单表扫描，MySQL索引选择不正确 并 详细解析OPTIMIZER_TRACE格式

一 表结构如下: 

MySQL  5.5.30  5.6.20 版本, 表大概有815万行

CREATE TABLE t_audit_operate_log (
  Fid bigint(16) AUTO_INCREMENT,
  Fcreate_time int(10) unsigned NOT NULL DEFAULT '0',
  Fuser varchar(50) DEFAULT '',
  Fip bigint(16) DEFAULT NULL,
  Foperate_object_id bigint(20) DEFAULT '0',
  PRIMARY KEY (Fid),
  KEY indx_ctime (Fcreate_time),
  KEY indx_user (Fuser),
  KEY indx_objid (Foperate_object_id),
  KEY indx_ip (Fip)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

执行查询:

mysql> explain select count(*) from t_audit_operate_log where Fuser='XX@XX.com' and Fcreate_time>=1407081600 and Fcreate_time<=1407427199\G

*************************** 1. row ***************************

id: 1

select_type: SIMPLE

table: t_audit_operate_log

type: ref

possible_keys: indx_ctime,indx_user

key: indx_user

key_len: 153

ref: const

rows: 2007326

Extra: Using where

发现,使用了一个不合适的索引, 不是很理想，于是改成指定索引：

mysql> explain select count(*) from t_audit_operate_log use index(indx_ctime) where Fuser='CY6016@cyou-inc.com' and Fcreate_time>=1407081600 and Fcreate_time<=1407427199\G

*************************** 1. row ***************************

id: 1

select_type: SIMPLE

table: t_audit_operate_log

type: range

possible_keys: indx_ctime

key: indx_ctime

key_len: 5

ref: NULL

rows: 670092

Extra: Using where

实际执行耗时，后者比前者快了接近10

问题: 很奇怪，优化器为何不选择使用 indx_ctime 索引，而选择了明显会扫描更多行的 indx_user 索引。

分析2个索引的数据量如下:  两个条件的唯一性对比：

select count(*) from t_audit_operate_log where Fuser='XX@XX.com';
+----------+
| count(*) |
+----------+
| 1238382 | 
+----------+

select count(*) from t_audit_operate_log where Fcreate_time>=1407254400 and Fcreate_time<=1407427199;
+----------+
| count(*) |
+----------+
| 198920 | 
+----------+

显然,使用索引indx_ctime好于indx_user,但MySQL却选择了indx_user. 为什么?

于是,使用 OPTIMIZER_TRACE进一步探索.

二  OPTIMIZER_TRACE的过程说明

以本处事例简要说明OPTIMIZER_TRACE的过程.

