查看原文
其他

cost量化分析

xryz GreatSQL社区 2024-07-08

* GreatSQL使


前言:

我们在日常维护数据库的时候,经常会遇到查询慢的语句,这时候一般会通过执行EXPLAIN去查看它的执行计划,但是执行计划往往只给我们带来了最基础的分析信息,比如是否有使用索引,还有一些其他供我们分析的信息,比如使用了临时表、排序等等,却无法展示为什么一些其他的执行计划未被选择,比如说明明有索引,或者好几个索引,但是为什么查询时未使用到期望的索引等

explain select * from basic_person_info t1 join basic_person_info2 t2 on t1.id_num=t2.id_num where t1.age >10 and t2.age<20;
+----+-------------+-------+------------+--------+--------------------------------------+---------------+---------+----------------+------+----------+-----------------------+
| id | select_type | table | partitions | type   | possible_keys                        | key           | key_len | ref            | rows | filtered | Extra                 |
+----+-------------+-------+------------+--------+--------------------------------------+---------------+---------+----------------+------+----------+-----------------------+
|  1 | SIMPLE      | t2    | NULL       | range  | id_num_unique,idx_age,idx_age_id_num | idx_age       | 1       | NULL           | 9594 |   100.00 | Using index condition |
|  1 | SIMPLE      | t1    | NULL       | eq_ref | id_num_unique,idx_age                | id_num_unique | 60      | test.t2.id_num |    1 |    50.00 | Using where           |
+----+-------------+-------+------------+--------+--------------------------------------+---------------+---------+----------------+------+----------+-----------------------+
2 rows in set, 1 warning (0.01 sec)

如上面这个例子,为什么t2表上列出了多个可能使用的索引,却选择了idx_age,优化器为什么选择了指定的索引,这时候并不能直观的看出问题,这时候我们就可以开启optimizer_trace跟踪分析MySQL具体是怎么选择出最优的执行计划的。

OPTIMIZER_TRACE:

optimizer_trace是什么:

optimizer_trace是一个具有跟踪功能的工具,可以跟踪执行的语句的解析优化执行过程,并将跟踪到的信息记录到INFORMATION_SCHEMA.OPTIMIZER_TRACE表中,但是每个会话都只能跟踪它自己执行的语句,并且表中默认只记录最后一个查询的跟踪结果

使用方法:

# 打开optimizer trace功能 (默认情况下它是关闭的):
set optimizer_trace="enabled=on";
select ...; # 这里输入你自己的查询语句
SELECT * FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
# 当你停止查看语句的优化过程时,把optimizer trace功能关闭
set optimizer_trace="enabled=off";

相关参数:

mysql>  show variables like '%optimizer_trace%';
+------------------------------+----------------------------------------------------------------------------+
| Variable_name                | Value                                                                      |
+------------------------------+----------------------------------------------------------------------------+
| optimizer_trace              | enabled=on,one_line=off                                                    |
| optimizer_trace_features     | greedy_search=on,range_optimizer=on,dynamic_range=on,repeated_subselect=on |
| optimizer_trace_limit        | 1                                                                          |
| optimizer_trace_max_mem_size | 1048576                                                                    |
| optimizer_trace_offset       | -1                                                                         |
+------------------------------+----------------------------------------------------------------------------+
  • optimizer_trace: enabled 开启/关闭optimizer_trace,one_line 是否单行显示,关闭为json模式,一般不开启
  • optimizer_trace_features:跟踪信息中可打印的项,一般不调整默认打印所有项
  • optimizer_trace_limit:存储的跟踪sql条数
  • optimizer_trace_offset:开始记录的sql语句的偏移量,负数表示从最近执行倒数第几条开始记录
  • optimizer_trace_max_mem_size:optimizer_trace内存的大小,如果跟踪信息超过这个大小,信息将会被截断

optimizer_trace表信息:

该表总共有4个字段

  • QUERY 表示我们的查询语句。
  • TRACE 表示优化过程的JSON格式文本。(重点关注)
  • MISSING_BYTES_BEYOND_MAX_MEM_SIZE 由于优化过程可能会输出很多,如果超过某个限制时,多余的文本将不会被显示,这个字段展示了被忽略的文本字节数。
  • INSUFFICIENT_PRIVILEGES 表示是否没有权限查看优化过程,默认值是0,只有某些特殊情况下才会是 1,我们暂时不关心这个字段的值。

信息解读:

通过 optimizer_trace表的query字段可以看到,一条语句的执行过程主要分为三个步骤:

"join_preparation": {},(准备阶段)
"join_optimization": {},(优化阶段)
"join_execution": {},(执行阶段)

各个步骤的详细内容解读:

  • preparation:
expanded_query :将语句进行格式化,补充隐藏的列名和表名等
transformations_to_nested_joins :查询重写,比如join的on改为where语句
  • optimization:
condition_processing{ :条件句处理。
    transformation{:转换类型句。这三次的转换分别是
        equality_propagation(等值条件句转换),如:a = b and b = c and c = 5
        constant_propagation(常量条件句转换),如:a = 1 AND b > a
        trivial_condition_removal(无效条件移除的转换),如:1 = 1
    }
}
substitute_generated_columns :替换虚拟生成列,测试了很多sql,这一列都没有看到有用的信息
table_dependencies :梳理表之间的依赖关系。
ref_optimizer_key_uses :如果优化器认为查询可以使用ref的话,在这里列出可以使用的索引。
rows_estimation{ :估算表行数和扫描的代价。如果查询中存在range扫描的话,对range扫描进行计划分析及代价估算。
  table_scan:全表扫描的行数(rows)以及所需要的代价(cost)。
  potential_range_indexes:该阶段会列出表中所有的索引并分析其是否可用,并且还会列出索引中可用的列字段。
  analyzing_range_alternatives :分析可选方案的代价。
}
considered_execution_plans{ :对比各可行计划的代价,选择相对最优的执行计划。
  plan_prefix:前置的执行计划。
  best_access_path:当前最优的执行顺序信息结果集。
  access_type表示使用索引的方式,可参照为explain中的type字段。
  condition_filtering_pct:类似于explain中的filtered列,这是一个估算值。
  rows_for_plan:该执行计划最终的扫描行数,这里的行数其实也是估算值,是由considered_access_paths的resulting_rows相乘之后再乘以condition_filtering_pct获得。
  cost_for_plan:该执行计划的执行代价,由considered_access_paths的cost相加而得。
  chosen:是否选择了该执行计划。
}
attaching_conditions_to_tables :添加附加条件,使得条件尽可能筛选单表数据。
refine_plan :优化后的执行计划。

trace信息中的json信息很长,因为我们关心的是不同执行计划的cost区别,所以只需要重点关注两个部分rows_estimation 和considered_execution_plans

代价模型计算:

统计信息和cost计算参数:

计算cost会涉及到表的主键索引数据页(聚簇索引)数量和表中的记录数,两个信息都可以通过innodb的表统计信息mysql.innodb_table_stats查到,n_rows是记录数,clustered_index_size是聚簇索引页数。

mysql> select * from mysql.innodb_table_stats where table_name='basic_person_info';
+---------------+-------------------+---------------------+--------+----------------------+--------------------------+
| database_name | table_name        | last_update         | n_rows | clustered_index_size | sum_of_other_index_sizes |
+---------------+-------------------+---------------------+--------+----------------------+--------------------------+
test          | basic_person_info | 2022-12-23 18:27:24 |  86632 |                  737 |                     1401 |
+---------------+-------------------+---------------------+--------+----------------------+--------------------------+
1 row in set (0.01 sec)

代价模型将操作分为Server层和Engine(存储引擎)层两类,Server层主要是CPU代价,Engine层主要是IO代价,比如MySQL从磁盘读取一个数据页的代价io_block_read_cost为1,从buffer pool读取的代价memory_block_read_cost为0.25,计算符合条件的行代价为row_evaluate_cost为0.1,除此之外还有:

  • memory_temptable_create_cost (default 1.0) 内存临时表的创建代价。
  • memory_temptable_row_cost (default 0.1) 内存临时表的行代价。
  • key_compare_cost (default 0.1) 键比较的代价,例如排序。
  • disk_temptable_create_cost (default 20.0) 内部myisam或innodb临时表的创建代价。
  • disk_temptable_row_cost (default 0.5) 内部myisam或innodb临时表的行代价。

这些都可以通过mysql.server_cost、mysql.engine_cost查看defalt值和设置值

mysql> select * from mysql.server_cost;
+------------------------------+------------+---------------------+---------+---------------+
| cost_name                    | cost_value | last_update         | comment | default_value |
+------------------------------+------------+---------------------+---------+---------------+
| disk_temptable_create_cost   |       NULL | 2022-05-11 16:09:37 | NULL    |            20 |
| disk_temptable_row_cost      |       NULL | 2022-05-11 16:09:37 | NULL    |           0.5 |
| key_compare_cost             |       NULL | 2022-05-11 16:09:37 | NULL    |          0.05 |
| memory_temptable_create_cost |       NULL | 2022-05-11 16:09:37 | NULL    |             1 |
| memory_temptable_row_cost    |       NULL | 2022-05-11 16:09:37 | NULL    |           0.1 |
| row_evaluate_cost            |       NULL | 2022-05-11 16:09:37 | NULL    |           0.1 |
+------------------------------+------------+---------------------+---------+---------------+
mysql> select * from mysql.engine_cost;
+-------------+-------------+------------------------+------------+---------------------+---------+---------------+
| engine_name | device_type | cost_name              | cost_value | last_update         | comment | default_value |
+-------------+-------------+------------------------+------------+---------------------+---------+---------------+
| default     |           0 | io_block_read_cost     |       NULL | 2022-05-11 16:09:37 | NULL    |             1 |
| default     |           0 | memory_block_read_cost |       NULL | 2023-01-09 11:17:39 | NULL    |          0.25 |
+-------------+-------------+------------------------+------------+---------------------+---------+---------------+

计算公式:

如上面介绍的一样,代价模型将操作分为两类io_cost和cpu_cost,io_cost+cpu_cost就是总的cost,下面是具体的计算方法:

全表扫描:

全表扫描成本 = io_cost + 1.1 + cpu_cost + 1  (io_cost +1.1和cpu_cost +1在代码里是直接硬加上的,不知道为什么,计算的时候直接加上)

io_cost = clustered_index_size (统计信息中的主键页数) * avg_single_page_cost(读取一个页的平均成本)

avg_single_page_cost = pages_in_memory_percent * 0.25(memory_block_read_cost) + pages_on_disk_percent * 1.0(io_block_read_cost)

pages_in_memory_percent 表示已经加载到 Buffer Pool 中的叶结点占所有叶结点的比例 pages_on_disk_percent 表示没有加载到 Buffer Pool 中的叶结点占所有叶结点的比例

所以当数据已经全部读取到buffer pool中的时候:

io_cost=clustered_index_size * 0.25

都没有读取到buffer pool中的时候:

io_cost=clustered_index_size * 1.0

当部分数据在buffer pool中,部分数据需要从磁盘读取时,这时的系数介于0.25到1之间

cpu_cost = n_rows(统计信息中记录数) * 0.1(row_evaluate_cost)

走索引的成本:

和全表扫描的计算方法类似,其中io_cost与搜索的区间数有关,比如扫描三个区间where a between 1 and 10  or  a between 20 and 30 or a between 40 and 50,此时:

io_cost=3 * avg_single_page_cost

cpu_cost=记录数 * 0.1(row_evaluate_cost)+0.01(代码中的微调参数)

针对二级索引还会有回表的操作:

MySQL认为每次回表都相当于是访问一个页面,所以每次回表都会进行一次IO,这部分成本:

io_cost=rows(记录数)*avg_single_page_cost

对回表查询的数据还需要进行一次计算:

cpu_cost=rows(记录数) *  0.1(row_evaluate_cost)(需要注意的是当索引需要回表扫描时,在rows_estimation阶段并不会计算这个值,在considered_execution_plans阶段会重新加上这部分成本)

所以针对需要回表的查询:

io_cost=查询区间 * avg_single_page_cost + rows(记录数) * avg_single_page_cost

cpu_cost=记录数 * 0.1(row_evaluate_cost) + 0.01(代码中的微调参数) + rows(记录数) * 0.1(row_evaluate_cost)

例子:

mysql> set optimizer_trace='enabled=on';
Query OK, 0 rows affected (0.00 sec)

mysql>explain select * from basic_person_info t1 join basic_person_info2 t2 on t1.id_num=t2.id_num where t1.age >10 and t2.age<20;
+----+-------------+-------+------------+--------+--------------------------------------+---------------+---------+----------------+------+----------+-----------------------+
| id | select_type | table | partitions | type   | possible_keys                        | key           | key_len | ref            | rows | filtered | Extra                 |
+----+-------------+-------+------------+--------+--------------------------------------+---------------+---------+----------------+------+----------+-----------------------+
|  1 | SIMPLE      | t2    | NULL       | range  | id_num_unique,idx_age,idx_age_id_num | idx_age       | 1       | NULL           | 9594 |   100.00 | Using index condition |
|  1 | SIMPLE      | t1    | NULL       | eq_ref | id_num_unique,idx_age                | id_num_unique | 60      | test.t2.id_num |    1 |    50.00 | Using where           |
+----+-------------+-------+------------+--------+--------------------------------------+---------------+---------+----------------+------+----------+-----------------------+
2 rows in set, 1 warning (0.04 sec)

