPLAN SCAN |ANONYMOUS_VIEW1|1 |7 | |
| |4 | └─TABLE FULL SCAN |t2(t2_c1) |1 |7 | |
| ====================================================================== |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([t1.c1], [t1.c2], [ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]), filter(nil), rowset=16 |
| conds(nil), nl_params_(nil), use_batch=false |
| 1 - output([t1.c1], [t1.c2]), filter([t1.c1 = 1]), rowset=16 |
| access([t1.c1], [t1.c2]), partitions(p0) |
| is_index_back=false, is_global_index=false, filter_before_indexback[false], |
| range_key([t1.__pk_increment]), range(MIN ; MAX)always true |
| 2 - output([ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]), filter(nil), rowset=16 |
| 3 - output([ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]), filter(nil), rowset=16 |
| access([ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]) |
| 4 - output([t2.c1], [t2.c2]), filter([t2.c2 = 1]), rowset=16 |
| access([t2.__pk_increment], [t2.c2], [t2.c1]), partitions(p0) |
| limit(5), offset(nil), is_index_back=true, is_global_index=false, filter_before_indexback[false], |
| range_key([t2.c1], [t2.__pk_increment]), range(MIN,MIN ; MAX,MAX)always true |
+----------------------------------------------------------------------------------------------------------+
24 rows in set
说明
由于 INDEX(t1 PRIMARY) 本来就存在于 SEL$1,可以不用再指定 Hint 作用的 Query Block。
上述示例中 SQL 也可以写成如下方式:
SELECT /*+INDEX(t1 PRIMARY) INDEX(@SEL$2 t2@SEL$2 t2_c1)*/ * FROM t1 , (SELECT * FROM t2 WHERE c2 = 1 LIMIT 5) WHERE t1.c1 = 1;
或:
SELECT /*+INDEX(t1 PRIMARY)*/ * FROM t1 , (SELECT /*+INDEX(t2 t2_c1)*/ * FROM t2 WHERE c2 = 1 LIMIT 5) WHERE t1.c1 = 1;
或:
SELECT /*+INDEX(@SEL$1 t1 PRIMARY) INDEX(@SEL$2 t2 t2_c1)*/ * FROM t1 , (SELECT * FROM t2 WHERE c2 = 1 LIMIT 5) WHERE t1.c1 = 1;
对于该 Hint,可以通过 EXPLAIN EXTENDED 命令的执行结果中的 Outline Data 来查看它的所有信息。
obclient> EXPLAIN EXTENDED SELECT * FROM t1, (SELECT * FROM t2 WHERE c2 = 1 LIMIT 5)
WHERE t1.c1 = 1;
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Query Plan |
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| ====================================================================== |
| |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| |
| ---------------------------------------------------------------------- |
| |0 |NESTED-LOOP JOIN CARTESIAN | |1 |7 | |
| |1 |├─TABLE RANGE SCAN |t1(t1_c1) |1 |7 | |
| |2 |└─MATERIAL | |1 |3 | |
| |3 | └─SUBPLAN SCAN |ANONYMOUS_VIEW1|1 |3 | |
| |4 | └─TABLE FULL SCAN |t2 |1 |3 | |
| ====================================================================== |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([t1.c1(0x7f20d7035330)], [t1.c2(0x7f20d70358b0)], [ANONYMOUS_VIEW1.c1(0x7f20d7035be0)], [ANONYMOUS_VIEW1.c2(0x7f20d7035f10)]), filter(nil), rowset=16 |
| conds(nil), nl_params_(nil), use_batch=false |
| 1 - output([t1.c1(0x7f20d7035330)], [t1.c2(0x7f20d70358b0)]), filter(nil), rowset=16 |
| access([t1.__pk_increment(0x7f20d7036b10)], [t1.c1(0x7f20d7035330)], [t1.c2(0x7f20d70358b0)]), partitions(p0) |
| is_index_back=true, is_global_index=false, |
| range_key([t1.c1(0x7f20d7035330)], [t1.__pk_increment(0x7f20d7036b10)]), range(1,MIN ; 1,MAX), |
| range_cond([t1.c1(0x7f20d7035330) = 1(0x7f20d7034b70)]) |
| 2 - output([ANONYMOUS_VIEW1.c1(0x7f20d7035be0)], [ANONYMOUS_VIEW1.c2(0x7f20d7035f10)]), filter(nil), rowset=16 |
| 3 - output([ANONYMOUS_VIEW1.c1(0x7f20d7035be0)], [ANONYMOUS_VIEW1.c2(0x7f20d7035f10)]), filter(nil), rowset=16 |
| access([ANONYMOUS_VIEW1.c1(0x7f20d7035be0)], [ANONYMOUS_VIEW1.c2(0x7f20d7035f10)]) |
| 4 - output([t2.c1(0x7f20d7033a10)], [t2.c2(0x7f20d7033490)]), filter([t2.c2(0x7f20d7033490) = 1(0x7f20d7032cd0)]), rowset=16 |
| access([t2.c2(0x7f20d7033490)], [t2.c1(0x7f20d7033a10)]), partitions(p0) |
| limit(5), offset(nil), is_index_back=false, is_global_index=false, filter_before_indexback[false], |
| range_key([t2.__pk_increment(0x7f20d70365e0)]), range(MIN ; MAX)always true |
| Used Hint: |
| ------------------------------------- |
| /*+ |
| |
| */ |
| Qb name trace: |
| ------------------------------------- |
| stmt_id:0, stmt_type:T_EXPLAIN |
| stmt_id:1, SEL$1 |
| stmt_id:2, SEL$2 |
| Outline Data: |
| ------------------------------------- |
| /*+ |
| BEGIN_OUTLINE_DATA |
| LEADING(@"SEL$1" ("aabb"."t1"@"SEL$1" "ANONYMOUS_VIEW1"@"SEL$1")) |
| USE_NL(@"SEL$1" "ANONYMOUS_VIEW1"@"SEL$1") |
| USE_NL_MATERIALIZATION(@"SEL$1" "ANONYMOUS_VIEW1"@"SEL$1") |
| INDEX(@"SEL$1" "aabb"."t1"@"SEL$1" "t1_c1") |
| FULL(@"SEL$2" "aabb"."t2"@"SEL$2") |
| OPTIMIZER_FEATURES_ENABLE('4.3.0.0') |
| END_OUTLINE_DATA |
| */ |
| Optimization Info: |
| ------------------------------------- |
| t1: |
| table_rows:1 |
| physical_range_rows:1 |
| logical_range_rows:1 |
| index_back_rows:1 |
| output_rows:1 |
| table_dop:1 |
| dop_method:Table DOP |
| avaiable_index_name:[t1_c1, t1] |
| unstable_index_name:[t1] |
| stats version:0 |
| dynamic sampling level:0 |
| estimation method:[DEFAULT, STORAGE] |
| t2: |
| table_rows:1 |
| physical_range_rows:1 |
| logical_range_rows:1 |
| index_back_rows:0 |
| output_rows:1 |
| table_dop:1 |
| dop_method:Table DOP |
| avaiable_index_name:[t2_c1, t2] |
| pruned_index_name:[t2_c1] |
| stats version:0 |
| dynamic sampling level:0 |
| estimation method:[DEFAULT, STORAGE] |
| Plan Type: |
| LOCAL |
| Note: |
| Degree of Parallelisim is 1 because of table property |
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------+
79 rows in set
Hint 使用规则
Hint 的一般使用规则如下:
对于没有指定 Query Block 的 Hint 表明作用于本 Query Block。
示例 1:由于