{\
  "steps": [\
    {\
      "join_preparation": {\  ---优化准备工作
        "select#": 1,\
        "steps": [\
          {\
            "expanded_query": "/* select#1 */ select count(0) AS `count(*)` from `t_audit_operate_log` where ((`t_audit_operate_log`.`Fuser` = 'XX@XX.com') and (`t_audit_operate_log`.`Fcreate_time` >= 1407081600) and (`t_audit_operate_log`.`Fcreate_time` <= 1407427199))"\
          }\
        ] /* steps */\
      } /* join_preparation */\
    },\
    {\
      "join_optimization": {\  ---优化工作的主要阶段,包括逻辑优化和物理优化两个阶段
        "select#": 1,\
        "steps": [\  ---优化工作的主要阶段, 逻辑优化阶段
          {\
            "condition_processing": {\  ---逻辑优化,条件化简
              "condition": "WHERE",\
              "original_condition": "((`t_audit_operate_log`.`Fuser` = 'XX@XX.com') and (`t_audit_operate_log`.`Fcreate_time` >= 1407081600) and (`t_audit_operate_log`.`Fcreate_time` <= 1407427199))",\
              "steps": [\
                {\
                  "transformation": "equality_propagation",\  ---逻辑优化,条件化简,等式处理
                  "resulting_condition": "((`t_audit_operate_log`.`Fuser` = 'XX@XX.com') and (`t_audit_operate_log`.`Fcreate_time` >= 1407081600) and (`t_audit_operate_log`.`Fcreate_time` <= 1407427199))"\
                },\
                {\
                  "transformation": "constant_propagation",\  ---逻辑优化,条件化简,常量处理
                  "resulting_condition": "((`t_audit_operate_log`.`Fuser` = 'XX@XX.com') and (`t_audit_operate_log`.`Fcreate_time` >= 1407081600) and (`t_audit_operate_log`.`Fcreate_time` <= 1407427199))"\
                },\
                {\
                  "transformation": "trivial_condition_removal",\  ---逻辑优化,条件化简,条件去除
                  "resulting_condition": "((`t_audit_operate_log`.`Fuser` = 'XX@XX.com') and (`t_audit_operate_log`.`Fcreate_time` >= 1407081600) and (`t_audit_operate_log`.`Fcreate_time` <= 1407427199))"\
                }\
              ] /* steps */\
            } /* condition_processing */\
          },\  ---逻辑优化,条件化简,结束
          {\
            "table_dependencies": [\  ---逻辑优化, 找出表之间的相互依赖关系. 非直接可用的优化方式.
              {\
                "table": "`t_audit_operate_log`",\
                "row_may_be_null": false,\
                "map_bit": 0,\
                "depends_on_map_bits": [\
                ] /* depends_on_map_bits */\
              }\
            ] /* table_dependencies */\
          },\
          {\
            "ref_optimizer_key_uses": [\   ---逻辑优化,  找出备选的索引
              {\
                "table": "`t_audit_operate_log`",\
                "field": "Fuser",\
                "equals": "'XX@XX.com'",\
                "null_rejecting": false\
              }\
            ] /* ref_optimizer_key_uses */\
          },\
          {\
            "rows_estimation": [\   ---逻辑优化, 估算每个表的元组个数. 单表上进行全表扫描和索引扫描的代价估算. 每个索引都估算索引扫描代价
              {\
                "table": "`t_audit_operate_log`",\
                "range_analysis": {\
                  "table_scan": {\---逻辑优化, 估算每个表的元组个数. 单表上进行全表扫描的代价
                    "rows": 8150516,\
                    "cost": 1.73e6\
                  } /* table_scan */,\
                  "potential_range_indices": [\ ---逻辑优化, 列出备选的索引. 后续版本字符串变为potential_range_indexes
                    {\
                      "index": "PRIMARY",\---逻辑优化, 本行表明主键索引不可用
                      "usable": false,\
                      "cause": "not_applicable"\
                    },\
                    {\
                      "index": "indx_ctime",\---逻辑优化, 索引indx_ctime
                      "usable": true,\
                      "key_parts": [\
                        "Fcreate_time",\
                        "Fid"\
                      ] /* key_parts */\
                    },\
                    {\
                      "index": "indx_user",\---逻辑优化, 索引indx_user
                      "usable": true,\
                      "key_parts": [\
                        "Fuser",\
                        "Fid"\
                      ] /* key_parts */\
                    },\
                    {\
                      "index": "indx_objid",\---逻辑优化, 索引
                      "usable": false,\
                      "cause": "not_applicable"\
                    },\
                    {\
                      "index": "indx_ip",\---逻辑优化, 索引
                      "usable": false,\
                      "cause": "not_applicable"\
                    }\
                  ] /* potential_range_indices */,\
                  "setup_range_conditions": [\ ---逻辑优化,如果有可下推的条件,则带条件考虑范围查询
                  ]/* setup_range_conditions */,\
                  "group_index_range":{\---逻辑优化,如带有GROUPBY或DISTINCT,则考虑是否有索引可优化这种操作.并考虑带有MIN/MAX的情况
                    "chosen":false,\
