研发leader突然给了1条sql说这条sql在测试环境库db-stage执行非常的慢，放到线上db-read从库上执行非常的快。而且线上库从库的表的数据远多于测试环境库的表数据。让我分析下是什么问题？？
二、sql内容如下 select odtl.id from db_order.t_order_device_trans_log odtl left join db_order.t_order_items_detail oid on odtl.order_id=oid.order_id left join db_order.t_orders o on oid.order_id=o.order_id where oid.customer_id= 40933 group by o.order_id;对此sql简单分析如下:
此sql包含了3个表的 left join 查询
db_order.t_order_device_trans_log db_order.t_order_items_detail 和db_order.t_orders 表连接组成，分别用了odtl、 oid、 o 别名表示。因为都是用的 LEFT JOIN，所以表连接顺序应该是 odtl-->oid-->o，和explain 执行计划中显示的一致。
本以为测试环境库配置参数和线上db-read库配置参数都是一样的。研发的话不能全信。还是拿着sql在测试库db-stage和线上db-read1 (Mariadb库版本10.2.17)，db-read4(mysql5.7.32) 跑了下
3.1、db-stage环境上的sql的执行计划：mysql>explain select odtl.id from db_order.t_order_device_trans_log odtl left join db_order.t_order_items_detail oid on odtl.order_id=oid.order_id left join db_order.t_orders o on oid.order_id=o.order_id where oid.customer_id= 40933 group by o.order_id;+----+-------------+-------+------------+--------+------------------------------------------------------------------------------------+-----------+---------+-----------------------+--------+----------+----------------------------------------------+| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |+----+-------------+-------+------------+--------+------------------------------------------------------------------------------------+-----------+---------+-----------------------+--------+----------+----------------------------------------------+| 1 | SIMPLE | odtl | NULL | index | NULL | idx_order | 62 | NULL | 785867 | 100.00 | Using index; Using temporary; Using filesort || 1 | SIMPLE | oid | NULL | ref | order_id,customer_id | order_id | 82 | func | 3 | 1.39 | Using index condition; Using where || 1 | SIMPLE | o | NULL | eq_ref | PRIMARY,order_id,main_order_id,customer_id_bill,source_order_id,idx_is_sale_status | order_id | 82 | db_order.oid.order_id | 1 | 100.00 | Using index |+----+-------------+-------+------------+--------+------------------------------------------------------------------------------------+-----------+---------+-----------------------+--------+----------+----------------------------------------------+3 rows in set, 1 warning (0.00 sec)分析：
发现别名表 odtl type 类型为index,说明一开始是选择了索引的 possible_keys为NULL key:存在索引idx_order,但是没走索引，并且进行了全表扫描。
'db-read4>explain select odtl.id from db_order.t_order_device_trans_log odtl left join db_order.t_order_items_detail oid on odtl.order_id=oid.order_id left join db_order.t_orders o on oid.order_id=o.order_id where oid.customer_id= 40933 group by o.order_id;+----+-------------+-------+------------+--------+-----------------------------------------------------------------+-------------+---------+-----------------------+------+----------+----------------------------------------------+| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |+----+-------------+-------+------------+--------+-----------------------------------------------------------------+-------------+---------+-----------------------+------+----------+----------------------------------------------+| 1 | SIMPLE | oid | NULL | ref | order_id,customer_id | customer_id | 4 | const | 279 | 100.00 | Using where; Using temporary; Using filesort || 1 | SIMPLE | o | NULL | eq_ref | PRIMARY,order_id,main_order_id,customer_id_bill,source_order_id | order_id | 82 | db_order.oid.order_id | 1 | 100.00 | Using index || 1 | SIMPLE | odtl | NULL | ref | idx_order | idx_order | 82 | db_order.oid.order_id | 7 | 100.00 | Using index |+----+-------------+-------+------------+--------+-----------------------------------------------------------------+-------------+---------+-----------------------+------+----------+----------------------------------------------+3 rows in set, 1 warning (0.00 sec)分析:
发现别名表 odtl type类型为ref ,说明选择了ref级别索引; possible_keys为 idx_order; key:存在索引idx_order ，而且走了索引的。
root@db-read1 13:17: [(none)]> explain select odtl.id from db_order.t_order_device_trans_log odtl left join db_order.t_order_items_detail oid on odtl.order_id=oid.order_id left join db_order.t_orders o on oid.order_id=o.order_id where oid.customer_id= 40933 group by o.order_id;+------+-------------+-------+--------+----------------------+-------------+---------+-----------------------+------+---------------------------------+| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |+------+-------------+-------+--------+----------------------+-------------+---------+-----------------------+------+---------------------------------+| 1 | SIMPLE | oid | ref | order_id,customer_id | customer_id | 4 | const | 279 | Using temporary; Using filesort || 1 | SIMPLE | o | eq_ref | order_id | order_id | 82 | db_order.oid.order_id | 1 | Using index || 1 | SIMPLE | odtl | ref | idx_order | idx_order | 82 | db_order.oid.order_id | 3 | Using index |+------+-------------+-------+--------+----------------------+-------------+---------+-----------------------+------+---------------------------------+3 rows in set (0.01 sec)分析:
发现别名表 odtl type类型为ref ,说明选择了ref级别索引; possible_keys为 idx_order; key:存在索引idx_order ，而且走了索引的。
结果：db-stage库确实是全表扫描了。确实比较奇怪。到底是什么问题导致的呢？
检查了下这条sql涉及到的这3个表的索引结构（测试环境和线上环境）。没发现有任何不同。太奇怪了。
换了一种方式查看sql的执行计划:找到了线索
指定explain执行计划的线上格式为JSON。可以更详细的显示出执行计划的具体信息（建议生产上多多使用此方法分析sql执行计划）
在db-stage库上执行分析：
mysql> explain format=json select odtl.id from db_order.t_order_device_trans_log odtl left join db_order.t_order_items_detail oid on odtl.order_id=oid.order_id left join db_order.t_orderss o on oid.order_id=o.order_id where oid.customer_id= 40933 group by o.order_id\G*************************** 1. row ***************************EXPLAIN: { "query_block": { "select_id": 1, "cost_info": { "query_cost": "3634954.72" }, "grouping_operation": { "using_temporary_table": true, "using_filesort": true, "cost_info": { "sort_cost": "39293.35" }, "nested_loop": [ { "table": { "table_name": "odtl", "access_type": "index", "key": "idx_order", "used_key_parts": [ "order_id" ], "key_length": "62", "rows_examined_per_scan": 785867, "rows_produced_per_join": 785867, "filtered": "100.00", "using_index": true, "cost_info": { "read_cost": "4007.00", "eval_cost": "157173.40", "prefix_cost": "161180.40", "data_read_per_join": "179M" }, "used_columns": [ "id", "order_id" ] } }, { "table": { "table_name": "oid", "access_type": "ref", "possible_keys": [ "order_id", "customer_id" ], "key": "order_id", "used_key_parts": [ "order_id" ], "key_length": "82", "ref": [ "func" ], "rows_examined_per_scan": 3, "rows_produced_per_join": 39293, "filtered": "1.39", "index_condition": "(convert(`db_order`.`odtl`.`order_id` using utf8mb4) = `db_order`.`oid`.`order_id`)", "cost_info": { "read_cost": "2822774.13", "eval_cost": "7858.67", "prefix_cost": "3548509.35", "data_read_per_join": "218M" }, "used_columns": [ "id", "order_id", "customer_id" ], "attached_condition": "(`db_order`.