查看optimizer_trace的内容

select * from basic_person_info t1 join basic_person_info2 t2 on t1.id_num=t2.id_num where t1.age >10 and t2.age<20 | {
  "steps": [
    {
      "join_preparation": {
        "select#": 1,
        "steps": [
          {
            "expanded_query""/* select#1 */ select `t1`.`id` AS `id`,`t1`.`id_num` AS `id_num`,`t1`.`lastname` AS `lastname`,`t1`.`firstname` AS `firstname`,`t1`.`mobile` AS `mobile`,`t1`.`sex` AS `sex`,`t1`.`birthday` AS `birthday`,`t1`.`age` AS `age`,`t1`.`top_education` AS `top_education`,`t1`.`address` AS `address`,`t1`.`income_by_year` AS `income_by_year`,`t1`.`create_time` AS `create_time`,`t1`.`update_time` AS `update_time`,`t2`.`id` AS `id`,`t2`.`id_num` AS `id_num`,`t2`.`lastname` AS `lastname`,`t2`.`firstname` AS `firstname`,`t2`.`mobile` AS `mobile`,`t2`.`sex` AS `sex`,`t2`.`birthday` AS `birthday`,`t2`.`age` AS `age`,`t2`.`top_education` AS `top_education`,`t2`.`address` AS `address`,`t2`.`income_by_year` AS `income_by_year`,`t2`.`create_time` AS `create_time`,`t2`.`update_time` AS `update_time` from (`basic_person_info` `t1` join `basic_person_info2` `t2` on((`t1`.`id_num` = `t2`.`id_num`))) where ((`t1`.`age` > 10) and (`t2`.`age` < 20))"
          },
          {
            "transformations_to_nested_joins": {
              "transformations": [
                "JOIN_condition_to_WHERE",
                "parenthesis_removal"
              ],
              "expanded_query""/* select#1 */ select `t1`.`id` AS `id`,`t1`.`id_num` AS `id_num`,`t1`.`lastname` AS `lastname`,`t1`.`firstname` AS `firstname`,`t1`.`mobile` AS `mobile`,`t1`.`sex` AS `sex`,`t1`.`birthday` AS `birthday`,`t1`.`age` AS `age`,`t1`.`top_education` AS `top_education`,`t1`.`address` AS `address`,`t1`.`income_by_year` AS `income_by_year`,`t1`.`create_time` AS `create_time`,`t1`.`update_time` AS `update_time`,`t2`.`id` AS `id`,`t2`.`id_num` AS `id_num`,`t2`.`lastname` AS `lastname`,`t2`.`firstname` AS `firstname`,`t2`.`mobile` AS `mobile`,`t2`.`sex` AS `sex`,`t2`.`birthday` AS `birthday`,`t2`.`age` AS `age`,`t2`.`top_education` AS `top_education`,`t2`.`address` AS `address`,`t2`.`income_by_year` AS `income_by_year`,`t2`.`create_time` AS `create_time`,`t2`.`update_time` AS `update_time` from `basic_person_info` `t1` join `basic_person_info2` `t2` where ((`t1`.`age` > 10) and (`t2`.`age` < 20) and (`t1`.`id_num` = `t2`.`id_num`))"
            }
          }
        ]
      }
    },
    {
      "join_optimization": {
        "select#": 1,
        "steps": [
          {
            "condition_processing": {
              "condition""WHERE",
              "original_condition""((`t1`.`age` > 10) and (`t2`.`age` < 20) and (`t1`.`id_num` = `t2`.`id_num`))",
              "steps": [
                {
                  "transformation""equality_propagation",
                  "resulting_condition""((`t1`.`age` > 10) and (`t2`.`age` < 20) and multiple equal(`t1`.`id_num`, `t2`.`id_num`))"
                },
                {
                  "transformation""constant_propagation",
                  "resulting_condition""((`t1`.`age` > 10) and (`t2`.`age` < 20) and multiple equal(`t1`.`id_num`, `t2`.`id_num`))"
                },
                {
                  "transformation""trivial_condition_removal",
                  "resulting_condition""((`t1`.`age` > 10) and (`t2`.`age` < 20) and multiple equal(`t1`.`id_num`, `t2`.`id_num`))"
                }
              ]
            }
          },
          {
            "substitute_generated_columns": {
            }
          },
          {
            "table_dependencies": [
              {
                "table""`basic_person_info` `t1`",
                "row_may_be_null"false,
                "map_bit": 0,
                "depends_on_map_bits": [
                ]
              },
              {
                "table""`basic_person_info2` `t2`",
                "row_may_be_null"false,
                "map_bit": 1,
                "depends_on_map_bits": [
                ]
              }
            ]
          },
          {
            "ref_optimizer_key_uses": [
              {
                "table""`basic_person_info` `t1`",
                "field""id_num",
                "equals""`t2`.`id_num`",
                "null_rejecting"true
              },
              {
                "table""`basic_person_info2` `t2`",
                "field""id_num",
                "equals""`t1`.`id_num`",
                "null_rejecting"true
              }
            ]
          },
          {
            "rows_estimation": [
              {
                "table""`basic_person_info` `t1`",
                "range_analysis": {
                  "table_scan": {
                    "rows": 86734,
                    "cost": 8859.75
                    
                    t1表的scan成本=聚簇索引页数*0.25 + 行数 * 0.1 +1.1+1
                    737*0.25+1.1+86734*0.1+1=8859.75
                    
                  },
                  "potential_range_indexes": [
                    {
                      "index""PRIMARY",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""id_num_unique",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""mobile_unique",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""idx_basic_person_info_name",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""idx_basic_person_info_top_education",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""idx_basic_person_info_create_time",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""idx_basic_person_info_mobile",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""idx_age",
                      "usable"true,
                      "key_parts": [
                        "age",
                        "id"
                      ]
                    }
                  ],
                  "setup_range_conditions": [
                  ],
                  "group_index_range": {
                    "chosen"false,
                    "cause""not_single_table"
                  },
                  "skip_scan_range": {
                    "chosen"false,
                    "cause""not_single_table"
                  },
                  "analyzing_range_alternatives": {
                    "range_scan_alternatives": [
                      {
                        "index""idx_age",
                        "ranges": [
                          "10 < age"
                        ],
                        "index_dives_for_eq_ranges"true,
                        "rowid_ordered"false,
                        "using_mrr"false,
                        "index_only"false,
                        "rows": 43367,
                        "cost": 15178.7,
                        