                    "cause":"not_group_by_or_distinct"\
                  }/* group_index_range */,\
                  "analyzing_range_alternatives":{\---逻辑优化,开始计算每个索引做范围扫描的花费(等值比较是范围扫描的特例)
                    "range_scan_alternatives":[\
                      {\
                        "index":"indx_ctime",\ ---[A]
                        "ranges":[\
                          "1407081600 <= Fcreate_time <= 1407427199"\
                        ]/* ranges */,\
                        "index_dives_for_eq_ranges":true,\
                        "rowid_ordered":false,\
                        "using_mrr":true,\
                        "index_only":false,\
                        "rows":688362,\
                        "cost":564553,\ ---逻辑优化,这个索引的代价最小
                        "chosen":true\ ---逻辑优化,这个索引的代价最小,被选中.(比前面的table_scan 和其他索引的代价都小)
                      },\
                      {\
                        "index":"indx_user",\
                        "ranges":[\
                          "XX@XX.com <= Fuser <= XX@XX.com"\
                        ]/* ranges */,\
                        "index_dives_for_eq_ranges":true,\
                        "rowid_ordered":true,\
                        "using_mrr":true,\
                        "index_only":false,\
                        "rows":1945894,\
                        "cost":1.18e6,\
                        "chosen":false,\
                        "cause":"cost"\
                      }\
                    ]/* range_scan_alternatives */,\
                    "analyzing_roworder_intersect":{\
                      "usable":false,\
                      "cause":"too_few_roworder_scans"\
                    }/* analyzing_roworder_intersect */\
                  }/* analyzing_range_alternatives */,\---逻辑优化,开始计算每个索引做范围扫描的花费.这项工作结算
                  "chosen_range_access_summary":{\---逻辑优化,开始计算每个索引做范围扫描的花费.总结本阶段最优的.
                    "range_access_plan":{\
                      "type":"range_scan",\
                      "index":"indx_ctime",\
                      "rows":688362,\
                      "ranges":[\
                        "1407081600 <= Fcreate_time <= 1407427199"\
                      ]/* ranges */\
                    }/* range_access_plan */,\
                    "rows_for_plan":688362,\
                    "cost_for_plan":564553,\
                    "chosen":true\    --这里看到的cost和rows都比 indx_user 要来的小很多---这个和[A]处是一样的,是信息汇总.
                  }/* chosen_range_access_summary */\
                }/* range_analysis */\
              }\
            ]/* rows_estimation */\ ---逻辑优化,估算每个表的元组个数.行估算结束
          },\
          {\
            "considered_execution_plans":[\ ---物理优化,开始多表连接的物理优化计算
              {\
                "plan_prefix":[\
                ]/* plan_prefix */,\
                "table":"`t_audit_operate_log`",\
                "best_access_path":{\
                  "considered_access_paths":[\
                    {\
                      "access_type":"ref",\ ---物理优化,计算indx_user索引上使用ref方查找的花费,
                      "index":"indx_user",\
                      "rows":1.95e6,\
                      "cost":683515,\
                      "chosen":true\
                    },\ ---物理优化,本应该比较所有的可用索引,即打印出多个格式相同的但索引名不同的内容,这里却没有。推测是bug--没有遍历每一个索引.
                    {\
                      "access_type":"range",\---物理优化,猜测对应的是indx_time（没有实例可进行调试，对比5.7的跟踪信息猜测而得）
                      "rows":516272,\
                      "cost":702225,\---物理优化，代价大于了ref方式的683515，所以没有被选择
                      "chosen":false\   -- cost比上面看到的增加了很多，但rows没什么变化---物理优化，此索引没有被选择
                    }\
                  ]/* considered_access_paths */\
                }/* best_access_path */,\
                "cost_for_plan":683515,\ ---物理优化，汇总在best_access_path 阶段得到的结果
                "rows_for_plan":1.95e6,\
                "chosen":true\   -- cost比上面看到的竟然小了很多？虽然rows没啥变化  ---物理优化，汇总在best_access_path 阶段得到的结果
              }\
            ]/* considered_execution_plans */\
          },\
          {\
            "attaching_conditions_to_tables":{\---逻辑优化，尽量把条件绑定到对应的表上
              }/* attaching_conditions_to_tables */\
          },\
          {\
            "refine_plan":[\
              {\
                "table":"`t_audit_operate_log`",\---逻辑优化，下推索引条件"pushed_index_condition"；其他条件附加到表上做为过滤条件"table_condition_attached"
              }\
            ]/* refine_plan */\
          }\
        ]/* steps */\
      }/* join_optimization */\ \---逻辑优化和物理优化结束
    },\
    {\
      "join_explain":{}/* join_explain */\
    }\
  ]/* steps */\

三 其他一个相似问题

单表扫描，使用ref和range从索引获取数据一例  

http://blog.163.com/li_hx/blog/static/183991413201461853637715/

四 问题的解决方式

遇到单表上有多个索引的时候,在MySQL5.6.20版本之前的版本,需要人工强制使用索引,以达到最好的效果.