首页 文章 精选 留言 我的

精选列表

搜索[智能问数]，共10000篇文章
优秀的个人博客,低调大师

VLAN 和广播域到底怎么数，关键不在交换机有几台

一台交换机接着财务、研发和访客区的电脑。财务电脑发出一条 ARP 广播，询问默认网关的 MAC 地址，研发电脑和访客电脑会不会收到？这个问题比数交换机端口更能说明 VLAN 的作用。假如三组电脑都处于同一个普通二层网络，交换机会把这条广播转发到该网络的其他相关端口。假如三组电脑分别属于 VLAN 10、VLAN 20 和 VLAN 30，财务电脑的广播就留在 VLAN 10 内，不会因为另外两组电脑也插在同一台交换机上，就跑到它们的端口。
2026-09-26
优秀的个人博客,低调大师

数仓实践丨常量标量子查询做全连接导致整体慢

本文分享自华为云社区《GaussDB(DWS)性能调优：常量标量子查询做全连接导致整体慢》，作者： Zawami 。 问题描述 由于SQL中存在标量子查询同另一查询做笛卡尔积使SQL整体慢。标量子查询，即结果集只有一行一列的子查询。这里导致的SQL语句执行慢不只是在于做笛卡尔积慢，也会使后续聚合更慢。 原始语句 WITH TMP AS( SELECT case when length('[“202309“]') = 6 then '[“202309“]' || '01' WHEN length('[“202309“]') <> 8 THEN TO_CHAR(CURRENT_DATE, 'YYYYMMDD') END AS V_DATE from DUAL ) SELECT BG_CODE, BG_CN_NAME, BG_EN_NAME, METRIC_CODE --指标ID , METRIC_CN_NAME --指标中文名称 , METRIC_EN_NAME --指标英文名称 , CURRENCY --币种 , OVERSEAS_FLAG, REGION_CODE, REGION_CN_NAME, REGION_EN_NAME, REPOFFICE_CODE, REPOFFICE_CN_NAME, REPOFFICE_EN_NAME, OFFICE_CODE, OFFICE_CN_NAME, OFFICE_EN_NAME, REGION_CUSTCATG_CODE, REGION_CUSTCATG_CN_NAME, REGION_CUSTCATG_EN_NAME, TOP_CUST_CATEGORY_CODE, TOP_CUST_CATEGORY_EN_NAME, TOP_CUST_CATEGORY_CN_NAME, ACCTCUST_HQ_CODE, ACCTCUST_HQ_CN_NAME, ACCTCUST_HQ_EN_NAME, ACCTCUST_BRANCH_CODE, ACCTCUST_BRANCH_CN_NAME, ACCTCUST_BRANCH_EN_NAME, ACCTCUST_SUBSIDIARY_CODE, ACCTCUST_SUBSIDIARY_CN_NAM, ACCTCUST_SUBSIDIARY_EN_NAM, COUNTRY_CODE --新增加入参 , COUNTRY_CN_NAME --新增加入参 , COUNTRY_EN_NAME --新增加入参 , AGREE_AMOUNT --BUSI_DSCT_00001 总优惠 , AGREE_REMAIN_AMOUNT --BUSI_DSCT_00002 即期优惠 , SIGN_AMOUNT --BUSI_DSCT_00003 即期优惠/一次性优惠 , USE_AMOUNT --BUSI_DSCT_00004 即期优惠/单价量折扣 , NOT_USED_VALID_AMOUNT --BUSI_DSCT_00005 延期优惠 , NOT_USED_INVALID_AMOUNT --BUSI_DSCT_00006 voucher , NEW_SIGN_AMOUNT --BUSI_DSCT_00007 其他延期优惠 , NEW_USE_AMOUNT --BUSI_DSCT_00008 本月新使用金额 , EXPIRED_AMOUNT --BUSI_DSCT_00009 本月已过期金额 , IMMED_EXPIRED_AMOUNT --BUSI_DSCT_00010 半年内即将过期金额 FROM ( SELECT C.BG_CODE, C.BG_CN_NAME, C.BG_EN_NAME, C.M_ID AS METRIC_CODE --指标ID , C.M_CN AS METRIC_CN_NAME --指标中文名称 , C.M_EN AS METRIC_EN_NAME --指标英文名称 , C.CURRENCY_CODE AS CURRENCY --币种 ,CASE WHEN 1 = 0 THEN C.OVERSEA_FLAG ELSE NULL END AS OVERSEAS_FLAG,CASE WHEN 1 = 0 THEN C.REGION_CODE ELSE NULL END AS REGION_CODE,CASE WHEN 1 = 0 THEN C.REGION_CN_NAME ELSE NULL END AS REGION_CN_NAME,CASE WHEN 1 = 0 THEN C.REGION_EN_NAME ELSE NULL END AS REGION_EN_NAME,CASE WHEN 1 = 0 THEN C.REPOFFICE_CODE ELSE NULL END AS REPOFFICE_CODE,CASE WHEN 1 = 0 THEN C.REPOFFICE_CN_NAME ELSE NULL END AS REPOFFICE_CN_NAME,CASE WHEN 1 = 0 THEN C.REPOFFICE_EN_NAME ELSE NULL END AS REPOFFICE_EN_NAME,CASE WHEN 1 = 0 THEN C.OFFICE_CODE ELSE NULL END AS OFFICE_CODE,CASE WHEN 1 = 0 THEN C.OFFICE_CN_NAME ELSE NULL END AS OFFICE_CN_NAME,CASE WHEN 1 = 0 THEN C.OFFICE_EN_NAME ELSE NULL END AS OFFICE_EN_NAME,CASE WHEN 1 = 0 THEN C.REGION_CUSTCATG_CODE ELSE NULL END AS REGION_CUSTCATG_CODE,CASE