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
直接上代码:
# coding=utf-8 from __future__ import absolute_import from __future__ import division from __future__ import print_function from __future__ import unicode_literals import sqlparse from sqlparse.sql import Identifier, IdentifierList from sqlparse.tokens import Keyword, Name RESULT_OPERATIONS = {'UNION', 'INTERSECT', 'EXCEPT', 'SELECT'} ON_KEYWORD = 'ON' PRECEDES_TABLE_NAME = {'FROM', 'JOIN', 'DESC', 'DESCRIBE', 'WITH'} class BaseExtractor(object): def __init__(self, sql_statement): self.sql = sqlparse.format(sql_statement, reindent=True, keyword_case='upper') self._table_names = set() self._alias_names = set() self._limit = None self._parsed = sqlparse.parse(self.stripped()) for statement in self._parsed: self.__extract_from_token(statement) self._limit = self._extract_limit_from_query(statement) self._table_names = self._table_names - self._alias_names @property def tables(self): return self._table_names @property def limit(self): return self._limit def is_select(self): return self._parsed[0].get_type() == 'SELECT' def is_explain(self): return self.stripped().upper().startswith('EXPLAIN') def is_readonly(self): return self.is_select() or self.is_explain() def stripped(self): return self.sql.strip(' \t\n;') def get_statements(self): statements = [] for statement in self._parsed: if statement: sql = str(statement).strip(' \n;\t') if sql: statements.append(sql) return statements @staticmethod def __precedes_table_name(token_value): for keyword in PRECEDES_TABLE_NAME: if keyword in token_value: return True return False @staticmethod def get_full_name(identifier): if len(identifier.tokens) > 1 and identifier.tokens[1].value == '.': return '{}.{}'.format(identifier.tokens[0].value, identifier.tokens[2].value) return identifier.get_real_name() @staticmethod def __is_result_operation(keyword): for operation in RESULT_OPERATIONS: if operation in keyword.upper(): return True return False @staticmethod def __is_identifier(token): return isinstance(token, (IdentifierList, Identifier)) def __process_identifier(self, identifier): if '(' not in '{}'.format(identifier): self._table_names.add(self.get_full_name(identifier)) return # store aliases if hasattr(identifier, 'get_alias'): self._alias_names.add(identifier.get_alias()) if hasattr(identifier, 'tokens'): # some aliases are not parsed properly if identifier.tokens[0].ttype == Name: self._alias_names.add(identifier.tokens[0].value) self.__extract_from_token(identifier) def as_create_table(self, table_name, overwrite=False): exec_sql = '' sql = self.stripped() if overwrite: exec_sql = 'DROP TABLE IF EXISTS {};\n'.format(table_name) exec_sql += 'CREATE TABLE {} AS \n{}'.format(table_name, sql) return exec_sql def __extract_from_token(self, token): if not hasattr(token, 'tokens'): return table_name_preceding_token = False for item in token.tokens: if item.is_group and not self.__is_identifier(item): self.__extract_from_token(item) if item.ttype in Keyword: if self.__precedes_table_name(item.value.upper()): table_name_preceding_token = True continue if not table_name_preceding_token: continue if item.ttype in Keyword or item.value == ',': if (self.__is_result_operation(item.value) or item.value.upper() == ON_KEYWORD): table_name_preceding_token = False continue # FROM clause is over break if isinstance(item, Identifier): self.