`oid`.`customer_id` = 40933)" } }, { "table": { "table_name": "o", "access_type": "eq_ref", "possible_keys": [ "PRIMARY", "order_id", "main_order_id", "customer_id_bill", "source_order_id", "idx_is_sale_status" ], "key": "order_id", "used_key_parts": [ "order_id" ], "key_length": "82", "ref": [ "db_order.oid.order_id" ], "rows_examined_per_scan": 1, "rows_produced_per_join": 39293, "filtered": "100.00", "using_index": true, "cost_info": { "read_cost": "39293.35", "eval_cost": "7858.67", "prefix_cost": "3595661.37", "data_read_per_join": "474M" }, "used_columns": [ "id", "order_id" ] } } ] } }}1 row in set, 1 warning (0.00 sec)计划index_condition条件中显示如下:发现了别名表odtl字段order_id =匹配别名表oid字段order_id时出现了utf8mb4字符集的转换
"index_condition": "(convert(`db_order`.`odtl`.`order_id` using utf8mb4) = `db_order`.`oid`.`order_id`)",豁然开朗，很有可能是表db_order.t_order_device_trans_log 字段order_id字符集和db_order.t_order_items_detail表字段order_id字符集不一样,导致执行sql过程中，使得原本的索引失效，造成db_order.t_order_device_trans_log全表的扫描。
于是在db-read1和db-read4库也同样执行json格式的explain计划：指标都是正常走索引的
'db-read4' root@localhost 23:08:05 (none)>explain format=json select odtl.id from db_order.t_order_device_trans_log odtl left join db_order.t_order_items_detail oid on odtl.order_id=oid.order_id left join db_order.t_orders o on oid.order_id=o.order_id where oid.customer_id= 40933 group by o.order_id\G*************************** 1. row ***************************EXPLAIN: { "query_block": { "select_id": 1, "cost_info": { "query_cost": "3395.51" }, "grouping_operation": { "using_temporary_table": true, "using_filesort": true, "cost_info": { "sort_cost": "2023.29" }, "nested_loop": [ { "table": { "table_name": "oid", "access_type": "ref", "possible_keys": [ "order_id", "customer_id" ], "key": "customer_id", "used_key_parts": [ "customer_id" ], "key_length": "4", "ref": [ "const" ], "rows_examined_per_scan": 279, "rows_produced_per_join": 279, "filtered": "100.00", "cost_info": { "read_cost": "279.00", "eval_cost": "55.80", "prefix_cost": "334.80", "data_read_per_join": "1M" }, "used_columns": [ "id", "order_id", "customer_id" ], "attached_condition": "(`db_order`.`oid`.`order_id` is not null)" } }, { "table": { "table_name": "o", "access_type": "eq_ref", "possible_keys": [ "PRIMARY", "order_id", "main_order_id", "customer_id_bill", "source_order_id" ], "key": "order_id", "used_key_parts": [ "order_id" ], "key_length": "82", "ref": [ "db_order.oid.order_id" ], "rows_examined_per_scan": 1, "rows_produced_per_join": 279, "filtered": "100.00", "using_index": true, "cost_info": { "read_cost": "279.00", "eval_cost": "55.80", "prefix_cost": "669.60", "data_read_per_join": "3M" }, "used_columns": [ "id", "order_id" ] } }, { "table": { "table_name": "odtl", "access_type": "ref", "possible_keys": [ "idx_order" ], "key": "idx_order", "used_key_parts": [ "order_id" ], "key_length": "82", "ref": [ "db_order.oid.order_id" ], "rows_examined_per_scan": 7, "rows_produced_per_join": 2023, "filtered": "100.00", "using_index": true, "cost_info": { "read_cost": "297.96", "eval_cost": "404.66", "prefix_cost": "1372.22", "data_read_per_join": "616K" }, "used_columns": [ "id", "order_id" ] } } ] } }}1 row in set, 1 warning (0.00 sec)root@db-read1 13:38: [(none)]> explain format=json select odtl.id from db_order.t_order_device_trans_log odtl left join db_order.t_order_items_detail oid on odtl.order_id=oid.order_id left join db_order.t_orders o on oid.order_id=o.order_id where