WHEN 1 = 0 THEN C.REGION_CUSTCATG_CN_NAME ELSE NULL END AS REGION_CUSTCATG_CN_NAME,CASE WHEN 1 = 0 THEN C.REGION_CUSTCATG_EN_NAME ELSE NULL END AS REGION_CUSTCATG_EN_NAME,CASE WHEN 1 = 0 THEN C.TOP_CUST_CATEGORY_CODE ELSE NULL END AS TOP_CUST_CATEGORY_CODE,CASE WHEN 1 = 0 THEN C.TOP_CUST_CATEGORY_EN_NAME ELSE NULL END AS TOP_CUST_CATEGORY_EN_NAME,CASE WHEN 1 = 0 THEN C.TOP_CUST_CATEGORY_CN_NAME ELSE NULL END AS TOP_CUST_CATEGORY_CN_NAME,CASE WHEN 1 = 0 THEN C.ACCTCUST_HQ_CODE ELSE NULL END AS ACCTCUST_HQ_CODE,CASE WHEN 1 = 0 THEN C.ACCTCUST_HQ_CN_NAME ELSE NULL END AS ACCTCUST_HQ_CN_NAME,CASE WHEN 1 = 0 THEN C.ACCTCUST_HQ_EN_NAME ELSE NULL END AS ACCTCUST_HQ_EN_NAME,CASE WHEN 1 = 0 THEN C.ACCTCUST_BRANCH_CODE ELSE NULL END AS ACCTCUST_BRANCH_CODE,CASE WHEN 1 = 0 THEN C.ACCTCUST_BRANCH_CN_NAME ELSE NULL END AS ACCTCUST_BRANCH_CN_NAME,CASE WHEN 1 = 0 THEN C.ACCTCUST_BRANCH_EN_NAME ELSE NULL END AS ACCTCUST_BRANCH_EN_NAME,CASE WHEN 1 = 0 THEN C.ACCTCUST_SUBSIDIARY_CODE ELSE NULL END AS ACCTCUST_SUBSIDIARY_CODE,CASE WHEN 1 = 0 THEN C.ACCTCUST_SUBSIDIARY_CN_NAM ELSE NULL END AS ACCTCUST_SUBSIDIARY_CN_NAM,CASE WHEN 1 = 0 THEN C.ACCTCUST_SUBSIDIARY_EN_NAM ELSE NULL END AS ACCTCUST_SUBSIDIARY_EN_NAM,CASE WHEN 1 = 0 THEN C.COUNTRY_CODE ELSE NULL END AS COUNTRY_CODE --新增加入参 ,CASE WHEN 1 = 0 THEN C.COUNTRY_CN_NAME ELSE NULL END AS COUNTRY_CN_NAME --新增加入参 ,CASE WHEN 1 = 0 THEN C.COUNTRY_EN_NAME ELSE NULL END AS COUNTRY_EN_NAME --新增加入参 , SUM(C.AGREE_AMOUNT) AS AGREE_AMOUNT --协议金额 , SUM(C.AGREE_REMAIN_AMOUNT) AS AGREE_REMAIN_AMOUNT --协议剩余金额 , SUM(C.SIGN_AMOUNT) AS SIGN_AMOUNT --可用金额 , SUM(C.USE_AMOUNT) AS USE_AMOUNT --已使用金额 , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND NVL( C.EXPIRED_DATE, add_months(C.EFFECTIVE_DATE, C.VALID_MONTH) ) >= to_date(T.V_DATE, 'yyyymmdd') THEN C.AGREE_REMAIN_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_END_DATE >= to_date(T.V_DATE, 'yyyymmdd') THEN C.EFFECTIVE_TOTAL_AMOUNT ELSE NULL END ) AS NOT_USED_VALID_AMOUNT --未使用金额(有效期外) , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' THEN C.SIGN_AMOUNT ELSE C.EFFECTIVE_TOTAL_AMOUNT END - CASE WHEN C.DSCT_TYPE = 'VOUCHER' THEN C.USE_AMOUNT ELSE C.USED_TOTAL_AMOUNT END - CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND NVL( C.EXPIRED_DATE, add_months(C.EFFECTIVE_DATE, C.VALID_MONTH) ) >= to_date(T.V_DATE, 'yyyymmdd') THEN C.AGREE_REMAIN_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_END_DATE >= to_date(T.V_DATE, 'yyyymmdd') THEN C.EFFECTIVE_TOTAL_AMOUNT ELSE NULL END ) AS NOT_USED_INVALID_AMOUNT --未使用金额(有效期内) , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND C.EXPIRED_DATE >= to_date(substr(T.V_DATE, 1, 6), 'yyyymm') and C.EXPIRED_DATE <= LAST_DAY(to_date(T.V_DATE, 'yyyymmdd')) THEN C.SIGN_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_START_DATE >= to_date(substr(T.V_DATE, 