                        通过索引idx_age读取数据:
                        io_cost=区间数* 0.25 +记录数* 0.25
                        io_cost=1*0.25+43367*0.25=10,842  
                        cpu_cost=记录数* 0.1 (没有回表的cost)
                        cpu_cost=43367*0.1=4,336.7 
                        cost=10842+4,336.7=15178.7
                        
                        "chosen"false,
                        "cause""cost"
                      }
                    ],
                    "analyzing_roworder_intersect": {
                      "usable"false,
                      "cause""too_few_roworder_scans"
                    }
                  }
                }
              },
              {
                "table""`basic_person_info2` `t2`",
                "range_analysis": {
                  "table_scan": {
                    "rows": 73845,
                    "cost": 7538.85
                    
                    t2表的scan成本=聚簇索引页数*0.25 + 行数 * 0.1 +1.1+1
                    609*0.25+1+73845*0.1+1.1=7538.85
                    
                  },
                  "potential_range_indexes": [
                    {
                      "index""PRIMARY",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""id_num_unique",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""mobile_unique",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""idx_basic_person_info_name",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""idx_basic_person_info_top_education",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""idx_basic_person_info_create_time",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""idx_basic_person_info_mobile",
                      "usable"false,
                      "cause""not_applicable"
                    },
                    {
                      "index""idx_age",
                      "usable"true,
                      "key_parts": [
                        "age",
                        "id"
                      ]
                    },
                    {
                      "index""idx_age_id_num",
                      "usable"true,
                      "key_parts": [
                        "age",
                        "id_num",
                        "id"
                      ]
                    }
                  ],
                  "setup_range_conditions": [
                  ],
                  "group_index_range": {
                    "chosen"false,
                    "cause""not_single_table"
                  },
                  "skip_scan_range": {
                    "chosen"false,
                    "cause""not_single_table"
                  },
                  "analyzing_range_alternatives": {
                    "range_scan_alternatives": [
                      {
                        "index""idx_age",
                        "ranges": [
                          "age < 20"
                        ],
                        "index_dives_for_eq_ranges"true,
                        "rowid_ordered"false,
                        "using_mrr"false,
                        "index_only"false,
                        "rows": 9594,
                        "cost": 3358.16,
                        
                        通过索引idx_age读取数据:
                        io_cost=区间数* 0.25 +记录数* 0.25
                        io_cost=1*0.25+9594*0.25=2,398.75        
                        cpu_cost=记录数* 0.1   (没有回表的cost) 
                        cpu_cost=9594*0.1959.4  
                        cost=2,398.75+959.4=3,358.15
                        
                        "chosen"true
                      },
                      {
                        "index""idx_age_id_num",
                        "ranges": [
                          "age < 20"
                        ],
                        "index_dives_for_eq_ranges"true,
                        "rowid_ordered"false,
                        "using_mrr"false,
                        "index_only"false,
                        "rows": 19086,
                        "cost": 6680.36,
                        
                        通过索引idx_age_id_num读取数据:
                        io_cost=区间数* 0.25 +记录数* 0.25
                        io_cost=1*0.25+19086*0.25=4,771.75           
                        cpu_cost=记录数* 0.1  (没有回表的cost)  
                        cpu_cost=19086*0.1=1908.6
                        cost=4,771.75+1908.6=6,680.35
                        
                        "chosen"false,
                        "cause""cost"
                      }
                    ],
                    "analyzing_roworder_intersect": {
                      "usable"false,
                      "cause""too_few_roworder_scans"
                    }
                  },
                  "chosen_range_access_summary": {
                    "range_access_plan": {
                      "type""range_scan",
                      "index""idx_age",
                      "rows": 9594,
                      "ranges": [
                        "age < 20"
                      ]
                    },
                    "rows_for_plan": 9594,
                    "cost_for_plan": 3358.16,
                    "chosen"true
                  }
                }
              }
            ]
          },
          {
            "considered_execution_plans": [
              {
                "plan_prefix": [
                ],
                "table""`basic_person_info2` `t2`",
                "best_access_path": {
                  "considered_access_paths": [
                    {
                      "access_type""ref",
                      "index""id_num_unique",
                      "usable"false,
                      "chosen"false
                    },
                    {
                      "rows_to_scan": 9594,
                      "filtering_effect": [
                      ],
                      "final_filtering_effect": 1,
                      "access_type""range",
                      "range_details": {
                        "used_index""idx_age"
                      },
                      "resulting_rows": 9594,
                      "cost": 4317.56,
                      
                        通过索引idx_age读取数据:
                        io_cost=区间数* 0.25 +记录数* 0.25
                        io_cost=1*0.25+9594*0.25=2,398.75        
                        cpu_cost=记录数* 0.1 + 记录数* 0.1   
                        cpu_cost=9594*0.1*2=1,918.8  
                        cost=2,398.75+1,918.8=4317.56
                      