__process_identifier(item) if isinstance(item, IdentifierList): for token in item.tokens: if self.__is_identifier(token): self.__process_identifier(token) def _get_limit_from_token(self, token): if token.ttype == sqlparse.tokens.Literal.Number.Integer: return int(token.value) elif token.is_group: return int(token.get_token_at_offset(1).value) def _extract_limit_from_query(self, statement): limit_token = None for pos, item in enumerate(statement.tokens): if item.ttype in Keyword and item.value.lower() == 'limit': limit_token = statement.tokens[pos + 2] return self._get_limit_from_token(limit_token) def get_query_with_new_limit(self, new_limit): if not self._limit: return self.sql + ' LIMIT ' + str(new_limit) limit_pos = None tokens = self._parsed[0].tokens # Add all items to before_str until there is a limit for pos, item in enumerate(tokens): if item.ttype in Keyword and item.value.lower() == 'limit': limit_pos = pos break limit = tokens[limit_pos + 2] if limit.ttype == sqlparse.tokens.Literal.Number.Integer: tokens[limit_pos + 2].value = new_limit elif limit.is_group: tokens[limit_pos + 2].value = ( '{}, {}'.format(next(limit.get_identifiers()), new_limit) ) str_res = '' for i in tokens: str_res += str(i.value) return str_res class SqlExtractor(BaseExtractor): """提取sql语句""" @staticmethod def get_full_name(identifier, including_dbs=False): if len(identifier.tokens) > 1 and identifier.tokens[1].value == '.': a = identifier.tokens[0].value b = identifier.tokens[2].value db_table = (a, b) full_tree = '{}.{}'.format(a, b) if len(identifier.tokens) == 3: return full_tree else: i = identifier.tokens[3].value c = identifier.tokens[4].value if i == ' ': return full_tree full_tree = '{}.{}.{}'.format(a, b, c) return full_tree return None, None if __name__ == '__main__': 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""" sql_extractor = SqlExtractor(sql) print(sql_extractor.sql) print(sql_extractor.tables)
输出结果:
{'mysql8.dataview_tprc.tprc_task', 'hive.bdc_dwd.dw_mk_order', 'mysql4.dataview_fenxiao.fx_order_task', 'mysql4.dataview_fenxiao.fx_order', 'hive.bdc_dwd.dw_mk_order_business', 'mysql7.dataview_trade.mk_order_merchant', 'mysql4.dataview_scrm.sc_tprc_product_info', 'hive.bdc_dwd.dw_fn_deal_asyn_order', 'hive.bdc_dwd.dw_fact_task_ss_daily', 'mysql7.dataview_trade.ddc_product_info', 'hive.bdc_dwd.dw_mk_order_status'}
格式化结果:
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
以上这篇python如何解析复杂sql,实现数据库和表的提取的实例剖析就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持。
免责声明:本站文章均来自网站采集或用户投稿,网站不提供任何软件下载或自行开发的软件! 如有用户或公司发现本站内容信息存在侵权行为,请邮件告知! 858582#qq.com
RTX 5090要首发 性能要翻倍!三星展示GDDR7显存
三星在GTC上展示了专为下一代游戏GPU设计的GDDR7内存。
首次推出的GDDR7内存模块密度为16GB,每个模块容量为2GB。其速度预设为32 Gbps(PAM3),但也可以降至28 Gbps,以提高产量和初始阶段的整体性能和成本效益。
据三星表示,GDDR7内存的能效将提高20%,同时工作电压仅为1.1V,低于标准的1.2V。通过采用更新的封装材料和优化的电路设计,使得在高速运行时的发热量降低,GDDR7的热阻比GDDR6降低了70%。
更新日志
- 群星.2003-存为爱2CD【环球】【WAV+CUE】
- 韩磊《试音天碟》高清音频[WAV+CUE]
- 邓涛《寂寞蒲公英(黑胶CD)》[WAV]
- 江志丰.2011-爱你的理由【豪记】【WAV+CUE
- 群星《传承-太平洋影音45周年纪念版 (CD2)》[320K/MP3][140.01MB]
- 群星《传承-太平洋影音45周年纪念版 (CD2)》[FLAC/分轨][293.29MB]
- 首首经典《滚石红人堂I 一人一首成名曲 4CD》[WAV+CUE][2.5G]
- s14上单t0梯度怎么排名 s14世界赛上单t0梯度排行榜
- tes目前进了几次s赛 LPL队伍tes参加全球总决赛次数总览
- 英雄联盟巅峰礼赠什么时候开始 2024巅峰礼赠活动时间介绍
- 冯骥发文谈睡觉重要性 网友打趣:求求你先做DLC
- 博主惊叹《少女前线2》万圣节大雷皮肤:这真能过审吗?
- 《生化危机8》夫人比基尼Mod再引骂战:夸张身材有错吗?
- 江蕙.1994-悲情歌声【点将】【WAV+CUE】
- 戴娆.2006-绽放【易柏文化】【WAV+CUE】