1, 6), 'yyyymm') and C.DSCT_START_DATE <= LAST_DAY(to_date(T.V_DATE, 'yyyymmdd')) THEN C.EFFECTIVE_TOTAL_AMOUNT ELSE NULL END ) AS NEW_SIGN_AMOUNT --本月新增可用金额 , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND C.EFFECTIVE_DATE >= to_date(substr(T.V_DATE, 1, 6), 'yyyymm') and C.EFFECTIVE_DATE <= LAST_DAY(to_date(T.V_DATE, 'yyyymmdd')) THEN C.USE_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_START_DATE >= to_date(substr(T.V_DATE, 1, 6), 'yyyymm') and C.DSCT_START_DATE <= LAST_DAY(to_date(T.V_DATE, 'yyyymmdd')) THEN C.USED_TOTAL_AMOUNT ELSE NULL END ) AS NEW_USE_AMOUNT --本月新使用金额 , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND C.EXPIRED_DATE < to_date(T.V_DATE, 'yyyymmdd') THEN C.AGREE_REMAIN_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_END_DATE < to_date(T.V_DATE, 'yyyymmdd') THEN C.EFFECTIVE_TOTAL_AMOUNT ELSE NULL END ) AS EXPIRED_AMOUNT --本月已过期金额 , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND C.EXPIRED_DATE BETWEEN to_date(T.V_DATE, 'yyyymmdd') AND add_months(to_date(T.V_DATE, 'yyyymmdd'), 6) THEN C.AGREE_REMAIN_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_END_DATE BETWEEN to_date(T.V_DATE, 'yyyymmdd') AND add_months(to_date(T.V_DATE, 'yyyymmdd'), 6) THEN C.EFFECTIVE_TOTAL_AMOUNT ELSE NULL END ) AS IMMED_EXPIRED_AMOUNT --半年内即将过期金额 FROM DMSALESW.DM_SALE_BUSI_DSCT_SUM_F C LEFT JOIN TMP T ON 1 = 1 WHERE C.CURRENCY_CODE IN ('USD') --改为多值 AND C.BG_CODE IN ('PDCG901159') AND C.M_ID IN ( 'BUSI_DSCT_00001', 'BUSI_DSCT_00002', 'BUSI_DSCT_00003', 'BUSI_DSCT_00004', 'BUSI_DSCT_00005', 'BUSI_DSCT_00006', 'BUSI_DSCT_00007' ) --新增加字段 --AND C.M_CN IN ('#[#P_REPORT_ITEM_NAME#]#') --新增加字段 --新增加字段 GROUP BY C.BG_CODE, C.BG_CN_NAME, C.BG_EN_NAME, C.M_ID --指标ID , C.M_CN --指标中文名称 , C.M_EN --指标英文名称 , C.CURRENCY_CODE --币种 ,CASE WHEN 1 = 0 THEN C.OVERSEA_FLAG ELSE NULL END,CASE WHEN 1 = 0 THEN C.REGION_CODE ELSE NULL END,CASE WHEN 1 = 0 THEN C.REGION_CN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.REGION_EN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.REPOFFICE_CODE ELSE NULL END,CASE WHEN 1 = 0 THEN C.REPOFFICE_CN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.REPOFFICE_EN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.OFFICE_CODE ELSE NULL END,CASE WHEN 1 = 0 THEN C.OFFICE_CN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.OFFICE_EN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.REGION_CUSTCATG_CODE ELSE NULL END,CASE WHEN 1 = 0 THEN C.REGION_CUSTCATG_CN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.REGION_CUSTCATG_EN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.TOP_CUST_CATEGORY_CODE ELSE NULL END,CASE WHEN 1 = 0 THEN C.TOP_CUST_CATEGORY_EN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.TOP_CUST_CATEGORY_CN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.ACCTCUST_HQ_CODE ELSE