t2表在 Query Block2,并且无法通过改写提升到 Query Block1,所以 Hint 无法生效。obclient> EXPLAIN SELECT /*+INDEX(t2 t2_c1)*/ * FROM t1 , (SELECT * FROM t2 WHERE c2 = 1 LIMIT 5) WHERE t1.c1 = 1; +-----------------------------------------------------------------------------------------------------------+ | Query Plan | +-----------------------------------------------------------------------------------------------------------+ | ====================================================================== | | |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| | | ---------------------------------------------------------------------- | | |0 |NESTED-LOOP JOIN CARTESIAN | |1 |7 | | | |1 |├─TABLE RANGE SCAN |t1(t1_c1) |1 |7 | | | |2 |└─MATERIAL | |1 |3 | | | |3 | └─SUBPLAN SCAN |ANONYMOUS_VIEW1|1 |3 | | | |4 | └─TABLE FULL SCAN |t2 |1 |3 | | | ====================================================================== | | Outputs & filters: | | ------------------------------------- | | 0 - output([t1.c1], [t1.c2], [ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]), filter(nil), rowset=16 | | conds(nil), nl_params_(nil), use_batch=false | | 1 - output([t1.c1], [t1.c2]), filter(nil), rowset=16 | | access([t1.__pk_increment], [t1.c1], [t1.c2]), partitions(p0) | | is_index_back=true, is_global_index=false, | | range_key([t1.c1], [t1.__pk_increment]), range(1,MIN ; 1,MAX), | | range_cond([t1.c1 = 1]) | | 2 - output([ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]), filter(nil), rowset=16 | | 3 - output([ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]), filter(nil), rowset=16 | | access([ANONYMOUS_VIEW1.c1], [ANONYMOUS_VIEW1.c2]) | | 4 - output([t2.c1], [t2.c2]), filter([t2.c2 = 1]), rowset=16 | | access([t2.c2], [t2.c1]), partitions(p0) | | limit(5), offset(nil), is_index_back=false, is_global_index=false, filter_before_indexback[false], | | range_key([t2.__pk_increment]), range(MIN ; MAX)always true | +-----------------------------------------------------------------------------------------------------------+ 25 rows in set示例 2:如果优化器能够重新整合子查询与外层查询,从而只使用一个 Query Block,那么 Hint 可能生效。
obclient> EXPLAIN SELECT /*+INDEX(t2 t2_c1)*/ * FROM t1 , (SELECT * FROM t2 WHERE c2 = 1) WHERE t1.c1 = 1; +------------------------------------------------------------------------------------+ | Query Plan | +------------------------------------------------------------------------------------+ | ================================================================ | | |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| | | ---------------------------------------------------------------- | | |0 |NESTED-LOOP JOIN CARTESIAN | |1 |7 | | | |1 |├─TABLE RANGE SCAN |t1(t1_c1)|1 |7 | | | |2 |└─MATERIAL | |1 |7 | | | |3 | └─TABLE FULL SCAN |t2(t2_c1)|1 |7 | | | ================================================================ | | Outputs & filters: | | ------------------------------------- | | 0 - output([t1.c1], [t1.c2], [t2.c1], [t2.c2]), filter(nil), rowset=16 | | conds(nil), nl_params_(nil), use_batch=false | | 1 - output([t1.c1], [t1.c2]), filter(nil), rowset=16 | | access([t1.__pk_increment], [t1.c1], [t1.c2]), partitions(p0) | | is_index_back=true, is_global_index=false, | | range_key([t1.c1], [t1.__pk_increment]), range(1,MIN ; 1,MAX), | | range_cond([t1.c1 = 1]) | | 2 - output([t2.c1], [t2.c2]), filter(nil), rowset=16 | | 3 - output([t2.c2], [t2.c1]), filter([t2.c2 = 1]), rowset=16 | | access([t2.__pk_increment], [t2.c2], [t2.c1]), partitions(p0) | | is_index_back=true, is_global_index=false, filter_before_indexback[false], | | range_key([t2.c1], [t2.__pk_increment]), range(MIN,MIN ; MAX,MAX)always true | +------------------------------------------------------------------------------------+ 22 rows in set
对于上面的示例,如果优化器由于某些原因(比如子查询内部操作阻止了提升合并)无法将子查询 (SELECT * FROM t2 WHERE c2 = 1) 合并到外层查询,那么 t2 表仍然被视为子查询的一部分,因此外层查询中的 Hint 无效。
如下示例,由于外层查询和子查询均引用了同一张表 t1,并且都试图利用 PRIMARY 索引,这可能会导致优化器在解释和处理 Hint 时遇到冲突,从而使得 Hint 可能无效。
obclient> EXPLAIN SELECT /*+INDEX(t1 PRIMARY)*/ *
FROM t1 , (SELECT * FROM t1 WHERE c1 = 1)
WHERE t1.c1 = 1;
+--------------------------------------------------------------------------+
| Query Plan |
+--------------------------------------------------------------------------+
| ================================================================ |
| |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| |
| ---------------------------------------------------------------- |
| |0 |NESTED-LOOP JOIN CARTESIAN | |1 |7 | |
| |1 |├─TABLE RANGE SCAN |t1(t1_c1)|1 |7 | |
| |2 |└─MATERIAL | |1 |7 | |
| |3 | └─TABLE RANGE SCAN |t1(t1_c1)|1 |7 | |
| ================================================================ |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([t1.c1], [t1.c2], [t1.c1], [t1.c2]), filter(nil), rowset=16 |
| conds(nil), nl_params_(nil), use_batch=false |
| 1 - output([t1.c1], [t1.c2]), filter(nil), rowset=16 |
| access([t1.__pk_increment], [t1.c1], [t1.c2]), partitions(p0) |
| is_index_back=true, is_global_index=false, |
| range_key([t1.c1], [t1.__pk_increment]), range(1,MIN ; 1,MAX), |
| range_cond([t1.c1 = 1]) |
| 2 - output([t1.c1], [t1.c2]), filter(nil), rowset=16 |
| 3 - output([t1.c1], [t1.c2]), filter(nil), rowset=16 |
| access([t1.__pk_increment], [t1.c1], [t1.c2]), partitions(p0) |
| is_index_back=true, is_global_index=false, |
| range_key([t1.c1], [t1.__pk_increment]), range(1,MIN ; 1,MAX), |
| range_cond([t1.c1 = 1]) |
+--------------------------------------------------------------------------+
23 rows in set
联接方法的 Hint 中指定的表如果找不到,则忽略该表,其他的指定依然生效;如果优化器不能生成指定的联接方法,就会选择其他方法,Hint 无效。
联接顺序的 Hint 中如果存在表无法找到的情况,则该 Hint 完全失效。
常用 Hint 及语法
与其他数据库的行为相比,OceanBase 数据库优化器是动态规划的,已经考虑了所有可能的最优路径,Hint 主要作用是指定优化器的行为,并按照 Hint 执行 SQL 查询。
INDEX Hint
INDEX Hint 的语法同时支持 MySQL 模式和 Oracle 模式。
INDEXHint 的 Oracle 语法如下:
SELECT/*+INDEX(table_name index_name) */ * FROM table_name;
INDEXHint 的 MySQL 语法如下:
table_name [[AS] alias] [index_hint_list]
index_hint_list:
index_hint [, index_hint] ...
index_hint:
USE {INDEX|KEY}
[FOR {JOIN|ORDER BY|GROUP BY}] ([index_list])
| IGNORE {INDEX|KEY}
[FOR {JOIN|ORDER BY|GROUP BY}] (index_list)
| FORCE {INDEX|KEY}
[FOR {JOIN|ORDER BY|GROUP BY}] (index_list)
index_list:
index_name [, index_name] ...
