select count(*)使用全表扫描？闲聊下其中的成本计算...

0、关键字

• mysql
• explain
• optimizer_trace
• 成本
• 全表扫描

SELECT count(*) FROM table;
SELECT count(1) FROM table;

1、sql执行成本

sql查询是否选择索引还是全表，执行引擎会通过成本核算出最佳的方式，如果全表扫描成本最低，则使用全表扫描。mysql执行成本组成如下：

• IO成本：即从磁盘把数据加载到内存的成本，默认情况下，读取数据页的IO成本是1，mysql是以页的形式读取数据。即当用到某个数据时，并非仅读取该数据，会把这个数据相邻的数据一起读到内存中，这就是有名的程序局部性原理，所以mysql每次会读取一整页数据（一页大小16kb），一页成本即1。
• CPU成本：数据读入内存后，还要检测数据是否满足条件和排序等CPU操作成本，显然与行数有关，默认情况下，检查成本0.2。

成本核算公式如下：

cost = rows*0.2 + data_length/(1024*16)

在了解mysql查询成本核算公式后，计算下table表的全表扫描成本，通过以下命令查询table表属性信息。

show table status like 'table'

Name	Engine	Version	Row_format	Rows	Avg_row_length	Data_length	Max_data_length	Index_length	Data_free	Auto_increment	Create_time	Update_time	Check_time	Collation	Checksum	Create_options	Comment
table	InnoDB	10	Dynamic	8139	324	2637824	0	0	4194304	200000000000008199	2020-06-03 16:56:50	2020-06-03 11:24:01		utf8_general_ci			样例信息表

依据查询结果rows、data_length可以计算出全表扫描成本，如下：

2、sql执行过程跟踪

sql优化工具使用，开启optimizer_trace

set optimizer_trace = "enabled=on";
explain select count(*) from table;
select * from information_schema.`OPTIMIZER_TRACE`;
set optimizer_trace = "enabled=off";

通过对执行计划过程跟踪，查看其执行成本。

{
  "steps": [
    {
      "join_preparation": {
        "select#": 1,
        "steps": [
          {
            "expanded_query": "/* select#1 */ select count(0) AS `count(*)` from `table` limit 0,200"
          }
        ]
      }
    },
    {
      "join_optimization": {
        "select#": 1,
        "steps": [
          {
            "table_dependencies": [
              {
                "table": "`table`",
                "row_may_be_null": false,
                "map_bit": 0,
                "depends_on_map_bits": [
                ]
              }
            ]
          },
          {
            "rows_estimation": [
              {
                "table": "`table`",
                "table_scan": {
                  "rows": 8139,
                  "cost": 161
                }
              }
            ]
          },
          {
            "considered_execution_plans": [
              {
                "plan_prefix": [
                ],
                "table": "`table`",
                "best_access_path": {
                  "considered_access_paths": [
                    {
                      "rows_to_scan": 8139,
                      "access_type": "scan",
                      "resulting_rows": 8139,
                      "cost": 1788.8,
                      "chosen": true
                    }
                  ]
                },
                "condition_filtering_pct": 100,
                "rows_for_plan": 8139,
                "cost_for_plan": 1788.8,
                "chosen": true
              }
            ]
          },
          {
            "attaching_conditions_to_tables": {
              "original_condition": null,
              "attached_conditions_computation": [
              ],
              "attached_conditions_summary": [
                {
                  "table": "`table`",
                  "attached": null
                }
              ]
            }
          },
          {
            "refine_plan": [
              {
                "table": "`table`"
              }
            ]
          }
        ]
      }
    },
    {
      "join_execution": {
        "select#": 1,
        "steps": [
        ]
      }
    }
  ]
}

3、具体场景分析

序号	SQL	执行计划
1	select created_date from table where enabled_flag = ‘Y’ order by created_date desc;	全表扫描
2	select created_date from table force index(created_date_idx) where enabled_flag = ‘Y’ order by created_date desc;	使用索引

备注：table表created_date字段已创建索引。

3.1、select created_date from table where enabled_flag = ‘Y’ order by created_date desc;

explain select created_date from table where enabled_flag = ‘Y’ order by created_date desc;
查看执行计划，该sql为全表（All）扫描。成本为1786.2。

{
  "considered_execution_plans": [
    {
      "plan_prefix": [
      ],
      "table": "`do_task_ver`",
      "best_access_path": {
        "considered_access_paths": [
          {
            "rows_to_scan": 8126,
            "access_type": "scan",
            "resulting_rows": 812.6,
            "cost": 1786.2,
            "chosen": true,
            "use_tmp_table": true
          }
        ]
      },
      "condition_filtering_pct": 100,
      "rows_for_plan": 812.6,
      "cost_for_plan": 1786.2,
      "sort_cost": 812.6,
      "new_cost_for_plan": 2598.8,
      "chosen": true
    }
  ]
}

3.2、select created_date from table force index(created_date_idx) where enabled_flag = ‘Y’ order by created_date desc;

explain select created_date from table force index(created_date_idx) where enabled_flag = ‘Y’ order by created_date desc;
查看执行计划，该sql为使用了索引index（created_date_idx）。成本为9752.2。

{
  "considered_execution_plans": [
    {
      "plan_prefix": [
      ],
      "table": "`do_task_ver` FORCE INDEX (`created_date_idx`)",
      "best_access_path": {
        "considered_access_paths": [
          {
            "rows_to_scan": 8126,
            "access_type": "scan",
            "resulting_rows": 812.6,
            "cost": 9752.2,
            "chosen": true,
            "use_tmp_table": true
          }
        ]
      },
      "condition_filtering_pct": 100,
      "rows_for_plan": 812.6,
      "cost_for_plan": 9752.2,
      "sort_cost": 812.6,
      "new_cost_for_plan": 10565,
      "chosen": true
    }
  ]
}

通过对比以上两sql，发现使用索引的成本大于全表扫描成本，因此默认情况select created_date from table where enabled_flag = ‘Y’ order by created_date desc;语句不会使用索引。

mysql选择的执行计划未必是最佳的，原因大多为行数统计不准确，mysql认为的最优是成本最低，我们认为最优是时间最短。

本篇文章仅一个浅显的记录，若想了解更深入，需要参考更多资料。

4、参考

1、8.9.5 The Optimizer Cost Model 成本模型