                      "chosen"true
                    }
                  ]
                },
                "condition_filtering_pct": 100,
                "rows_for_plan": 9594,
                "cost_for_plan": 4317.56,
                "rest_of_plan": [
                  {
                    "plan_prefix": [
                      "`basic_person_info2` `t2`"
                    ],
                    "table""`basic_person_info` `t1`",
                    "best_access_path": {
                      "considered_access_paths": [
                        {
                          "access_type""eq_ref",
                          "index""id_num_unique",
                          "rows": 1,
                          "cost": 3357.9,
                          
                          io_cost=t2表记录数*0.25=9594*0.25=2398.5
                          cpu_cost=记录数*0.1=9594*0.1=959.4
                          cost=2398.5+959.4=3357.9
                          
                          "chosen"true
                        },
                        {
                          "rows_to_scan": 86734,
                          "filtering_effect": [
                          ],
                          "final_filtering_effect": 0.5,
                          "access_type""scan",
                          "using_join_cache"true,
                          "buffers_needed": 14,
                          "resulting_rows": 43367,
                          "cost": 4.16701e+07,
                          "chosen"false
                        }
                      ]
                    },
                    "condition_filtering_pct": 100,
                    "rows_for_plan": 9594,
                    "cost_for_plan": 7675.46,
                    
                   总cost=4,317.56+3,357.9=7,675.46
                   
                    "chosen"true
                  }
                ]
              },
              {
                "plan_prefix": [
                ],
                "table""`basic_person_info` `t1`",
                "best_access_path": {
                  "considered_access_paths": [
                    {
                      "access_type""ref",
                      "index""id_num_unique",
                      "usable"false,
                      "chosen"false
                    },
                    {
                      "rows_to_scan": 86734,
                      "filtering_effect": [
                      ],
                      "final_filtering_effect": 0.5,
                      "access_type""scan",
                      "resulting_rows": 43367,
                      "cost": 8857.65,
                      
                      t1的scan成本
                      
                      "chosen"true
                    }
                  ]
                },
                "condition_filtering_pct": 100,
                "rows_for_plan": 43367,
                "cost_for_plan": 8857.65,
                "pruned_by_cost"true
                
                放弃后续的计算
                
              }
            ]
          },
          {
            "attaching_conditions_to_tables": {
              "original_condition""((`t1`.`id_num` = `t2`.`id_num`) and (`t1`.`age` > 10) and (`t2`.`age` < 20))",
              "attached_conditions_computation": [
              ],
              "attached_conditions_summary": [
                {
                  "table""`basic_person_info2` `t2`",
                  "attached""(`t2`.`age` < 20)"
                },
                {
                  "table""`basic_person_info` `t1`",
                  "attached""((`t1`.`id_num` = `t2`.`id_num`) and (`t1`.`age` > 10))"
                }
              ]
            }
          },
          {
            "finalizing_table_conditions": [
              {
                "table""`basic_person_info2` `t2`",
                "original_table_condition""(`t2`.`age` < 20)",
                "final_table_condition   ""(`t2`.`age` < 20)"
              },
              {
                "table""`basic_person_info` `t1`",
                "original_table_condition""((`t1`.`id_num` = `t2`.`id_num`) and (`t1`.`age` > 10))",
                "final_table_condition   ""(`t1`.`age` > 10)"
              }
            ]
          },
          {
            "refine_plan": [
              {
                "table""`basic_person_info2` `t2`",
                "pushed_index_condition""(`t2`.`age` < 20)",
                "table_condition_attached": null
              },
              {
                "table""`basic_person_info` `t1`"
              }
            ]
          }
        ]
      }
    },
    {
      "join_execution": {
        "select#": 1,
        "steps": [
        ]
      }
    }
  ]
}

成本常数修改:

前面已经介绍了成本常量值实际上存放在MySQL自带的系统库MySQL中的server_cost和engine_cost表中,其中server_cost表存放server层的成本常量,engine_cost表存放engine层成本常量

mysql> select * from mysql.server_cost;
+------------------------------+------------+---------------------+---------+---------------+
| cost_name                    | cost_value | last_update         | comment | default_value |
+------------------------------+------------+---------------------+---------+---------------+
| disk_temptable_create_cost   |       NULL | 2022-05-11 16:09:37 | NULL    |            20 |
| disk_temptable_row_cost      |       NULL | 2022-05-11 16:09:37 | NULL    |           0.5 |
| key_compare_cost             |       NULL | 2022-05-11 16:09:37 | NULL    |          0.05 |
| memory_temptable_create_cost |       NULL | 2022-05-11 16:09:37 | NULL    |             1 |
| memory_temptable_row_cost    |       NULL | 2022-05-11 16:09:37 | NULL    |           0.1 |
| row_evaluate_cost            |       NULL | 2022-05-11 16:09:37 | NULL    |           0.1 |
+------------------------------+------------+---------------------+---------+---------------+

mysql> select * from mysql.engine_cost;
+-------------+-------------+------------------------+------------+---------------------+---------+---------------+
| engine_name | device_type | cost_name              | cost_value | last_update         | comment | default_value |
+-------------+-------------+------------------------+------------+---------------------+---------+---------------+
| default     |           0 | io_block_read_cost     |       NULL | 2022-05-11 16:09:37 | NULL    |             1 |
| default     |           0 | memory_block_read_cost |       NULL | 2023-01-09 11:17:39 | NULL    |          0.25 |
+-------------+-------------+------------------------+------------+---------------------+---------+---------------+