Oracle 模式的语法中一个表只能指定一个 INDEX,而 MySQL 模式的语法可以指定多个。但是 OceanBase 数据库中 MySQL 模式的语法虽然支持指定多个 INDEX,但是对于 USE 和 FORCE 方式,只会用第一个 INDEX 生成 PATH,即使 SQL 语句中没有该 INDEX 的 filter 而导致全部扫描同时回表(即 OceanBase 数据库当前设计是认为写 Hint 的人比程序更明白那条路径是更好的)。IGNORE 类型会忽略所有指定的 INDEX。USE、FORCE 方式和 Oracle 的Hint 方式实际是一样的,如果该方式的 INDEX 不存在或者处于 invalid 状态,则 Hint 无效。对于 IGNORE 方式,如果将包括主表在内的所有 INDEX 忽略,则 Hint 无效。
在 SQL 语句中,表名存在别名即 table_name [AS] alias,必须写表别名,才能使 INDEX 生效。示例如下:
obclient> create table t1(c1 int, c2 int, c3 int);
Query OK, 0 rows affected
obclient> create index idx1 on t1(c1);
Query OK, 0 rows affected
obclient> create index idx2 on t1(c2);
Query OK, 0 rows affected
obclient> insert into t1 with recursive cte(n) as (select 1 from dual union all select n+1 from cte where n < 1000) select n, mod(n, 3), n from cte;
Query OK, 1 row affected
obclient> analyze table t1 COMPUTE STATISTICS for all columns size 128;
Query OK, 0 rows affected
obclient> explain select * from t1 where c1 = 1 and c2 = 1;
+-----------------------------------------------------------------------------------+
| Query Plan |
+-----------------------------------------------------------------------------------+
| ==================================================== |
| |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| |
| ---------------------------------------------------- |
| |0 |TABLE RANGE SCAN|t1(idx1)|1 |7 | |
| ==================================================== |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([t1.c1], [t1.c2], [t1.c3]), filter([t1.c2 = 1]), rowset=16 |
| access([t1.__pk_increment], [t1.c1], [t1.c2], [t1.c3]), partitions(p0) |
| is_index_back=true, is_global_index=false, filter_before_indexback[false], |
| range_key([t1.c1], [t1.__pk_increment]), range(1,MIN ; 1,MAX), |
| range_cond([t1.c1 = 1]) |
+-----------------------------------------------------------------------------------+
12 rows in set
-------生效的 index
obclient> explain select /*+index(t idx2)*/ * from t1 t where c1 = 1 and c2 = 1;
+-----------------------------------------------------------------------------------+
| Query Plan |
+-----------------------------------------------------------------------------------+
| =================================================== |
| |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| |
| --------------------------------------------------- |
| |0 |TABLE RANGE SCAN|t(idx2)|1 |812 | |
| =================================================== |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([t.c1], [t.c2], [t.c3]), filter([t.c1 = 1]), rowset=16 |
| access([t.__pk_increment], [t.c1], [t.c2], [t.c3]), partitions(p0) |
| is_index_back=true, is_global_index=false, filter_before_indexback[false], |
| range_key([t.c2], [t.__pk_increment]), range(1,MIN ; 1,MAX), |
| range_cond([t.c2 = 1]) |
+-----------------------------------------------------------------------------------+
12 rows in set
-------不生效的 index
obclient> explain select /*+index(t1 idx2)*/ * from t1 t where c1 = 1 and c2 = 1;
+-----------------------------------------------------------------------------------+
| Query Plan |
+-----------------------------------------------------------------------------------+
| =================================================== |
| |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| |
| --------------------------------------------------- |
| |0 |TABLE RANGE SCAN|t(idx1)|1 |7 | |
| =================================================== |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([t.c1], [t.c2], [t.c3]), filter([t.c2 = 1]), rowset=16 |
| access([t.__pk_increment], [t.c1], [t.c2], [t.c3]), partitions(p0) |
| is_index_back=true, is_global_index=false, filter_before_indexback[false], |
| range_key([t.c1], [t.__pk_increment]), range(1,MIN ; 1,MAX), |
| range_cond([t.c1 = 1]) |
+-----------------------------------------------------------------------------------+
12 rows in set
根据示例结果,c1 = 1 过滤性优于 c2 = 1,INDEX Hint 不生效时使用优化器索引选择机制使用了 IDX1。
FULL Hint
FULL Hint 的语法是用于指定表使用主表扫描,语法如下:
/*+ FULL(table_name)*/
FULL Hint 用于为指定表选择主表扫描,等价于 INDEX Hint /*+ INDEX(table_name PRIMARY)*/。
ORDERED Hint
ORDERED Hint 可以指定按照 FROM 后面的表顺序作为联接顺序,语法如下:
/*+ ORDERED */
如果指定该 Hint 后发生 SQL 改写,那么就按照改写后的 stmt 中 From Items 的顺序联接,因为改写时候 sub_query 会在 From Items 中对应位置填放新的 Table Item。
LEADING Hint
LEADING Hint 可以指定表的联接顺序,语法如下:
/*+ LEADING(table_name_list)*/
在 table_name_list 中可以使用 () 表示内部各表的联接优先级,指定复杂联接。示例如下:
obclient> EXPLAIN BASIC SELECT /*+LEADING(d c b a)*/ * FROM t1 a, t1 b, t1 c, t1 d;
+---------------------------------------------------------------------------------------------------------------------------------------+
| Query Plan |
+---------------------------------------------------------------------------------------------------------------------------------------+
| ========================================= |
| |ID|OPERATOR |NAME| |
| ----------------------------------------- |
| |0 |NESTED-LOOP JOIN CARTESIAN | | |
| |1 |├─NESTED-LOOP JOIN CARTESIAN | | |
| |2 |│ ├─NESTED-LOOP JOIN CARTESIAN | | |
| |3 |│ │ ├─TABLE FULL SCAN |d | |
| |4 |│ │ └─MATERIAL | | |
| |5 |│ │ └─TABLE FULL SCAN |c | |
| |6 |│ └─MATERIAL | | |
| |7 |│ └─TABLE FULL SCAN |b | |
| |8 |└─MATERIAL | | |
| |9 | └─TABLE FULL SCAN |a | |
| ========================================= |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([a.c1], [a.c2], [a.c3], [b.c1], [b.c2], [b.c3], [c.c1], [c.c2], [c.c3], [d.c1], [d.c2], [d.c3]), filter(nil), rowset=256 |
| conds(nil), nl_params_(nil), use_batch=false |