NULL END,CASE WHEN 1 = 0 THEN C.ACCTCUST_HQ_CN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.ACCTCUST_HQ_EN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.ACCTCUST_BRANCH_CODE ELSE NULL END,CASE WHEN 1 = 0 THEN C.ACCTCUST_BRANCH_CN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.ACCTCUST_BRANCH_EN_NAME ELSE NULL END,CASE WHEN 1 = 0 THEN C.ACCTCUST_SUBSIDIARY_CODE ELSE NULL END,CASE WHEN 1 = 0 THEN C.ACCTCUST_SUBSIDIARY_CN_NAM ELSE NULL END,CASE WHEN 1 = 0 THEN C.ACCTCUST_SUBSIDIARY_EN_NAM ELSE NULL END,CASE WHEN 1 = 0 THEN C.COUNTRY_CODE ELSE NULL END --新增加入参 ,CASE WHEN 1 = 0 THEN C.COUNTRY_CN_NAME ELSE NULL END --新增加入参 ,CASE WHEN 1 = 0 THEN C.COUNTRY_EN_NAME ELSE NULL END ) T --新增加入参 从SQL中可以看到TMP为标量子查询，并且在子查询T中和物理表C做了笛卡尔积。 下面是该SQL的执行计划： id | operation | A-time | A-rows | E-rows | E-distinct | Peak Memory | E-memory | A-width | E-width | E-costs ----+-------------------------------------------------------------------------+----------------------+---------+---------+------------+----------------+----------+-----------+---------+----------- 1 | -> Row Adapter | 3037.648 | 7 | 245 | | 419KB | | | 1318 | 117210.62 2 | -> Vector Streaming (type: GATHER) | 3037.633 | 7 | 245 | | 777KB | | | 1318 | 117210.62 3 | -> Vector Hash Aggregate | [3031.872, 3032.516] | 7 | 245 | | [4MB, 4MB] | 16MB | [0,870] | 557 | 117128.41 4 | -> Vector Streaming(type: REDISTRIBUTE) | [3031.560, 3032.232] | 112 | 3920 | | [1MB, 1MB] | 2MB | | 557 | 116852.33 5 | -> Vector Hash Aggregate | [2728.059, 2909.255] | 112 | 3920 | | [8MB, 8MB] | 16MB | [833,833] | 557 | 116699.48 6 | -> Vector Nest Loop Left Join (7, 8) | [441.050, 471.725] | 3007901 | 2106919 | | [1MB, 1MB] | 1MB | | 237 | 67316.28 7 | -> CStore Scan on dmsalesw.dm_sale_busi_dsct_sum_f c | [145.354, 158.560] | 3007901 | 2106919 | | [5MB, 5MB] | 1MB | | 205 | 65011.82 8 | -> Vector Materialize | [32.034, 38.902] | 3007901 | 1 | | [288KB, 288KB] | 16MB | [21,21] | 32 | 0.03 9 | -> Vector Subquery Scan on dual | [0.067, 0.093] | 16 | 1 | | [128KB, 128KB] | 1MB | | 32 | 0.02 10 | -> Vector Adapter | [0.005, 0.006] | 16 | 1 | | [40KB, 40KB] | 1MB | | 0 | 0.01 11 | -> Result | [0.001, 0.002] | 16 | 1 | | [8KB, 8KB] | 1MB | | 0 | 0.01 把TMP作为一列放到T中后，性能有明显提升。 EXPLAIN PERFORMANCE SELECT BG_CODE, BG_CN_NAME, BG_EN_NAME, METRIC_CODE --指标ID , METRIC_CN_NAME --指标中文名称 , METRIC_EN_NAME --指标英文名称 , CURRENCY --币种 , OVERSEAS_FLAG, REGION_CODE, REGION_CN_NAME, REGION_EN_NAME, REPOFFICE_CODE, REPOFFICE_CN_NAME, REPOFFICE_EN_NAME, OFFICE_CODE, OFFICE_CN_NAME, OFFICE_EN_NAME, REGION_CUSTCATG_CODE, REGION_CUSTCATG_CN_NAME, REGION_CUSTCATG_EN_NAME, TOP_CUST_CATEGORY_CODE, TOP_CUST_CATEGORY_EN_NAME, TOP_CUST_CATEGORY_CN_NAME, ACCTCUST_HQ_CODE, ACCTCUST_HQ_CN_NAME, ACCTCUST_HQ_EN_NAME, ACCTCUST_BRANCH_CODE, ACCTCUST_BRANCH_CN_NAME, ACCTCUST_BRANCH_EN_NAME, ACCTCUST_SUBSIDIARY_CODE, ACCTCUST_SUBSIDIARY_CN_NAM, ACCTCUST_SUBSIDIARY_EN_NAM, COUNTRY_CODE --新增加入参 , COUNTRY_CN_NAME --新增加入参 , COUNTRY_EN_NAME --新增加入参 , AGREE_AMOUNT --BUSI_DSCT_00001 总优惠 , AGREE_REMAIN_AMOUNT --BUSI_DSCT_00002 即期优惠 , SIGN_AMOUNT --BUSI_DSCT_00003 即期优惠/一次性优惠 , USE_AMOUNT --BUSI_DSCT_00004 即期优惠/单价量折扣 , NOT_USED_VALID_AMOUNT --BUSI_DSCT_00005 延期优惠 , NOT_USED_INVALID_AMOUNT --BUSI_DSCT_00006 voucher , NEW_SIGN_AMOUNT --BUSI_DSCT_00007 其他延期优惠 , NEW_USE_AMOUNT --BUSI_DSCT_00008 本月新使用金额 , EXPIRED_AMOUNT --BUSI_DSCT_00009 本月已过期金额 , IMMED_EXPIRED_AMOUNT --BUSI_DSCT_00010 半年内即将过期金额 FROM ( SELECT case when length('[“202309“]') = 6 then '[“202309“]' || '01' WHEN length('[“202309“]') <> 8 THEN TO_CHAR(CURRENT_DATE, 'YYYYMMDD') END AS V_DATE, C.BG_CODE, C.BG_CN_NAME, C.BG_EN_NAME, C.M_ID AS METRIC_CODE --指标ID , C.M_CN AS METRIC_CN_NAME --指标中文名称 , C.M_EN AS METRIC_EN_NAME --指标英文名称 , C.CURRENCY_CODE AS CURRENCY --币种 ,CASE WHEN 1 = 0 THEN C.OVERSEA_FLAG ELSE NULL END AS OVERSEAS_FLAG,CASE WHEN 1 = 0 THEN C.REGION_CODE ELSE NULL END AS REGION_CODE,CASE WHEN 1 = 0 THEN C.REGION_CN_NAME ELSE NULL END AS REGION_CN_NAME,CASE WHEN 1 = 0 THEN C.REGION_EN_NAME ELSE NULL END AS REGION_EN_NAME,CASE WHEN 1 = 0 THEN C.REPOFFICE_CODE ELSE NULL END AS REPOFFICE_CODE,CASE WHEN 1 = 0 THEN C.REPOFFICE_CN_NAME ELSE NULL END AS REPOFFICE_CN_NAME,CASE WHEN 1 = 0 THEN C.REPOFFICE_EN_NAME ELSE NULL END AS REPOFFICE_EN_NAME,CASE WHEN 1 = 0 THEN C.OFFICE_CODE ELSE NULL END AS OFFICE_CODE,CASE WHEN 1 = 0 THEN C.OFFICE_CN_NAME ELSE NULL END AS OFFICE_CN_NAME,CASE WHEN 1 = 0 THEN C.OFFICE_EN_NAME ELSE NULL END AS OFFICE_EN_NAME,CASE WHEN 1 = 0 THEN C.REGION_CUSTCATG_CODE ELSE NULL END AS REGION_CUSTCATG_CODE,CASE WHEN 1 = 0 THEN C.REGION_CUSTCATG_CN_NAME ELSE NULL END AS REGION_CUSTCATG_CN_NAME,CASE WHEN 1 = 0 THEN C.REGION_CUSTCATG_EN_NAME ELSE NULL END AS REGION_CUSTCATG_EN_NAME,CASE WHEN 1 = 0 THEN C.TOP_CUST_CATEGORY_CODE ELSE NULL END AS TOP_CUST_CATEGORY_CODE,CASE WHEN 1 = 0 THEN C.TOP_CUST_CATEGORY_EN_NAME ELSE NULL END AS TOP_CUST_CATEGORY_EN_NAME,CASE WHEN 1 = 0 THEN C.TOP_CUST_CATEGORY_CN_NAME ELSE NULL END AS TOP_CUST_CATEGORY_CN_NAME,CASE WHEN 1 = 0 THEN C.ACCTCUST_HQ_CODE ELSE NULL END AS ACCTCUST_HQ_CODE,CASE WHEN 1 = 0 THEN C.ACCTCUST_HQ_CN_NAME ELSE NULL END AS ACCTCUST_HQ_CN_NAME,CASE WHEN 1 = 0 THEN C.ACCTCUST_HQ_EN_NAME ELSE NULL END AS ACCTCUST_HQ_EN_NAME,CASE WHEN 1 = 0 THEN C.ACCTCUST_BRANCH_CODE ELSE NULL END AS ACCTCUST_BRANCH_CODE,CASE WHEN 1 = 0 THEN C.ACCTCUST_BRANCH_CN_NAME ELSE NULL END AS ACCTCUST_BRANCH_CN_NAME,CASE WHEN 1 = 0 THEN C.ACCTCUST_BRANCH_EN_NAME ELSE NULL END AS ACCTCUST_BRANCH_EN_NAME,CASE WHEN 1 = 0 THEN C.ACCTCUST_SUBSIDIARY_CODE ELSE NULL END AS ACCTCUST_SUBSIDIARY_CODE,CASE WHEN 1 = 0 THEN C.ACCTCUST_SUBSIDIARY_CN_NAM ELSE NULL END AS