其中 default_value的值是系统默认的,不能修改,cost_value列的值我们可以修改,如果cost_value列的值不为空系统将用该值覆盖默认值,我们可以通过update语句来修改

mysql> update mysql.engine_cost set cost_value=10 where cost_name='memory_block_read_cost';
Query OK, 0 rows affected (0.00 sec)
mysql> update mysql.engine_cost set cost_value=10 where cost_name='io_block_read_cost';
Query OK, 0 rows affected (0.00 sec)

很多资料都说执行flush optimizer_costs就可以生效,不过我在修改完后并执行flush optimizer_costs并不能马上生效,最后是通过重启数据库实例才生效,这个可能是数据库版本的差异,大家可以自行验证。

mysql> explain select * from basic_person_info t1 join basic_person_info2 t2 on t1.id_num=t2.id_num where t1.age >10 and t2.age<20;
+----+-------------+-------+------------+--------+--------------------------------------+---------------+---------+----------------+-------+----------+-------------+
| id | select_type | table | partitions | type   | possible_keys                        | key           | key_len | ref            | rows  | filtered | Extra       |
+----+-------------+-------+------------+--------+--------------------------------------+---------------+---------+----------------+-------+----------+-------------+
|  1 | SIMPLE      | t2    | NULL       | ALL    | id_num_unique,idx_age,idx_age_id_num | NULL          | NULL    | NULL           | 73990 |    12.97 | Using where |
|  1 | SIMPLE      | t1    | NULL       | eq_ref | id_num_unique,idx_age                | id_num_unique | 60      | test.t2.id_num |     1 |    50.00 | Using where |
+----+-------------+-------+------------+--------+--------------------------------------+---------------+---------+----------------+-------+----------+-------------+

"table""`basic_person_info2` `t2`",
                "range_analysis": {
                  "table_scan": {
                    "rows": 73990,
                    "cost": 13491.1
                   
                   全表扫描cost=609*10+73990*0.1+1.1+1= 13491.1
                   
                  },
"index""idx_age",
                        "ranges": [
                          "age < 20"
                        ],
                        "index_dives_for_eq_ranges"true,
                        "rowid_ordered"false,
                        "using_mrr"false,
                        "index_only"false,
                        "rows": 9594,
                        "cost": 96909.4,
                        
                        idx_age索引扫描cost=1*10+9594*10+9594*0.1=96,909.4
                        
                        "chosen"false,
                        "cause""cost"
                      },

修改后的执行计划,发现t2表走了全表扫描了而没有走idx_age索引,分别查看一下t2表走全表扫描和idx_age索引的cost发现全表扫描的cost为13491.1,而走索引的cost为96,909.4,因为全表扫描的cost比走索引低,所以优化器没有选择idx_age索引。

从这个例子可以看出,更改成本常量值会直接影响优化器的方案选择,所以一定要慎重,没有特殊原因建议不要修改。

explain format=json

虽然通过optimizer_trace可以看到很多详细的优化器选择过程,但是使用起来起来还是比较麻烦,需要过滤的信息很多,这时explain format=json输出json格式的分析数据也是一个不错的选择,它也包含语句将要执行的成本信息,如下:

query_cost  总查询成本
read_cost   IO成本+除 eval_cost以外cpu成本
eval_cost   检测rows * filter条记录的成本
prefix_cost 单次查询的成本,等于read_cost+eval_cost
mysql> explain format=json select * from basic_person_info t1 join basic_person_info2 t2 on t1.id_num=t2.id_num where t1.age >10 and t2.age<20;
{
  "query_block": {
    "select_id": 1,
    "cost_info": {
      "query_cost""7675.46"
    },
    "nested_loop": [
      {
        "table": {
          "table_name""t2",
          "access_type""range",
          "possible_keys": [
            "id_num_unique",
            "idx_age",
            "idx_age_id_num"
          ],
          "key""idx_age",
          "used_key_parts": [
            "age"
          ],
          "key_length""1",
          "rows_examined_per_scan": 9594,
          "rows_produced_per_join": 9594,
          "filtered""100.00",
          "index_condition""(`test`.`t2`.`age` < 20)",
          "cost_info": {
            "read_cost""3358.16",
            包含所有io成本+(cpu成本-eval_cost)
            "eval_cost""959.40",
            计算扇出的cpu成本,优化器利用启发式规则估算出满足所有条件的的比例(filtered)
            =rows_examined_per_scan*filtered*0.1
            "prefix_cost""4317.56",
            单表查询的总成本
            