| 1 - output([b.c1], [b.c2], [b.c3], [c.c1], [c.c2], [c.c3], [d.c1], [d.c2], [d.c3]), filter(nil), rowset=256 |
| conds(nil), nl_params_(nil), use_batch=false |
| 2 - output([c.c1], [c.c2], [c.c3], [d.c1], [d.c2], [d.c3]), filter(nil), rowset=256 |
| conds(nil), nl_params_(nil), use_batch=false |
| 3 - output([d.c1], [d.c2], [d.c3]), filter(nil), rowset=256 |
| access([d.c1], [d.c2], [d.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([d.__pk_increment]), range(MIN ; MAX)always true |
| 4 - output([c.c1], [c.c2], [c.c3]), filter(nil), rowset=256 |
| 5 - output([c.c1], [c.c2], [c.c3]), filter(nil), rowset=256 |
| access([c.c1], [c.c2], [c.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([c.__pk_increment]), range(MIN ; MAX)always true |
| 6 - output([b.c1], [b.c2], [b.c3]), filter(nil), rowset=256 |
| 7 - output([b.c1], [b.c2], [b.c3]), filter(nil), rowset=256 |
| access([b.c1], [b.c2], [b.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([b.__pk_increment]), range(MIN ; MAX)always true |
| 8 - output([a.c1], [a.c2], [a.c3]), filter(nil), rowset=256 |
| 9 - output([a.c1], [a.c2], [a.c3]), filter(nil), rowset=256 |
| access([a.c1], [a.c2], [a.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([a.__pk_increment]), range(MIN ; MAX)always true |
+---------------------------------------------------------------------------------------------------------------------------------------+
41 rows in set
obclient> EXPLAIN BASIC SELECT /*+LEADING((d c) (b a))*/ * FROM t1 a, t1 b, t1 c, t1 d;
+---------------------------------------------------------------------------------------------------------------------------------------+
| Query Plan |
+---------------------------------------------------------------------------------------------------------------------------------------+
| ========================================= |
| |ID|OPERATOR |NAME| |
| ----------------------------------------- |
| |0 |NESTED-LOOP JOIN CARTESIAN | | |
| |1 |├─NESTED-LOOP JOIN CARTESIAN | | |
| |2 |│ ├─TABLE FULL SCAN |d | |
| |3 |│ └─MATERIAL | | |
| |4 |│ └─TABLE FULL SCAN |c | |
| |5 |└─MATERIAL | | |
| |6 | └─NESTED-LOOP JOIN CARTESIAN | | |
| |7 | ├─TABLE FULL SCAN |b | |
| |8 | └─MATERIAL | | |
| |9 | └─TABLE FULL SCAN |a | |
| ========================================= |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([a.c1], [a.c2], [a.c3], [b.c1], [b.c2], [b.c3], [c.c1], [c.c2], [c.c3], [d.c1], [d.c2], [d.c3]), filter(nil), rowset=256 |
| conds(nil), nl_params_(nil), use_batch=false |
| 1 - output([c.c1], [c.c2], [c.c3], [d.c1], [d.c2], [d.c3]), filter(nil), rowset=256 |
| conds(nil), nl_params_(nil), use_batch=false |
| 2 - output([d.c1], [d.c2], [d.c3]), filter(nil), rowset=256 |
| access([d.c1], [d.c2], [d.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([d.__pk_increment]), range(MIN ; MAX)always true |
| 3 - output([c.c1], [c.c2], [c.c3]), filter(nil), rowset=256 |
| 4 - output([c.c1], [c.c2], [c.c3]), filter(nil), rowset=256 |
| access([c.c1], [c.c2], [c.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([c.__pk_increment]), range(MIN ; MAX)always true |
| 5 - output([a.c1], [a.c2], [a.c3], [b.c1], [b.c2], [b.c3]), filter(nil), rowset=256 |
| 6 - output([a.c1], [a.c2], [a.c3], [b.c1], [b.c2], [b.c3]), filter(nil), rowset=256 |
| conds(nil), nl_params_(nil), use_batch=false |
| 7 - output([b.c1], [b.c2], [b.c3]), filter(nil), rowset=256 |
| access([b.c1], [b.c2], [b.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([b.__pk_increment]), range(MIN ; MAX)always true |
| 8 - output([a.c1], [a.c2], [a.c3]), filter(nil), rowset=256 |
| 9 - output([a.c1], [a.c2], [a.c3]), filter(nil), rowset=256 |
| access([a.c1], [a.c2], [a.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([a.__pk_increment]), range(MIN ; MAX)always true |
+---------------------------------------------------------------------------------------------------------------------------------------+
41 rows in set
obclient> EXPLAIN BASIC SELECT /*+LEADING((d c b) a))*/ * FROM t1 a, t1 b, t1 c, t1 d;
+---------------------------------------------------------------------------------------------------------------------------------------+
| Query Plan |
+---------------------------------------------------------------------------------------------------------------------------------------+
| ========================================= |
| |ID|OPERATOR |NAME| |
| ----------------------------------------- |
| |0 |NESTED-LOOP JOIN CARTESIAN | | |
| |1 |├─NESTED-LOOP JOIN CARTESIAN | | |
| |2 |│ ├─NESTED-LOOP JOIN CARTESIAN | | |
| |3 |│ │ ├─TABLE FULL SCAN |d | |
| |4 |│ │ └─MATERIAL | | |
| |5 |│ │ └─TABLE FULL SCAN |c | |
| |6 |│ └─MATERIAL | | |
| |7 |│ └─TABLE FULL SCAN |b | |
| |8 |└─MATERIAL | | |
| |9 | └─TABLE FULL SCAN |a | |
| ========================================= |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([a.c1], [a.c2], [a.c3], [b.c1], [b.c2], [b.c3], [c.c1], [c.c2], [c.c3], [d.c1], [d.c2], [d.c3]), filter(nil), rowset=256 |
| conds(nil), nl_params_(nil), use_batch=false |
| 1 - output([b.c1], [b.c2], [b.c3], [c.c1], [c.c2], [c.c3], [d.c1], [d.c2], [d.c3]), filter(nil), rowset=256 |
| conds(nil), nl_params_(nil), use_batch=false |
| 2 - output([c.c1], [c.c2], [c.c3], [d.c1], [d.c2], [d.c3]), filter(nil), rowset=256 |
| conds(nil), nl_params_(nil), use_batch=false |
| 3 - output([d.c1], [d.c2], [d.c3]), filter(nil), rowset=256 |
| access([d.c1], [d.c2], [d.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([d.__pk_increment]), range(MIN ; MAX)always true |
