python如何解析复杂sql,实现数据库和表的提取的实例剖析-木庄网络博客

本文整理自网络，侵删。

需求：

公司的数据分析师，提交一个sql, 一般都三四百行。由于数据安全的需要，不能开放所有的数据库和数据表给数据分析师查询，所以需要解析sql中的数据库和表，与权限管理系统中记录的数据库和表权限信息比对，实现非法查询的拦截。

解决办法：

在解决这个问题前，现在github找了一下轮子，发现python下面除了sql parse没什么好的解析数据库和表的轮轮。到是在java里面找到presto-parser解析的比较准。于是自己结合sql parse源码写了个类，供大家参考，测试了一下，检测还是准的。

测试sql

select
b.product_name "产品",
count(a.order_id) "订单量",
b.selling_price_max "销售价",
b.gross_profit_rate_max/100 "毛利率",
case when b.business_type =1 then '自营消化' when b.business_type =2 then '服务商消化' end "消化模式"
from(select 'CRM签单' label,date(d.update_ymd) close_ymd,c.product_name,c.product_id,
  a.order_id,cast(a.recipient_amount as double) amt,d.cost
  from mysql4.dataview_fenxiao.fx_order a
  left join mysql4.dataview_fenxiao.fx_order_task b on a.order_id = b.order_id
  left join mysql7.dataview_trade.ddc_product_info c on cast(c.product_id as varchar) = a.product_ids and c.snapshot_version = 'SELLING'
  inner join (select t1.par_order_id,max(t1.update_ymd) update_ymd,
        sum(case when t4.product2_type = 1 and t5.shop_id is not null then t5.price else t1.order_hosted_price end) cost
        from hive.bdc_dwd.dw_mk_order t1
        left join hive.bdc_dwd.dw_mk_order_status t2 on t1.order_id = t2.order_id and t2.acct_day = substring(cast(DATE_ADD('day',-1,CURRENT_DATE) as varchar),9,2)
        left join mysql7.dataview_trade.mk_order_merchant t3 on t1.order_id = t3.order_id
        left join mysql7.dataview_trade.ddc_product_info t4 on t4.product_id = t3.MERCHANT_ID and t4.snapshot_version = 'SELLING'
        left join mysql4.dataview_scrm.sc_tprc_product_info t5 on t5.product_id = t4.product_id and t5.shop_id = t1.seller_id
        where t1.acct_day = substring(cast(DATE_ADD('day',-1,CURRENT_DATE) as varchar),9,2)
        and t2.valid_state in (100,200) ------有效订单
        and t1.order_mode = 10  --------产品消耗订单
        and t2.complete_state = 1 -----订单已经完成
        group by t1.par_order_id
  ) d on d.par_order_id = b.task_order_id
  where c.product_type = 0 and date(from_unixtime(a.last_recipient_time)) > date('2016-01-01') and a.payee_type <> 1 -----------已收款
  UNION ALL
  select '企业管家消耗' label,date(c.update_ymd) close_ymd,b.product_name,b.product_id,
  a.task_id,(case when a.yb_price = 0 and b.product2_type = 1 then b.selling_price_min else a.yb_price end) amt,
  (case when a.yb_price = 0 and b.product2_type = 2 then 0 when b.product2_type = 1 and e.shop_id is not null then e.price else c.order_hosted_price end) cost
  from mysql8.dataview_tprc.tprc_task a
  left join mysql7.dataview_trade.ddc_product_info b on a.product_id = b.product_id and b.snapshot_version = 'SELLING'
  inner join hive.bdc_dwd.dw_mk_order c on a.order_id = c.order_id and c.acct_day = substring(cast(DATE_ADD('day',-1,CURRENT_DATE) as varchar),9,2)
  left join hive.bdc_dwd.dw_mk_order_status d on d.order_id = c.order_id and d.acct_day = substring(cast(DATE_ADD('day',-1,CURRENT_DATE) as varchar),9,2)
  left join mysql4.dataview_scrm.sc_tprc_product_info e on e.product_id = b.product_id and e.shop_id = c.seller_id
  where d.valid_state in (100,200) and d.complete_state = 1 and c.order_mode = 10
  union ALL
  select '交易管理系统' label,date(t6.close_ymd) close_ymd,t4.product_name,t4.product_id,
  t1.order_id,(t1.order_hosted_price-t1.order_refund_price) amt,
  (case when t1.order_mode <> 11 then t7.user_amount when t1.order_mode = 11 and t4.product2_type = 1 and t5.shop_id is not null then t5.price else t8.cost end) cost
  from hive.bdc_dwd.dw_mk_order t1
  left join hive.bdc_dwd.dw_mk_order_business t2 on t1.order_id = t2.order_id and t2.acct_day=substring(cast(DATE_ADD('day',-1,CURRENT_DATE) as varchar),9,2)
  left join mysql7.dataview_trade.mk_order_merchant t3 on t1.order_id = t3.order_id
  left join mysql7.dataview_trade.ddc_product_info t4 on t4.product_id = t3.MERCHANT_ID and t4.snapshot_version = 'SELLING'
  left join mysql4.dataview_scrm.sc_tprc_product_info t5 on t5.product_id = t4.product_id and t5.shop_id = t1.seller_id
  left join hive.bdc_dwd.dw_fact_task_ss_daily t6 on t6.task_id = t2.task_id and t6.acct_time=date_format(date_add('day',-1,current_date),'%Y-%m-%d')
  left join (select a.task_id,sum(a.user_amount) user_amount
        from hive.bdc_dwd.dw_fn_deal_asyn_order a
        where a.is_new=1 and a.service='Trade_Payment' and a.state=1 and a.acct_day=substring(cast(DATE_ADD('day',-1,CURRENT_DATE) as varchar),9,2)
        group by a.task_id)t7 on t7.task_id = t2.task_id     
  left join (select t1.par_order_id,sum(t1.order_hosted_price - t1.order_refund_price) cost
        from hive.bdc_dwd.dw_mk_order t1
        where t1.acct_day = substring(cast(DATE_ADD('day',-1,CURRENT_DATE) as varchar),9,2) and t1.order_type = 1 and t1.order_stype = 4 and t1.order_mode = 12
        group by t1.par_order_id) t8 on t1.order_id = t8.par_order_id
  where t1.acct_day = substring(cast(DATE_ADD('day',-1,CURRENT_DATE) as varchar),9,2)
  and t1.order_type = 1 and t1.order_stype in (4,5) and t1.order_mode <> 12 and t4.product_id is not null and t1.order_hosted_price > 0 and t6.is_deal = 1 and t6.close_ymd >= '2018-12-31'
)a
left join mysql7.dataview_trade.ddc_product_info b on a.product_id = b.product_id and b.snapshot_version = 'SELLING'
where b.product2_type = 1 -------标品
and close_ymd between DATE_ADD('day',-7,CURRENT_DATE) and DATE_ADD('day',-1,CURRENT_DATE)
GROUP BY b.product_name,
b.selling_price_max,
b.gross_profit_rate_max/100,
b.actrul_supply_num,
case when b.business_type =1 then '自营消化' when b.business_type =2 then '服务商消化' end
order by count(a.order_id) desc
limit 10

可以看到该sql比较杂，也没有格式化，不太好提取数据库和表。所以第一步需要对sql进行格式化

阅读剩余部分