oid.customer_id= 40933 group by o.order_id\G*************************** 1. row ***************************EXPLAIN: { "query_block": { "select_id": 1, "filesort": { "sort_key": "db_order.o.order_id", "temporary_table": { "table": { "table_name": "oid", "access_type": "ref", "possible_keys": ["order_id", "customer_id"], "key": "customer_id", "key_length": "4", "used_key_parts": ["customer_id"], "ref": ["const"], "rows": 279, "filtered": 100 }, "table": { "table_name": "o", "access_type": "eq_ref", "possible_keys": ["order_id"], "key": "order_id", "key_length": "82", "used_key_parts": ["order_id"], "ref": ["db_order.oid.order_id"], "rows": 1, "filtered": 100, "using_index": true }, "table": { "table_name": "odtl", "access_type": "ref", "possible_keys": ["idx_order"], "key": "idx_order", "key_length": "82", "used_key_parts": ["order_id"], "ref": ["db_order.oid.order_id"], "rows": 3, "filtered": 100, "using_index": true } } } }}1 row in set (0.00 sec)问题就是db-stage 库 表db_order.t_order_device_trans_log 字符集居然是utf8导致的索引失效。
于是查看测试库db-stage 表db_order.t_order_device_trans_log 字符集居然是utf8mysql> show create table db_order.t_order_device_trans_log\G *************************** 1. row *************************** Table: t_order_device_trans_logCreate Table: CREATE TABLE `t_order_device_trans_log` ( `id` bigint(20) NOT NULL AUTO_INCREMENT, `order_id` varchar(20) NOT NULL COMMENT '订单号', `order_type` tinyint(2) NOT NULL COMMENT '订单类型', `rent_type` tinyint(2) NOT NULL COMMENT '租赁类型（1.随租随还2.固定租期3.租完即送4.无期限随租随还5.短期租赁）', `item_detail_id` bigint(20) NOT NULL COMMENT '订单设备id', `sn` varchar(50) NOT NULL COMMENT '序列号', `before_status` tinyint(2) DEFAULT NULL COMMENT '变更前设备状态', `change_type` tinyint(2) NOT NULL COMMENT '变更类型：0:已发货，1:起租，2:退租，3:融租完成，4:挂应收，5:换货, 99:取消发货', `cdate` datetime NOT NULL, PRIMARY KEY (`id`), KEY `idx_order` (`order_id`)) ENGINE=InnoDB AUTO_INCREMENT=3794673 DEFAULT CHARSET=utf81 row in set (0.00 sec)db-stage库其他2个表db_order.t_order_items_detail 和db_order.t_orders 字符集都是 utf8mb4
于是查看测试库db-read1 表db_order.t_order_device_trans_log db_order.t_order_items_detail 和db_order.t_orders 字符集utf8mb4
于是查看测试库db-read4 表db_order.t_order_device_trans_log db_order.t_order_items_detail 和db_order.t_orders 字符集utf8mb4
切记：修改表字符集前一定要对表提前做下备份
mysqldump -uroot -p'UEgzFO' -h 127.0.0.1 db_order t_order_device_trans_log > /data/dump/2021-04-13.t_order_device_trans_log.sql mysql> show create table db_order.t_order_device_trans_log\G *************************** 1. row *************************** Table: t_order_device_trans_logCreate Table: CREATE TABLE `t_order_device_trans_log` ( `id` bigint(20) NOT NULL AUTO_INCREMENT, `order_id` varchar(20) NOT NULL COMMENT '订单号', `order_type` tinyint(2) NOT NULL COMMENT '订单类型', `rent_type` tinyint(2) NOT NULL COMMENT '租赁类型（1.随租随还2.固定租期3.租完即送4.无期限随租随还5.短期租赁）', `item_detail_id` bigint(20) NOT NULL COMMENT '订单设备id', `sn` varchar(50) NOT NULL COMMENT '序列号', `before_status` tinyint(2) DEFAULT NULL COMMENT '变更前设备状态', `change_type` tinyint(2) NOT NULL COMMENT '变更类型：0:已发货，1:起租，2:退租，3:融租完成，4:挂应收，5:换货, 99:取消发货', `cdate` datetime NOT NULL, PRIMARY KEY (`id`), KEY `idx_order` (`order_id`)) ENGINE=InnoDB AUTO_INCREMENT=3794673 DEFAULT CHARSET=utf81 row in set (0.00 sec)占用磁盘的大小:
mysql> select count(*) from db_order.t_order_device_trans_log\G*************************** 1. row ***************************count(*): 9622341 row in set (0.19 sec)mysql> SELECT (`DATA_LENGTH`+ `INDEX_LENGTH`)/1024/1024 as `table_data_size` from `TABLES` WHERE TABLE_NAME ='t_order_device_trans_log' and TABLE_SCHEMA='db_order';+-----------------+| table_data_size |+-----------------+| 62.60937500 |+-----------------+1 row in set (0.00 sec)下面的操作是错误的，才导致后面执行sql报错
mysql > ALTER TABLE db_order.t_order_device_trans_log CONVERT TO CHARACTER SET utf8mb4;Query OK, 962234 rows affected (17.99 sec)Records: 962234 Duplicates: 0 Warnings: 0mysql> show create table db_order.t_order_device_trans_log\G*************************** 1. row *************************** Table: t_order_device_trans_logCreate Table: CREATE TABLE `t_order_device_trans_log` ( `id` bigint(20) NOT NULL AUTO_INCREMENT, `order_id` varchar(20) NOT NULL COMMENT '订单号', `order_type` tinyint(2) NOT NULL COMMENT '订单类型', `rent_type` tinyint(2) NOT NULL COMMENT '租赁类型（1.随租随还2.固定租期3.租完即送4.无期限随租随还5.短期租赁）', `item_detail_id` bigint(20) NOT NULL COMMENT '订单设备id', `sn` varchar(50) NOT NULL COMMENT '序列号', `before_status` tinyint(2) DEFAULT NULL COMMENT '变更前设备状态', `change_type` tinyint(2) NOT NULL COMMENT '变更类型：0:已发货，1:起租，2:退租，3:融租完成，4:挂应收，5:换货, 99:取消发货', `cdate` datetime NOT NULL, PRIMARY KEY (`id`), KEY `idx_order` (`order_id`)) ENGINE=InnoDB AUTO_INCREMENT=3794673 DEFAULT CHARSET=utf8mb41 row in set (0.00 sec)执行计划报错：mysql> explain select odtl.id from db_order.t_order_device_trans_log odtl left join db_order.t_order_items_detail oid on odtl.order_id=oid.order_id left join db_order.t_orders o on oid.order_id=o.order_id where oid.customer_id= 40933 group by o.order_id;ERROR 1267 (HY000): Illegal mix of collations (utf8mb4_general_ci,IMPLICIT) and (utf8mb4_unicode_ci,IMPLICIT) for operation '='mysql> ALTER TABLE t_order_device_trans_log DEFAULT CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci;Query OK, 0 rows affected (0.01 sec)Records: 0 Duplicates: 0 Warnings: 0查看执行计划，还在报错：
mysql> explain select odtl.id from db_order.t_order_device_trans_log odtl left join db_order.t_order_items_detail oid on odtl.order_id=oid.order_id left join db_order.t_orders o on oid.order_id=o.order_id where oid.customer_id= 40933 group by o.order_id; ERROR 1267 (HY000): Illegal mix of collations (utf8mb4_general_ci,IMPLICIT) and (utf8mb4_unicode_ci,IMPLICIT) for operation '='重要提示：下面的参数可以不重启MySQL服务进行动态设置。同时也要主要my.cnf配置文件字符集参数的设置
[root@db-stage ~]# egrep "character|collation_server" /etc/my.cnfdefault-character-set=utf8mb4#character-set-server=utf8character_set_server = 'utf8mb4'collation_server = 'utf8mb4_unicode_ci'#collation_server = 'utf8mb4_general_ci'下面是还原原来表字符集 utf8:
mysql> ALTER TABLE db_order.t_order_device_trans_log CONVERT TO CHARACTER SET utf8;Query OK, 962234 rows affected (17.99 sec)Records: 962234 Duplicates: 0 Warnings: 0再次执行下面的修改表字符集utf8mb4为命令：
mysql> ALTER TABLE db_order.t_order_device_trans_log CONVERT TO CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci;Query OK, 962784 rows affected (18.99 sec)Records: 962784 Duplicates: 0 Warnings: 0mysql> show create table db_order.t_order_device_trans_log\G*************************** 1. row *************************** Table: t_order_device_trans_logCreate Table: CREATE TABLE `t_order_device_trans_log` ( `id` bigint(20) NOT NULL AUTO_INCREMENT, `order_id` varchar(20) COLLATE utf8mb4_unicode_ci NOT NULL COMMENT '订单号', `order_type` tinyint(2) NOT NULL COMMENT '订单类型', `rent_type` tinyint(2) NOT NULL COMMENT '租赁类型（1.随租随还2.固定租期3.租完即送4.无期限随租随还5.