| 4 - output([c.c1], [c.c2], [c.c3]), filter(nil), rowset=256 |
| 5 - output([c.c1], [c.c2], [c.c3]), filter(nil), rowset=256 |
| access([c.c1], [c.c2], [c.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([c.__pk_increment]), range(MIN ; MAX)always true |
| 6 - output([b.c1], [b.c2], [b.c3]), filter(nil), rowset=256 |
| 7 - output([b.c1], [b.c2], [b.c3]), filter(nil), rowset=256 |
| access([b.c1], [b.c2], [b.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([b.__pk_increment]), range(MIN ; MAX)always true |
| 8 - output([a.c1], [a.c2], [a.c3]), filter(nil), rowset=256 |
| 9 - output([a.c1], [a.c2], [a.c3]), filter(nil), rowset=256 |
| access([a.c1], [a.c2], [a.c3]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([a.__pk_increment]), range(MIN ; MAX)always true |
+---------------------------------------------------------------------------------------------------------------------------------------+
41 rows in set
为确保按照用户指定的顺序进行联接,LEADING Hint 的检查比较严格,如果发现 Hint 指定的 table_name 不存在,LEADING Hint 失效;如果发现 Hint 中存在重复表,LEADING Hint 失效。如果在 Optimizer 联接期间,按 table_id 无法在 From Items 中找到对应的表,则可能发生改写,那么该表及后面的表指定的 JOIN 顺序失效,该表前面的 JOIN 顺序依然有效。
USE_MERGE Hint
USE_MERGE 可以指定表在联接时候使用 Merge Join 算法,将所指定的表作为右表。语法如下:
/*+ USE_MERGE(table_name_list) */
注意
OceanBase 数据库中 Merge Join 必须有等值条件的 join-condition,因此无等值条件的两个表联接,use_merge 会无效。
关于 Merge Join 算法是否认为 A Merge Join B 等效于 B Merge Join A,当前并没有最后结论。按照代价模型,Merge Join 计算代价时是区分左右表的,这也增加了 Hint 的灵活性,所以当前 Merge Join 区分左右表,即 use_merge 仅对表作为右表的时候生效。
USE_NL Hint
和 join 相关的 Hint 的基本结构如下: join_hint_name ( @ qb_name table_name_list) 基本语义为,当联接的右表匹配 table_name_list 时,按照 Hint 语义生成计划。一般需要使用 LEADING Hint 指定联接顺序,使 table_name_list 中的表为右表,否则 Hint 会随着联接顺序变化而失效。
其中 table_name_list 可有以下形式:
- 单表 use_nl ( t1 ): 以 t1 为右表时使用 Nested Loop Join。
- 多个单表 use_nl ( t1 t2 ... ): 以 t1 或 t2 等为右表时使用 Nested Loop Join。
- 多表 use_nl ( (t1 t2) ): 以 t1 join t2 为右表时使用 Nested Loop Join,忽略 t1/t2 连接顺序和方法。
- 多组表 use_nl ( t1 (t2 t3) (t4 t5 t6) ... ): 以 t1 / t2 join t3 / t4 join t5 join t6 等表为右表时使用 Nested Loop Join。
USE_NL 指定表作为右表,在联接的时候使用 Nested Loop Join 算法,语法如下:
/*+ USE_NL(table_name_list) */
示例如下:
obclient> CREATE TABLE t0(c1 INT, c2 INT, c3 INT);
obclient> CREATE TABLE t1(c1 INT, c2 INT, c3 INT);
obclient> CREATE TABLE t2(c1 INT, c2 INT, c3 INT);
obclient> EXPLAIN EXTENDED SELECT /*+LEADING(t0 t1) USE_NL(t1)*/ *
FROM t0, t1 WHERE t0.c1 = t1.c1;
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Query Plan |
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| =================================================== |
| |ID|OPERATOR |NAME|EST.ROWS|EST.TIME(us)| |
| --------------------------------------------------- |
| |0 |NESTED-LOOP JOIN | |1 |3 | |
| |1 |├─TABLE FULL SCAN |t0 |1 |3 | |
| |2 |└─MATERIAL | |1 |3 | |
| |3 | └─TABLE FULL SCAN|t1 |1 |3 | |
| =================================================== |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([t0.c1(0x7f218dc21640)], [t0.c2(0x7f218dc21f00)], [t0.c3(0x7f218dc22230)], [t1.c1(0x7f218dc21980)], [t1.c2(0x7f218dc22560)], [t1.c3(0x7f218dc22890)]), filter(nil), rowset=16 |
| conds([t0.c1(0x7f218dc21640) = t1.c1(0x7f218dc21980)(0x7f218dc20e80)]), nl_params_(nil), use_batch=false |
| 1 - output([t0.c1(0x7f218dc21640)], [t0.c2(0x7f218dc21f00)], [t0.c3(0x7f218dc22230)]), filter(nil), rowset=16 |
| access([t0.c1(0x7f218dc21640)], [t0.c2(0x7f218dc21f00)], [t0.c3(0x7f218dc22230)]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([t0.__pk_increment(0x7f218dc230f0)]), range(MIN ; MAX)always true |
| 2 - output([t1.c1(0x7f218dc21980)], [t1.c2(0x7f218dc22560)], [t1.c3(0x7f218dc22890)]), filter(nil), rowset=16 |
| 3 - output([t1.c1(0x7f218dc21980)], [t1.c2(0x7f218dc22560)], [t1.c3(0x7f218dc22890)]), filter(nil), rowset=16 |
| access([t1.c1(0x7f218dc21980)], [t1.c2(0x7f218dc22560)], [t1.c3(0x7f218dc22890)]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([t1.__pk_increment(0x7f218dc23420)]), range(MIN ; MAX)always true |
| Used Hint: |
| ------------------------------------- |
| /*+ |
| |
| LEADING(("t0" "t1")) |
| USE_NL("t1") |
| */ |
| Qb name trace: |
| ------------------------------------- |
| stmt_id:0, stmt_type:T_EXPLAIN |
| stmt_id:1, SEL$1 |
| Outline Data: |
| ------------------------------------- |
| /*+ |
| BEGIN_OUTLINE_DATA |
| LEADING(@"SEL$1" ("cccc"."t0"@"SEL$1" "cccc"."t1"@"SEL$1")) |
| USE_NL(@"SEL$1" "cccc"."t1"@"SEL$1") |
| USE_NL_MATERIALIZATION(@"SEL$1" "cccc"."t1"@"SEL$1") |
| FULL(@"SEL$1" "cccc"."t0"@"SEL$1") |
| FULL(@"SEL$1" "cccc"."t1"@"SEL$1") |
| OPTIMIZER_FEATURES_ENABLE('4.3.0.0') |
| END_OUTLINE_DATA |
| */ |
| Optimization Info: |
| ------------------------------------- |
| t0: |
| table_rows:1 |
| physical_range_rows:1 |
| logical_range_rows:1 |
| index_back_rows:0 |
| output_rows:1 |
| table_dop:1 |
| dop_method:Table DOP |
| avaiable_index_name:[t0] |
| stats version:0 |
| dynamic sampling level:1 |
| estimation method:[DYNAMIC SAMPLING FULL] |
| t1: |
| table_rows:1 |
| physical_range_rows:1 |
| logical_range_rows:1 |
| index_back_rows:0 |
| output_rows:1 |
| table_dop:1 |
| dop_method:Table DOP |
| avaiable_index_name:[t1] |
| stats version:0 |
| dynamic sampling level:1 |
| estimation method:[DYNAMIC SAMPLING FULL] |
| Plan Type: |
| LOCAL |
| Note: |
| Degree of Parallelisim is 1 because of table property |
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
74 rows in set
obclient> EXPLAIN EXTENDED SELECT /*+LEADING(t0 (t1 t2)) USE_NL((t1 t2))*/ *
FROM t0, t1, t2 WHERE t0.c1 = t1.c1 AND t0.c1 = t2.c1;
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Query Plan |