            "data_read_per_join""3M"
          },
          "used_columns": [
            "id",
            "id_num",
            "lastname",
            "firstname",
            "mobile",
            "sex",
            "birthday",
            "age",
            "top_education",
            "address",
            "income_by_year",
            "create_time",
            "update_time"
          ]
        }
      },
      {
        "table": {
          "table_name""t1",
          "access_type""eq_ref",
          "possible_keys": [
            "id_num_unique",
            "idx_age"
          ],
          "key""id_num_unique",
          "used_key_parts": [
            "id_num"
          ],
          "key_length""60",
          "ref": [
            "test.t2.id_num"
          ],
          "rows_examined_per_scan": 1,
          "rows_produced_per_join": 4797,
          "filtered""50.00",
          "cost_info": {
            "read_cost""2398.50",
            包含所有io成本+(cpu成本-eval_cost)
            "eval_cost""479.70",
            计算扇出的cpu成本,优化器利用启发式规则估算出满足所有条件的的比例(filtered)
            =rows_examined_per_scan*filtered*0.1
            "prefix_cost""7675.46",
            两表查询的总cost
            "data_read_per_join""1M"
          },
          "used_columns": [
            "id",
            "id_num",
            "lastname",
            "firstname",
            "mobile",
            "sex",
            "birthday",
            "age",
            "top_education",
            "address",
            "income_by_year",
            "create_time",
            "update_time"
          ],
          "attached_condition""(`test`.`t1`.`age` > 10)"
        }
      }
    ]
  }
}

另外,explain结合show warnings语句一起使用还可以得知优化器改写后的语句

mysql> show warnings;
+-------+------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Level | Code | Message                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          |
+-------+------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Note  | 1003 | /* select#1 */ select `test`.`t1`.`id` AS `id`,`test`.`t1`.`id_num` AS `id_num`,`test`.`t1`.`lastname` AS `lastname`,`test`.`t1`.`firstname` AS `firstname`,`test`.`t1`.`mobile` AS `mobile`,`test`.`t1`.`sex` AS `sex`,`test`.`t1`.`birthday` AS `birthday`,`test`.`t1`.`age` AS `age`,`test`.`t1`.`top_education` AS `top_education`,`test`.`t1`.`address` AS `address`,`test`.`t1`.`income_by_year` AS `income_by_year`,`test`.`t1`.`create_time` AS `create_time`,`test`.`t1`.`update_time` AS `update_time`,`test`.`t2`.`id` AS `id`,`test`.`t2`.`id_num` AS `id_num`,`test`.`t2`.`lastname` AS `lastname`,`test`.`t2`.`firstname` AS `firstname`,`test`.`t2`.`mobile` AS `mobile`,`test`.`t2`.`sex` AS `sex`,`test`.`t2`.`birthday` AS `birthday`,`test`.`t2`.`age` AS `age`,`test`.`t2`.`top_education` AS `top_education`,`test`.`t2`.`address` AS `address`,`test`.`t2`.`income_by_year` AS `income_by_year`,`test`.`t2`.`create_time` AS `create_time`,`test`.`t2`.`update_time` AS `update_time` from `test`.`basic_person_info` `t1` join `test`.`basic_person_info2` `t2` where ((`test`.`t1`.`id_num` = `test`.`t2`.`id_num`) and (`test`.`t1`.`age` > 10) and (`test`.`t2`.`age` < 20)) |
+-------+------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1 row in set (0.00 sec)

总结:

  • MySQL的优化器是基于成本来选择最优执行方案的,哪个成本最少就选哪个,所以重点在于计算出各个执行计划的cost
  • 成本由CPU成本和IO成本组成,每个成本常数值可以自己调整,非必要的情况下不要调整,以免影响整个数据库的执行计划选择
  • 通过开启optimizer_trace可以跟踪优化器的各个环节的分析步骤,可以判断有时候为什么没有走索引而走了全表扫描
  • explain加上format=json选项后可以查看成本信息分为read_cost和eval_cost,但只能看到当前已经选择的执行计划,另外通过show warnings可以看到优化器改写后的语句
Enjoy GreatSQL :)

《零基础学习MySQL》视频课程

B

https://www.bilibili.com/video/BV1Da411W7Va?spm_id_from=333.999.0.0&vd_source=ae1951b64ea7b9e6ba11f1d0bbcff0e4


文章推荐:



关于 GreatSQL

GreatSQL是由万里数据库维护的MySQL分支,专注于提升MGR可靠性及性能,支持InnoDB并行查询特性,是适用于金融级应用的MySQL分支版本。

GreatSQL社区官网: https://greatsql.cn/

Gitee: https://gitee.com/GreatSQL/GreatSQL

GitHub: https://github.com/GreatSQL/GreatSQL

Bilibili:https://space.bilibili.com/1363850082



社区有奖建议反馈:https://greatsql.cn/thread-54-1-1.html
社区博客有奖征稿详情:https://greatsql.cn/thread-100-1-1.html

(对文章有疑问或者有独到见解都可以去社区官网提出或分享哦~)


&QQ

可扫码添加GreatSQL社区助手微信好友,送验证信息“加群”加入GreatSQL/MGR交流微信群,亦可直接扫码加入GreatSQL/MGR交流QQ群


QQ


想看更多技术好文,点个“在看”吧!

继续滑动看下一个
向上滑动看下一个

您可能也对以下帖子感兴趣

文章有问题?点此查看未经处理的缓存