短期租赁）', `item_detail_id` bigint(20) NOT NULL COMMENT '订单设备id', `sn` varchar(50) COLLATE utf8mb4_unicode_ci NOT NULL COMMENT '序列号', `before_status` tinyint(2) DEFAULT NULL COMMENT '变更前设备状态', `change_type` tinyint(2) NOT NULL COMMENT '变更类型：0:已发货，1:起租，2:退租，3:融租完成，4:挂应收，5:换货, 99:取消发货', `cdate` datetime NOT NULL, PRIMARY KEY (`id`), KEY `idx_order` (`order_id`)) ENGINE=InnoDB AUTO_INCREMENT=3795223 DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci1 row in set (0.00 sec)mysql> s��̬,ʧ��how variables where Variable_name like 'collation%';+----------------------+--------------------+| Variable_name | Value |+----------------------+--------------------+| collation_connection | utf8mb4_general_ci || collation_database | utf8mb4_unicode_ci || collation_server | utf8mb4_unicode_ci |+----------------------+--------------------+3 rows in set (0.01 sec)查看表占用磁盘物理空间大小:
mysql> use information_schema;Reading table information for completion of table and column namesYou can turn off this feature to get a quicker startup with -ADatabase changedmysql> SELECT (`DATA_LENGTH`+ `INDEX_LENGTH`)/1024/1024 as `table_data_size` from `TABLES` WHERE TABLE_NAME ='t_order_device_trans_log' and TABLE_SCHEMA='db_order';+-----------------+| table_data_size |+-----------------+| 142.35937500 |+-----------------+1 row in set (0.00 sec)此时查看db-stage json格式的执行计划：
mysql> explain format=json select odtl.id from db_order.t_order_device_trans_log odtl left join db_order.t_order_items_detail oid on odtl.order_id=oid.order_id left join db_order.t_ordeers o on oid.order_id=o.order_id where oid.customer_id= 40933 group by o.order_id\G*************************** 1. row ***************************EXPLAIN: { "query_block": { "select_id": 1, "cost_info": { "query_cost": "1073.53" }, "grouping_operation": { "using_temporary_table": true, "using_filesort": true, "cost_info": { "sort_cost": "589.89" }, "nested_loop": [ { "table": { "table_name": "oid", "access_type": "ref", "possible_keys": [ "order_id", "customer_id" ], "key": "customer_id", "used_key_parts": [ "customer_id" ], "key_length": "4", "ref": [ "const" ], "rows_examined_per_scan": 106, "rows_produced_per_join": 106, "filtered": "100.00", "cost_info": { "read_cost": "106.00", "eval_cost": "21.20", "prefix_cost": "127.20", "data_read_per_join": "602K" }, "used_columns": [ "id", "order_id", "customer_id" ], "attached_condition": "(`db_order`.`oid`.`order_id` is not null)" } }, { "table": { "table_name": "o", "access_type": "eq_ref", "possible_keys": [ "PRIMARY", "order_id", "main_order_id", "customer_id_bill", "source_order_id", "idx_is_sale_status" ], "key": "order_id", "used_key_parts": [ "order_id" ], "key_length": "82", "ref": [ "db_order.oid.order_id" ], "rows_examined_per_scan": 1, "rows_produced_per_join": 106, "filtered": "100.00", "using_index": true, "cost_info": { "read_cost": "106.00", "eval_cost": "21.20", "prefix_cost": "254.40", "data_read_per_join": "1M" }, "used_columns": [ "id", "order_id" ] } }, { "table": { "table_name": "odtl", "access_type": "ref", "possible_keys": [ "idx_order" ], "key": "idx_order", "used_key_parts": [ "order_id" ], "key_length": "82", "ref": [ "db_order.oid.order_id" ], "rows_examined_per_scan": 5, "rows_produced_per_join": 589, "filtered": "100.00", "using_index": true, "cost_info": { "read_cost": "111.26", "eval_cost": "117.98", "prefix_cost": "483.64", "data_read_per_join": "179K" }, "used_columns": [ "id", "order_id" ] } } ] } }}1 row in set, 1 warning (0.00 sec)mysql> 到此处完成sql的优化