MYSQL 的10046 OPTIMIZER_TRACE

1.set optimizer_trace='enabled=on';    --- 开启trace

2.set optimizer_trace_max_mem_size=1000000;    --- 设置trace大小

3.set end_markers_in_json=on;    --- 增加trace中注释

4.select * from information_schema.optimizer_trace\G;

•  
  {\
•  
  "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 */\