从4.75s到0.6s，我仅动了一条SQL
I
京东物流技术团队约 14 分钟
京东物流技术团队约 14 分钟一、前言
软件在持续的开发和维护过程中，会不断添加新功能和修复旧的缺陷，这往往伴随着代码的快速增长和复杂性的提升。若代码库没有得到良好的管理和重构，就可能积累大量的技术债务，包括不一致的设计、冗余代码、过时的库和框架以及不再使用的功能。这些因素都会导致软件结构的脆弱，增加系统出错的可能性，我们俗称为“代码腐化”，持续性的重构是一种好的解决方案。SQL也是我们常用的代码语言，虽然SQL本身作为一种标准化的查询语言不会"腐化"，但是使用SQL编写的数据库应用程序、查询和架构确实可能会因时间推移而面临类似于代码腐化的问题。
平台技术部一直坚持做稳定性建设，其中慢SQL就作为一个核心指标在治理。在治理进入深水区时，就会啃到因“SQL腐化”引入的复杂SQL治理这种硬骨头。本文以一个案例为依托来看看怎样像重构Java等高级编程语言一样来重构SQL。
select count(*) total_count from (select * from (select a.line_store_goods_id as line_resource_id, a.group_num as group_num, a.approval_erp as approval_erp, a.approval_person as approval_person, a.approval_status as approval_status, a.approval_time as approval_time, a.approval_remark as approval_remark, a.master_slave as master_slave, a.parent_line_code as parent_line_code, a.remarks as remarks, a.operator_time, a.same_stowage as same_stowage, b.start_org_id, b.start_org_name, b.start_province_id, b.start_province_name, b.start_city_id, b.start_city_name, b.start_node_code, b.start_node_name, b.end_org_id, b.end_org_name, b.end_province_id, b.end_province_name, b.end_city_id, b.end_city_name, b.end_node_code, b.end_node_name, b.depart_time, b.arrive_time, b.depart_wave_code, b.arrive_wave_code, b.enable_time, b.disable_time, a.store_enable_time, a.store_disable_time, a.update_name operator_name, b.line_code, b.line_type, b.transport_type, IF(a.store_enable_time > b.enable_time, IF(a.store_enable_time > c.enable_time, a.store_enable_time, c.enable_time), IF(b.enable_time > c.enable_time, b.enable_time, c.enable_time)) as insect_start_time, IF(a.store_disable_time < b.disable_time, IF(a.store_disable_time < c.disable_time, a.store_disable_time, c.disable_time), IF(b.disable_time < c.disable_time, b.disable_time, c.disable_time)) as insect_end_time FROM (select * FROM line_store_goods WHERE yn = 1 and master_slave = 1) a join (select start_org_id, start_org_name, start_province_id, start_province_name, start_city_id, start_city_name, start_node_code, start_node_name, end_org_id, end_org_name, end_province_id, end_province_name, end_city_id, end_city_name, end_node_code, end_node_name, depart_time, arrive_time, depart_wave_code, arrive_wave_code, line_code, line_type, transport_type, min(enable_time) as enable_time, max(disable_time) as disable_time from line_resource where line_code in (select line_code from line_store_goods WHERE yn = 1 ) and yn=1 group by line_code) b ON a.line_code = b.line_code and a.start_node_code = b.start_node_code join (select line_code,start_node_code, min(enable_time) as enable_time, max(disable_time) as disable_time from line_resource WHERE yn = 1 group by line_code) c ON a.parent_line_code = c.line_code and a.start_node_code = c.start_node_code) temp WHERE start_node_code = '311F001' and disable_time > '2023-11-15 00:00:00' and enable_time < disable_time) t_total;select count(*) total_count from (select * from (select a.line_store_goods_id as line_resource_id, a.group_num as group_num, a.approval_erp as approval_erp, a.approval_person as approval_person, a.approval_status as approval_status, a.approval_time as approval_time, a.approval_remark as approval_remark, a.master_slave as master_slave, a.parent_line_code as parent_line_code, a.remarks as remarks, a.operator_time, a.same_stowage as same_stowage, b.start_org_id, b.start_org_name, b.start_province_id, b.start_province_name, b.start_city_id, b.start_city_name, b.start_node_code, b.start_node_name, b.end_org_id, b.end_org_name, b.end_province_id, b.end_province_name, b.end_city_id, b.end_city_name, b.end_node_code, b.end_node_name, b.depart_time, b.arrive_time, b.depart_wave_code, b.arrive_wave_code, b.enable_time, b.disable_time, a.store_enable_time, a.store_disable_time, a.update_name operator_name, b.line_code, b.line_type, b.transport_type, IF(a.store_enable_time > b.enable_time, IF(a.store_enable_time > c.enable_time, a.store_enable_time, c.enable_time), IF(b.enable_time > c.enable_time, b.enable_time, c.enable_time)) as insect_start_time, IF(a.store_disable_time < b.disable_time, IF(a.store_disable_time < c.disable_time, a.store_disable_time, c.disable_time), IF(b.disable_time < c.disable_time, b.disable_time, c.disable_time)) as insect_end_time FROM (select * FROM line_store_goods WHERE yn = 1 and master_slave = 1) a join (select start_org_id, start_org_name, start_province_id, start_province_name, start_city_id, start_city_name, start_node_code, start_node_name, end_org_id, end_org_name, end_province_id, end_province_name, end_city_id, end_city_name, end_node_code, end_node_name, depart_time, arrive_time, depart_wave_code, arrive_wave_code, line_code, line_type, transport_type, min(enable_time) as enable_time, max(disable_time) as disable_time from line_resource where line_code in (select line_code from line_store_goods WHERE yn = 1 ) and yn=1 group by line_code) b ON a.line_code = b.line_code and a.start_node_code = b.start_node_code join (select line_code,start_node_code, min(enable_time) as enable_time, max(disable_time) as disable_time from line_resource WHERE yn = 1 group by line_code) c ON a.parent_line_code = c.line_code and a.start_node_code = c.start_node_code) temp WHERE start_node_code = '311F001' and disable_time > '2023-11-15 00:00:00' and enable_time < disable_time) t_total;经过格式化之后，能简单判断出SQL的功能是检索满足某条件的线路数量统计。