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| ===================================================== |
| |ID|OPERATOR |NAME|EST.ROWS|EST.TIME(us)| |
| ----------------------------------------------------- |
| |0 |NESTED-LOOP JOIN | |1 |3 | |
| |1 |├─TABLE FULL SCAN |t0 |1 |3 | |
| |2 |└─MATERIAL | |1 |5 | |
| |3 | └─HASH JOIN | |1 |5 | |
| |4 | ├─TABLE FULL SCAN|t1 |1 |3 | |
| |5 | └─TABLE FULL SCAN|t2 |1 |3 | |
| ===================================================== |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([t0.c1(0x7f217d422a30)], [t0.c2(0x7f217d424700)], [t0.c3(0x7f217d424a30)], [t1.c1(0x7f217d422d70)], [t1.c2(0x7f217d424d60)], [t1.c3(0x7f217d425090)], |
| [t2.c1(0x7f217d424180)], [t2.c2(0x7f217d4253c0)], [t2.c3(0x7f217d4256f0)]), filter(nil), rowset=16 |
| conds([t0.c1(0x7f217d422a30) = t1.c1(0x7f217d422d70)(0x7f217d422270)]), nl_params_(nil), use_batch=false |
| 1 - output([t0.c1(0x7f217d422a30)], [t0.c2(0x7f217d424700)], [t0.c3(0x7f217d424a30)]), filter(nil), rowset=16 |
| access([t0.c1(0x7f217d422a30)], [t0.c2(0x7f217d424700)], [t0.c3(0x7f217d424a30)]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([t0.__pk_increment(0x7f217d426110)]), range(MIN ; MAX)always true |
| 2 - output([t1.c1(0x7f217d422d70)], [t1.c2(0x7f217d424d60)], [t1.c3(0x7f217d425090)], [t2.c1(0x7f217d424180)], [t2.c2(0x7f217d4253c0)], [t2.c3(0x7f217d4256f0)]), filter(nil), rowset=16 |
| 3 - output([t1.c1(0x7f217d422d70)], [t1.c2(0x7f217d424d60)], [t1.c3(0x7f217d425090)], [t2.c1(0x7f217d424180)], [t2.c2(0x7f217d4253c0)], [t2.c3(0x7f217d4256f0)]), filter(nil), rowset=16 |
| equal_conds([t1.c1(0x7f217d422d70) = t2.c1(0x7f217d424180)(0x7f217d4c2500)]), other_conds(nil) |
| 4 - output([t1.c1(0x7f217d422d70)], [t1.c2(0x7f217d424d60)], [t1.c3(0x7f217d425090)]), filter(nil), rowset=16 |
| access([t1.c1(0x7f217d422d70)], [t1.c2(0x7f217d424d60)], [t1.c3(0x7f217d425090)]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([t1.__pk_increment(0x7f217d426440)]), range(MIN ; MAX)always true |
| 5 - output([t2.c1(0x7f217d424180)], [t2.c2(0x7f217d4253c0)], [t2.c3(0x7f217d4256f0)]), filter(nil), rowset=16 |
| access([t2.c1(0x7f217d424180)], [t2.c2(0x7f217d4253c0)], [t2.c3(0x7f217d4256f0)]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([t2.__pk_increment(0x7f217d426770)]), range(MIN ; MAX)always true |
| Used Hint: |
| ------------------------------------- |
| /*+ |
| |
| LEADING(("t0" ("t1" "t2"))) |
| USE_NL(("t1" "t2")) |
| */ |
| Qb name trace: |
| ------------------------------------- |
| stmt_id:0, stmt_type:T_EXPLAIN |
| stmt_id:1, SEL$1 |
| Outline Data: |
| ------------------------------------- |
| /*+ |
| BEGIN_OUTLINE_DATA |
| LEADING(@"SEL$1" ("cccc"."t0"@"SEL$1" ("cccc"."t1"@"SEL$1" "cccc"."t2"@"SEL$1"))) |
| USE_NL(@"SEL$1" ("cccc"."t2"@"SEL$1" "cccc"."t1"@"SEL$1")) |
| USE_NL_MATERIALIZATION(@"SEL$1" ("cccc"."t2"@"SEL$1" "cccc"."t1"@"SEL$1")) |
| FULL(@"SEL$1" "cccc"."t0"@"SEL$1") |
| USE_HASH(@"SEL$1" "cccc"."t2"@"SEL$1") |
| FULL(@"SEL$1" "cccc"."t1"@"SEL$1") |
| FULL(@"SEL$1" "cccc"."t2"@"SEL$1") |
| OPTIMIZER_FEATURES_ENABLE('4.3.0.0') |
| END_OUTLINE_DATA |
| */ |
| Optimization Info: |
| ------------------------------------- |
| t0: |
| table_rows:1 |
| physical_range_rows:1 |
| logical_range_rows:1 |
| index_back_rows:0 |
| output_rows:1 |
| table_dop:1 |
| dop_method:Table DOP |
| avaiable_index_name:[t0] |
| stats version:0 |
| dynamic sampling level:1 |
| estimation method:[DYNAMIC SAMPLING FULL] |
| t1: |
| table_rows:1 |
| physical_range_rows:1 |
| logical_range_rows:1 |
| index_back_rows:0 |
| output_rows:1 |
| table_dop:1 |
| dop_method:Table DOP |
| avaiable_index_name:[t1] |
| stats version:0 |
| dynamic sampling level:1 |
| estimation method:[DYNAMIC SAMPLING FULL] |
| t2: |
| table_rows:1 |
| physical_range_rows:1 |
| logical_range_rows:1 |
| index_back_rows:0 |
| output_rows:1 |
| table_dop:1 |
| dop_method:Table DOP |
| avaiable_index_name:[t2] |
| stats version:0 |
| dynamic sampling level:1 |
| estimation method:[DYNAMIC SAMPLING FULL] |
| Plan Type: |
| LOCAL |
| Note: |
| Degree of Parallelisim is 1 because of table property |
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
97 rows in set
USE_HASH Hint
USE_HASH 指定表作为右表,在联接的时候使用 Hash Join 算法,语法如下:
/*+ USE_HASH(table_name_list) */
示例如下:
obclient> CREATE TABLE t0(c1 INT, c2 INT, c3 INT);
obclient> CREATE TABLE t1(c1 INT, c2 INT, c3 INT);
obclient> EXPLAIN EXTENDED SELECT /*+LEADING(t0 t1) USE_HASH(t1)*/ *
FROM t0, t1 WHERE t0.c1 = t1.c1;
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Query Plan |
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| ================================================= |
| |ID|OPERATOR |NAME|EST.ROWS|EST.TIME(us)| |
| ------------------------------------------------- |
| |0 |HASH JOIN | |1 |5 | |
| |1 |├─TABLE FULL SCAN|t0 |1 |3 | |
| |2 |└─TABLE FULL SCAN|t1 |1 |3 | |
| ================================================= |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([t0.c1(0x7f21a8421640)], [t0.c2(0x7f21a8421f00)], [t0.c3(0x7f21a8422230)], [t1.c1(0x7f21a8421980)], [t1.c2(0x7f21a8422560)], [t1.c3(0x7f21a8422890)]), filter(nil), rowset=16 |
| equal_conds([t0.c1(0x7f21a8421640) = t1.c1(0x7f21a8421980)(0x7f21a8420e80)]), other_conds(nil) |
| 1 - output([t0.c1(0x7f21a8421640)], [t0.c2(0x7f21a8421f00)], [t0.c3(0x7f21a8422230)]), filter(nil), rowset=16 |
| access([t0.c1(0x7f21a8421640)], [t0.c2(0x7f21a8421f00)], [t0.c3(0x7f21a8422230)]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([t0.__pk_increment(0x7f21a84230f0)]), range(MIN ; MAX)always true |