ACCTCUST_SUBSIDIARY_CN_NAM,CASE WHEN 1 = 0 THEN C.ACCTCUST_SUBSIDIARY_EN_NAM ELSE NULL END AS ACCTCUST_SUBSIDIARY_EN_NAM,CASE WHEN 1 = 0 THEN C.COUNTRY_CODE ELSE NULL END AS COUNTRY_CODE --新增加入参 ,CASE WHEN 1 = 0 THEN C.COUNTRY_CN_NAME ELSE NULL END AS COUNTRY_CN_NAME --新增加入参 ,CASE WHEN 1 = 0 THEN C.COUNTRY_EN_NAME ELSE NULL END AS COUNTRY_EN_NAME --新增加入参 , SUM(C.AGREE_AMOUNT) AS AGREE_AMOUNT --协议金额 , SUM(C.AGREE_REMAIN_AMOUNT) AS AGREE_REMAIN_AMOUNT --协议剩余金额 , SUM(C.SIGN_AMOUNT) AS SIGN_AMOUNT --可用金额 , SUM(C.USE_AMOUNT) AS USE_AMOUNT --已使用金额 , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND NVL( C.EXPIRED_DATE, add_months(C.EFFECTIVE_DATE, C.VALID_MONTH) ) >= to_date(V_DATE, 'yyyymmdd') THEN C.AGREE_REMAIN_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_END_DATE >= to_date(V_DATE, 'yyyymmdd') THEN C.EFFECTIVE_TOTAL_AMOUNT ELSE NULL END ) AS NOT_USED_VALID_AMOUNT --未使用金额(有效期外) , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' THEN C.SIGN_AMOUNT ELSE C.EFFECTIVE_TOTAL_AMOUNT END - CASE WHEN C.DSCT_TYPE = 'VOUCHER' THEN C.USE_AMOUNT ELSE C.USED_TOTAL_AMOUNT END - CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND NVL( C.EXPIRED_DATE, add_months(C.EFFECTIVE_DATE, C.VALID_MONTH) ) >= to_date(V_DATE, 'yyyymmdd') THEN C.AGREE_REMAIN_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_END_DATE >= to_date(V_DATE, 'yyyymmdd') THEN C.EFFECTIVE_TOTAL_AMOUNT ELSE NULL END ) AS NOT_USED_INVALID_AMOUNT --未使用金额(有效期内) , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND C.EXPIRED_DATE >= to_date(substr(V_DATE, 1, 6), 'yyyymm') and C.EXPIRED_DATE <= LAST_DAY(to_date(V_DATE, 'yyyymmdd')) THEN C.SIGN_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_START_DATE >= to_date(substr(V_DATE, 1, 6), 'yyyymm') and C.DSCT_START_DATE <= LAST_DAY(to_date(V_DATE, 'yyyymmdd')) THEN C.EFFECTIVE_TOTAL_AMOUNT ELSE NULL END ) AS NEW_SIGN_AMOUNT --本月新增可用金额 , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND C.EFFECTIVE_DATE >= to_date(substr(V_DATE, 1, 6), 'yyyymm') and C.EFFECTIVE_DATE <= LAST_DAY(to_date(V_DATE, 'yyyymmdd')) THEN C.USE_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_START_DATE >= to_date(substr(V_DATE, 1, 6), 'yyyymm') and C.DSCT_START_DATE <= LAST_DAY(to_date(V_DATE, 'yyyymmdd')) THEN C.USED_TOTAL_AMOUNT ELSE NULL END ) AS NEW_USE_AMOUNT --本月新使用金额 , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND C.EXPIRED_DATE < to_date(V_DATE, 'yyyymmdd') THEN C.AGREE_REMAIN_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_END_DATE < to_date(V_DATE, 'yyyymmdd') THEN C.EFFECTIVE_TOTAL_AMOUNT ELSE NULL END ) AS EXPIRED_AMOUNT --本月已过期金额 , SUM( CASE WHEN C.DSCT_TYPE = 'VOUCHER' AND C.EXPIRED_DATE BETWEEN to_date(V_DATE, 'yyyymmdd') AND add_months(to_date(V_DATE, 'yyyymmdd'), 6) THEN C.AGREE_REMAIN_AMOUNT WHEN C.DSCT_TYPE in ( 'FOC', 'Volume Based List Price Adjustment', 