注意：格式化作为一个重要的工具可以在任意阶段发生作用。
step2.分层拆解
level0
select count(*) total_count from t_totallevel1 - t_total
select * from temp WHERE start_node_code = '311F001' and disable_time > '2023-11-15 00:00:00' and enable_time < disable_timelevel2 - temp
select a.line_store_goods_id as line_resource_id, a.group_num as group_num, a.approval_erp as approval_erp, a.approval_person as approval_person, a.approval_status as approval_status, a.approval_time as approval_time, a.approval_remark as approval_remark, a.master_slave as master_slave, a.parent_line_code as parent_line_code, a.remarks as remarks, a.operator_time, a.same_stowage as same_stowage, b.start_org_id, b.start_org_name, b.start_province_id, b.start_province_name, b.start_city_id, b.start_city_name, b.start_node_code, b.start_node_name, b.end_org_id, b.end_org_name, b.end_province_id, b.end_province_name, b.end_city_id, b.end_city_name, b.end_node_code, b.end_node_name, b.depart_time, b.arrive_time, b.depart_wave_code, b.arrive_wave_code, b.enable_time, b.disable_time, a.store_enable_time, a.store_disable_time, a.update_name operator_name, b.line_code, b.line_type, b.transport_type, IF(a.store_enable_time > b.enable_time, IF(a.store_enable_time > c.enable_time, a.store_enable_time, c.enable_time), IF(b.enable_time > c.enable_time, b.enable_time, c.enable_time)) as insect_start_time, IF(a.store_disable_time < b.disable_time, IF(a.store_disable_time < c.disable_time, a.store_disable_time, c.disable_time), IF(b.disable_time < c.disable_time, b.disable_time, c.disable_time)) as insect_end_time FROM join_tablelevel3 - join_table
(select * FROM line_store_goods WHERE yn = 1 and master_slave = 1) a join (select start_org_id, start_org_name, start_province_id, start_province_name, start_city_id, start_city_name, start_node_code, start_node_name, end_org_id, end_org_name, end_province_id, end_province_name, end_city_id, end_city_name, end_node_code, end_node_name, depart_time, arrive_time, depart_wave_code, arrive_wave_code, line_code, line_type, transport_type, min(enable_time) as enable_time, max(disable_time) as disable_time from line_resource where line_code in (select line_code from line_store_goods WHERE yn = 1 ) and yn=1 group by line_code) b ON a.line_code = b.line_code and a.start_node_code = b.start_node_code join (select line_code,start_node_code, min(enable_time) as enable_time, max(disable_time) as disable_time from line_resource WHERE yn = 1 group by line_code) c ON a.parent_line_code = c.line_code and a.start_node_code = c.start_node_codelevel4 - a,b,c
select * FROM line_store_goods WHERE yn = 1 and master_slave = 1select start_org_id, start_org_name, start_province_id, start_province_name, start_city_id, start_city_name, start_node_code, start_node_name, end_org_id, end_org_name, end_province_id, end_province_name, end_city_id, end_city_name, end_node_code, end_node_name, depart_time, arrive_time, depart_wave_code, arrive_wave_code, line_code, line_type, transport_type, min(enable_time) as enable_time, max(disable_time) as disable_time from line_resource where line_code in (select line_code from line_store_goods WHERE yn = 1 ) and yn=1 group by line_codeselect line_code,start_node_code, min(enable_time) as enable_time, max(disable_time) as disable_time from line_resource WHERE yn = 1 group by line_code
层级合并：减少临时表个数 条件下推：减少检索行数&临时表大小 join优化：减少检索行数&临时表大小 子查询删除：减少临时表个数 子查询与join的相互转换：减少检索行数
重构1 - 层级合并
level0 & level1
select count(*) total_count from (select * from temp where a = "1")select count(*) from temp where a = "1"select count(*) from ( (select line_code FROM line_store_goods WHERE yn = 1 and parent_line_code = line_code and master_slave = 1 and start_node_code = '311F001') a join (select line_code, min(enable_time) as enable_time, max(disable_time) as disable_time from line_resource where yn=1 and start_node_code = '311F001' group by line_code) b ON a.line_code = b.line_code) where disable_time > '2023-11-15 00:00:00' and enable_time < disable_time重构后的SQL具备良好的可读性，基于此很容易反推出SQL的业务功能。基于此与其理论应用场景做是否匹配的理论判断很重要。有的时候生产上的SQL不一定是正确的，因为部分场景下可用性并不完全等价于正确性。
另外，上面的技术能不用就不用，好的前置设计胜过事后的十八般武艺。
作者：京东物流 崔立群
来源：京东云开发者社区 https://www.cnblogs.com/Jcloud/p/17994104
- END -
IT168与ITPUB技术社区强强联手，收集数百款主流数据库产品，重磅推出“数据库全景图”，旨在打造一款集知识普及、产品对比、选型参考于一体的综合性资源平台。“数据库全景图（11月版）”可扫描上方左侧二维码回复关键词获取，识别右侧二维码直达“数据库全景图”链接（右上角浏览器打开获取更好体验）。