| 2 - output([t1.c1(0x7f21a8421980)], [t1.c2(0x7f21a8422560)], [t1.c3(0x7f21a8422890)]), filter(nil), rowset=16 |
| access([t1.c1(0x7f21a8421980)], [t1.c2(0x7f21a8422560)], [t1.c3(0x7f21a8422890)]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([t1.__pk_increment(0x7f21a8423420)]), range(MIN ; MAX)always true |
| Used Hint: |
| ------------------------------------- |
| /*+ |
| |
| LEADING(("t0" "t1")) |
| USE_HASH("t1") |
| */ |
| Qb name trace: |
| ------------------------------------- |
| stmt_id:0, stmt_type:T_EXPLAIN |
| stmt_id:1, SEL$1 |
| Outline Data: |
| ------------------------------------- |
| /*+ |
| BEGIN_OUTLINE_DATA |
| LEADING(@"SEL$1" ("dddd"."t0"@"SEL$1" "dddd"."t1"@"SEL$1")) |
| USE_HASH(@"SEL$1" "dddd"."t1"@"SEL$1") |
| FULL(@"SEL$1" "dddd"."t0"@"SEL$1") |
| FULL(@"SEL$1" "dddd"."t1"@"SEL$1") |
| OPTIMIZER_FEATURES_ENABLE('4.3.0.0') |
| END_OUTLINE_DATA |
| */ |
| Optimization Info: |
| ------------------------------------- |
| t0: |
| table_rows:1 |
| physical_range_rows:1 |
| logical_range_rows:1 |
| index_back_rows:0 |
| output_rows:1 |
| table_dop:1 |
| dop_method:Table DOP |
| avaiable_index_name:[t0] |
| stats version:0 |
| dynamic sampling level:1 |
| estimation method:[DYNAMIC SAMPLING FULL] |
| t1: |
| table_rows:1 |
| physical_range_rows:1 |
| logical_range_rows:1 |
| index_back_rows:0 |
| output_rows:1 |
| table_dop:1 |
| dop_method:Table DOP |
| avaiable_index_name:[t1] |
| stats version:0 |
| dynamic sampling level:1 |
| estimation method:[DYNAMIC SAMPLING FULL] |
| Plan Type: |
| LOCAL |
| Note: |
| Degree of Parallelisim is 1 because of table property |
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
71 rows in set
PARALLEL Hint
PARALLEL 指定语句级别的并行度。语法如下:
/*+ PARALLEL(n) */
其中 n 为整数,表示并行度。
OceanBase 数据库也支持表级别的 PARALLEL Hint,语法如下:
/*+ PARALLEL(table_name n) */
如果同时指定了全局并行度和表级并行度,则表级并行度不会生效。示例如下:
obclient> CREATE TABLE tbl1 (col1 INT) PARTITION BY HASH(col1) ;
Query OK, 0 rows affected
obclient> EXPLAIN SELECT /*+ PARALLEL(3) PARALLEL(tbl1 5) */ * FROM tbl1;
+----------------------------------------------------------------------+
| Query Plan |
+----------------------------------------------------------------------+
| ========================================================= |
| |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| |
| --------------------------------------------------------- |
| |0 |PX COORDINATOR | |1 |1 | |
| |1 |└─EXCHANGE OUT DISTR |:EX10000|1 |1 | |
| |2 | └─PX BLOCK ITERATOR| |1 |1 | |
| |3 | └─TABLE FULL SCAN|tbl1 |1 |1 | |
| ========================================================= |
| Outputs & filters: |
| ------------------------------------- |
| 0 - output([INTERNAL_FUNCTION(tbl1.col1)]), filter(nil), rowset=16 |
| 1 - output([INTERNAL_FUNCTION(tbl1.col1)]), filter(nil), rowset=16 |
| dop=5 |
| 2 - output([tbl1.col1]), filter(nil), rowset=16 |
| 3 - output([tbl1.col1]), filter(nil), rowset=16 |
| access([tbl1.col1]), partitions(p0) |
| is_index_back=false, is_global_index=false, |
| range_key([tbl1.__pk_increment]), range(MIN ; MAX)always true |
+----------------------------------------------------------------------+
18 rows in set
UNION_MERGE Hint
注意
OceanBase 数据库从 V4.4.1 版本开始移除 UNION_MERGE Hint,不再支持 UNION_MERGE Hint 语句。
UNION_MERGE Hint 用于指定查询计划中使用 Index Merge 的执行方式。当提供有效的 UNION_MERGE Hint 时,OceanBase 数据库会直接选用对应的 Index Merge 计划。
语法如下:
/*+ UNION_MERGE(table_name index_name_list) */
参数解释:
table_name:指定表的名称。index_name_list:指定索引名称列表,多个索引名称之间用空格分隔。
示例如下:
创建一个包含全文索引的表
tbl1。CREATE TABLE tbl1 ( col1 INT PRIMARY KEY, col2 INT, col3 VARCHAR(100), col4 VARCHAR(100), FULLTEXT INDEX ftidx3(col3), FULLTEXT INDEX ftidx4(col4) );使用
UNION_MERGEHint 查看执行计划。在查询中使用UNION_MERGEHint,指定表tbl1和索引ftidx3、ftidx4。EXPLAIN SELECT /*+UNION_MERGE(tbl1 ftidx3 ftidx4)*/ * FROM tbl1 WHERE col1 = 1 AND (MATCH(col3) AGAINST ("word1") OR MATCH(col4) AGAINST ("word1"));返回结果如下:
+----------------------------------------------------------------------------------------------------------------------------------------------------------------+ | Query Plan | +----------------------------------------------------------------------------------------------------------------------------------------------------------------+ | =========================================================================== | | |ID|OPERATOR |NAME |EST.ROWS|EST.TIME(us)| | | --------------------------------------------------------------------------- | | |0 |DISTRIBUTED INDEX MERGE SCAN|tbl1(ftidx3,ftidx4)|1 |45 | | | =========================================================================== | | Outputs & filters: | | ------------------------------------- | | 0 - output([tbl1.col1], [tbl1.col2], [tbl1.col3], [tbl1.col4]), filter([MATCH(tbl1.col3) AGAINST('word1') OR MATCH(tbl1.col4) AGAINST('word1')], [tbl1.col1 | | = 1]), rowset=16 | | access([tbl1.col1], [tbl1.col3], [tbl1.col4], [tbl1.col2]), partitions(p0) | | is_index_back=true, is_global_index=false, keep_ordering=true, use_index_merge=true, filter_before_indexback[false,false], | | index_name: ftidx3, range_cond(nil), filter(nil) | | index_name: ftidx4, range_cond(nil), filter(nil) | | lookup_filter([tbl1.col1 = 1], [MATCH(tbl1.col3) AGAINST('word1') OR MATCH(tbl1.col4) AGAINST('word1')]) | +----------------------------------------------------------------------------------------------------------------------------------------------------------------+ 14 rows in set返回结果说明:
DISTRIBUTED INDEX MERGE SCAN:表示分布式环境下使用了 Index Merge 的扫描方式。is_index_back=true:表示启用了回表操作。use_index_merge=true:表示启用了 Index Merge。filter_before_indexback[false,false]:表示过滤条件在回表操作之前未被应用。
PUSH_SUBQ Hint
PUSH_SUBQ Hint 用于指示优化器尽早执行未改写成连接的子查询,通常这些子查询如果不被改写成连接的话,会被放在执行计划的所有表连接计算之后。如果子查询计算成本低且能过滤大量数据,提前执行可能提升计划性能。但是要注意,如果子查询被改写成连接(/+unnest/),这个提示无效。适用场景如下:
- 子查询的执行开销较小,但能快速过滤大量数据。
- 需要尽早利用子查询的结果缩小数据范围。
语法如下:
/*+ PUSH_SUBQ[(@qb_name)] */
参数解释:
@qb_name:可选参数,指定子查询的别名(详细介绍信息,可参见上文 QB_NAME 参数),用于明确提示作用的对象。如果省略,则默认对 Hint 所处的子查询生效。
示例如下:
在查询中使用 PUSH_SUBQ Hint 提示优化器尽早执行子查询,以提前过滤 tbl1 表的数据。
SELECT /*+ PUSH_SUBQ(@"SEL$2") */ *
FROM tbl1, tbl2
WHERE tbl1.col1 = (SELECT MAX(tbl3.col1)
FROM tbl3
WHERE tbl3.col2 = tbl2.col2);
NO_PUSH_SUBQ Hint
NO_PUSH_SUBQ Hint 是 PUSH_SUBQ 的反向操作,用于优化器将未改写成连接的子查询放在最后执行。适用于子查询成本高或无法显著减少行数的情况。适用场景如下:
- 子查询的执行开销较大,或其结果对数据量的影响较小。
- 需要将子查询的执行延迟到其他过滤条件生效后再执行,以减少其输入数据量。
语法如下:
/*+ NO_PUSH_SUBQ[(@qb_name)] */
参数解释:
@qb_name:可选参数,指定子查询的别名。
示例如下:
在查询中使用 NO_PUSH_SUBQ Hint 提示优化器将子查询推迟到最后执行。
SELECT /*+ NO_PUSH_SUBQ */ *
FROM tbl1, tbl2
WHERE tbl1.col1 = (SELECT MAX(tbl3.col1)
FROM tbl3
WHERE tbl3.col2 = tbl2.col2);
OPT_PARAM Hint
OPT_PARAM Hint 用于指定在查询级别更新一些优化器相关的配置项/系统变量。
语法如下:
/*+ OPT_PARAM(parameter_name [,] parameter_value) */
参数解释:
parameter_name:指配置项或系统变量名。parameter_value:指需要指定的变量值。OPT_PARAMHint 对如下参数有效:rowsets_enabled: 开启/关闭向量化,数据类型是VARCHAR,取值范围为'TRUE'和'FALSE',取值需要用单引号('')包裹。rowsets_max_rows: 设置批量返回行数batch_size大小,数据类型是INT,取值范围为[0, 65535],取值不能用单引号('')包裹。enable_newsort: 允许查询中开启/关闭 newsort 优化,数据类型是VARCHAR,取值范围为'TRUE'和'FALSE',取值需要用单引号('')包裹。use_part_sort_mgb: 允许查询中开启/关闭 part sort merge group by,数据类型是VARCHAR,取值范围为'TRUE'和'FALSE',取值需要用单引号('')包裹。enable_in_range_optimization: 允许查询中开启/关闭 in 优化,数据类型是VARCHAR,取值范围为'TRUE'和'FALSE',取值需要用单引号('')包裹。xsolapi_generate_with_clause: 允许查询中开启/关闭 cte 抽取改写,数据类型是VARCHAR,取值范围为'TRUE'和'FALSE',取值需要用单引号('')包裹。preserve_order_for_pagination: 运行为分页查询添加order by保序/禁止添加order by,数据类型是VARCHAR,取值范围为'TRUE'和'FALSE',取值需要用单引号('')包裹。storage_card_estimation: 是否使用存储层估行,数据类型是VARCHAR,取值范围为'TRUE'和'FALSE',取值需要用单引号('')包裹。workarea_size_policy: 用于设置手动或者自动调整 SQL 工作区大小的策略。,数据类型是VARCHAR,取值范围为'MANUAL'表示手动调整和'AUTO'表示自动调整,取值需要用单引号('')包裹。enable_rich_vector_format: 用于开关向量化 2.0(session 级别参数),数据类型是VARCHAR,取值范围为'TRUE'和'FALSE',取值需要用单引号('')包裹。spill_compression_codec: 用于指定需要临时物化的算子选用的压缩算法,数据类型是VARCHAR,取值为NONE、LZ4、SNAPPY、ZLIB、ZSTD,代表不同的压缩算法。默认为NONE,表示不进行压缩。inlist_rewrite_threshold: 允许触发inlist改写成values statement的常量个数阈值,数据类型是INT64,取值范围为[1, 2147483647]。orc_filter_pushdown_level:表示控制语句级别 ORC 外表条件下压级别。parameter_value取值如下:说明
OceanBase 数据库
OPT_PARAMHint 中parameter_name参数从 V4.4.0 版本开始支持orc_filter_pushdown_level取值。0:表示禁用过滤条件下压。1:表示下压至文件级别。2:表示下压至 stripe 级别。3:表示下压至 row index 级别。4:表示下压到 Encoding 级别。
parquet_filter_pushdown_level:表示控制语句级别 Parquet 外表条件下压级别。parameter_value取值如下:说明
OceanBase 数据库
OPT_PARAMHint 中parameter_name参数从 V4.4.0 版本开始支持parquet_filter_pushdown_level取值。0:表示禁用过滤条件下压。1:表示下压至文件级别。2:表示下压至 RowGroup 级别。3:表示下压至 Page 级。4:表示下压至 Encoding 级别。
示例如下:
控制语句级别 ORC 外表条件下压级别。
关闭 ORC 过滤条件下压能力。
SELECT /*+opt_param('orc_filter_pushdown_level', '0') */ * FROM ext_orc_tbl WHERE c1 = 1;设置 ORC 条件下压到文件级别。
SELECT /*+opt_param('orc_filter_pushdown_level', '1') */ * FROM ext_orc_tbl WHERE c1 = 1;设置 ORC 条件下压到 stripe 级别。
SELECT /*+opt_param('orc_filter_pushdown_level', '2') */ * FROM ext_orc_tbl WHERE c1 = 1;设置 ORC 条件下压到 row index 级别。
SELECT /*+opt_param('orc_filter_pushdown_level', '3') */ * FROM ext_orc_tbl WHERE c1 = 1;设置 ORC 条件下压到 encoding 级别。
SELECT /*+opt_param('orc_filter_pushdown_level', '4') */ * FROM ext_orc_tbl WHERE c1 = 1;
控制语句级别 Parquet 外表条件下压级别。
关闭 Parquet 过滤条件下压能力。
SELECT /*+opt_param('parquet_filter_pushdown_level', '0') */ * FROM ext_parquet_tbl WHERE c1 = 1;设置 Parquet 条件下压到文件级别。
SELECT /*+opt_param('parquet_filter_pushdown_level', '1') */ * FROM ext_parquet_tbl WHERE c1 = 1;设置 Parquet 条件下压到 RowGroup 级别。
SELECT /*+opt_param('parquet_filter_pushdown_level', '2') */ * FROM ext_parquet_tbl WHERE c1 = 1;设置 Parquet 条件下压到 Page 级别。
SELECT /*+opt_param('parquet_filter_pushdown_level', '3') */ * FROM ext_parquet_tbl WHERE c1 = 1;设置 Parquet 条件下压到 Encoding 级别。
SELECT /*+opt_param('parquet_filter_pushdown_level', '4') */ * FROM ext_parquet_tbl WHERE c1 = 1;
INDEX_MERGE Hint
INDEX_MERGE Hint 用于控制指定的表使用 index merge 计划。指定的索引不能生成 index merge 计划时忽略 Hint,没有指定索引则尽可能尝试 index merge 计划。
使用 INDEX_MERGE Hint 可以在 _enable_index_merge 配置项值为 False 时生成 index merge 计划。
语法如下:
/*+ INDEX_MERGE([@query_block_name] tbl_name [index_name [,index_name]...]) */
参数解释:
@query_block_name:可选项,表示该 Hint 作用的查询块(query block) 名称。如果省略,则默认对当前查询的主查询块生效。tbl_name:表示表名。index_name:可选项,表示索引名称,可以指定多个索引名。- 当指定的表后没有
index_name参数的时候,表示控制指定的表使用 index merge 计划。 - 当指定的表后有
index_name参数的时候,表示控制指定的表使用指定的索引(Index)生成 index merge 计划。
- 当指定的表后没有
示例如下:
INDEX_MERGEHint 中指定索引名称。obclient> SELECT /*+ INDEX_MERGE(t1 idx_c2 idx_c3) */ * FROM t1 WHERE c2 = 1 AND c3 = 1 AND c4 = 1;INDEX_MERGEHint 中不指定索引名称。obclient> SELECT /*+ INDEX_MERGE(t1)*/ * FROM t1 WHERE c2 = 1 AND c3 = 1 AND c4 = 1;
NO_INDEX_MERGE Hint
NO_INDEX_MERGE Hint 用于控制指定的表不使用指定的索引生成 index merge 计划。没有指定索引时禁止该表生成 index merge 计划。
当 NO_INDEX_MERGE Hint 与 INDEX_MERGE Hint 发生冲突时,忽略这两个 Hint。
语法如下:
/*+ NO_INDEX_MERGE([@query_block_name] tbl_name [index_name [,index_name]...]) */
参数解释:
@query_block_name:可选项,表示该 Hint 作用的查询块(query block) 名称。如果省略,则默认对当前查询的主查询块生效。tbl_name:表示表名。index_name:可选项,表示索引名称,可以指定多个索引名。- 当指定的表后没有
index_name参数的时候,表示不使用 index merge 计划。 - 当指定的表后有
index_name参数的时候,表示不使用指定的索引(Index)来生成 index merge 计划。
- 当指定的表后没有
示例如下:
obclient> SELECT /*+ NO_INDEX_MERGE(t1)*/ * FROM t1
WHERE c2 = 1
AND c3 = 1
AND c4 = 1;