'One-Time Discount' ) AND C.DSCT_END_DATE BETWEEN to_date(V_DATE, 'yyyymmdd') AND add_months(to_date(V_DATE, 'yyyymmdd'), 6) THEN C.EFFECTIVE_TOTAL_AMOUNT ELSE NULL END ) AS IMMED_EXPIRED_AMOUNT --半年内即将过期金额 FROM DMSALESW.DM_SALE_BUSI_DSCT_SUM_F C WHERE C.CURRENCY_CODE IN ('USD') --改为多值 AND C.BG_CODE IN ('PDCG901159') AND C.M_ID IN ( 'BUSI_DSCT_00001', 'BUSI_DSCT_00002', 'BUSI_DSCT_00003', 'BUSI_DSCT_00004', 'BUSI_DSCT_00005', 'BUSI_DSCT_00006', 'BUSI_DSCT_00007' ) --新增加字段 --AND C.M_CN IN ('#[#P_REPORT_ITEM_NAME#]#') --新增加字段 --新增加字段 GROUP BY 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36 ) T --新增加入参 下面是执行计划： id | operation | A-time | A-rows | E-rows | E-distinct | Peak Memory | E-memory | A-width | E-width | E-costs ----+-------------------------------------------------------------------------+----------------------+---------+---------+------------+----------------+----------+-----------+---------+----------- 1 | -> Row Adapter | 1139.637 | 7 | 245 | | 419KB | | | 1318 | 117002.27 2 | -> Vector Streaming (type: GATHER) | 1139.616 | 7 | 245 | | 777KB | | | 1318 | 117002.27 3 | -> Vector Subquery Scan on t | [1129.463, 1130.072] | 7 | 245 | | [504KB, 504KB] | 1MB | | 1318 | 116920.22 4 | -> Vector Hash Aggregate | [1129.459, 1130.067] | 7 | 245 | | [4MB, 4MB] | 16MB | [0,898] | 523 | 116920.07 5 | -> Vector Streaming(type: REDISTRIBUTE) | [1129.142, 1129.918] | 112 | 3920 | | [1MB, 1MB] | 2MB | | 523 | 116643.28 6 | -> Vector Hash Aggregate | [882.194, 987.474] | 112 | 3920 | | [8MB, 8MB] | 16MB | [861,861] | 523 | 116498.95 7 | -> CStore Scan on dmsalesw.dm_sale_busi_dsct_sum_f c | [126.343, 142.697] | 3080954 | 2135243 | | [5MB, 5MB] | 1MB | | 203 | 66116.77 可以看到，不但省去了Nest Loop的耗时，而且后面Aggregate的耗时也减少了不少。整体从3s+优化到1.2s。 点击关注，第一时间了解华为云新鲜技术~
2023-11-30
优秀的个人博客,低调大师

数仓性能优化：倾斜优化-表达式计算倾斜的hint优化

本文分享自华为云社区《GaussDB(DWS)性能调优：倾斜优化-表达式计算倾斜的hint优化》，作者： 譡里个檔 。 1.原始SQL SELECT TMP4.TAX_AMT, CATE.L1_PUR_ITEM_CATG_CN_NAME || '-' || CATE.L2_PUR_ITEM_CATG_CN_NAME || '-' || CATE.L3_PUR_ITEM_CATG_CN_NAME AS PRODUCT_CATEGORY, MATE.ITEM_CODE AS PRODUCT_CODE, INVEN.INVENTORY_ORG_NAME, TMP4.INVOICE_WITHHOLDING_TAX_GROUP, TMP4.PAYMENT_WITHHOLDING_TAX_GROUP, TMP4.PO_CHARGE_ACCOUNT_CODE, TMP4.CFS_INVOICE_NUMBER, APR.TAX_INVOICE_DATE FROM DWLTAX.DWL_TAX_TAXDP_ERP_AP_INVOICE_TMP5 TMP4, DWRDIM_DW1.DWR_DIM_PUR_ITEM_CATEGORY_D CATE, DWRDIM_DW1.DWR_DIM_MATERIAL_CODE_D MATE, DWRDIM_DW1.DWR_DIM_INVENTORY_ORG_D INVEN, DWTAXDI.DWI_AP_INVOICE_I AP, DWTAXDI.DWI_AP_INVOICE_REGSTN_I APR WHERE 1 = 1 AND TMP4.ITEM_CATEGORY_KEY = CATE.PUR_ITEM_CATG_KEY(+) AND CATE.DEL_FLAG(+) = 'N' AND TMP4.ITEM_ID = MATE.ITEM_ID(+) AND MATE.DEL_FLAG(+) = 'N' AND TMP4.PO_SHIPMENT_TARGET_INV_ORG_KEY = INVEN.INVENTORY_ORG_KEY(+) AND INVEN.DEL_FLAG(+) = 'N' AND TMP4.AP_INVOICE_ID = AP.AP_INVOICE_ID(+) AND 6600 || AP.ATTRIBUTE1 = TO_CHAR(APR.AP_INVOICE_REGSTN_ID(+)) 执行performance，查询具体执行情况和SQL自诊断信息(详细见附件case-step1-原始执行信息.txt) id | operation | A-time | A-rows | E-rows | E-distinct | Peak Memory | E-memory | A-width | E-width | E-costs ----+------------------------------------------------------------------------------------------------------+------------------------+------------+------------+------------+----------------+----------------+-----------+---------+------------- 1 | -> Row Adapter | 69922.773 | 69237018 | 69237018 | | 87KB | | | 573 | 15160857.61 2 | -> Vector Streaming (type: GATHER) | 65581.989 | 69237018 | 69237018 | | 536KB | | | 573 | 15160857.61 3 | -> Vector Hash Right Join (4, 6) | [61186.201, 73129.055] | 69237018 | 69237018 | | [306MB, 682MB] | 1113MB(9990MB) | | 573 | 15159431.83 4 | -> Vector Streaming(type: BROADCAST ng: LC_DL1->LC_DW1) | [554.217, 21008.078] | 1382000544 | 1381572384 | 282184 | [4MB, 4MB] | 3MB | | 16 | 7056095.88 5 | -> CStore Scan on dwifin.dwi_ap_invoice_regstn s | [5.354, 11.617] | 28791678 | 28782758 | | [1MB, 1MB] | 1MB | | 16 | 28004.18 6 | -> Vector Hash Left Join (7, 19) | [1728.008, 2017.488] | 69237018 | 69237018 | 79721 | [834KB, 834KB] | 16MB | [229,252] | 578 | 1832322.90 7 | -> Vector Hash Left Join (8, 17) | [1428.799, 1925.653] | 69237018 | 69237018 | 179 | [32MB, 32MB] | 28MB(8901MB) | | 576 | 1817105.07 8 | -> Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) | [996.780, 1635.826] | 69237018 | 69237018 | 4167 | [1MB, 1MB] | 2MB | | 570 | 1788113.85 9 | -> Vector Hash Left Join (10, 14) | [1086.903, 1780.641] | 69237018 | 69237018 | | [173MB, 174MB] | 227MB(9067MB) | | 570 | 1304897.12 10 | -> Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) | [153.628, 891.680] | 69237018 | 69237018 | 20271 | [1MB, 1MB] | 2MB | | 567 | 847160.16 11 | -> Vector Hash Left Join (12, 13) | [367.155, 465.821] | 69237018 | 69237018 | | [30MB, 30MB] | 22MB(8896MB) | | 567 | 363943.43 12 | -> CStore Scan on dwltax.dwl_tax_taxdp_erp_ap_invoice_tmp5 tmp4 | [150.676, 178.827] | 69237018 | 69237018 | 526 | [4MB, 4MB] | 1MB | | 553 | 340168.44 13 | -> CStore Scan on dwrdim_dw1.dwr_dim_pur_item_category_d cate | [14.549, 24.399] | 8228448 | 8228448 | 171426 | [2MB, 2MB] | 1MB | [104,104] | 26 | 9056.99 14 | -> Vector Streaming(type: PART REDISTRIBUTE PART BROADCAST ng: LC_DL1->LC_DW1) | [315.926, 339.782] | 117191217 | 117191170 | 2441483 | [1MB, 1MB] | 3MB | [47,47] | 22 | 406136.10 15 | -> Vector Partition Iterator | [118.307, 151.248] | 117191170 | 117191170 | | [41KB, 41KB] | 1MB | | 22 | 300641.93 16 | -> Partitioned CStore Scan on dwifin.dwi_ap_invoice s | [86.557, 111.947] | 117191170 | 117191170 | | [6MB, 6MB] | 1MB | | 22 | 300641.93 17 | -> Vector Streaming(type: PART LOCAL PART BROADCAST) | [60.429, 99.381] | 15442613 | 15442566 | 321720 | [584KB, 584KB] | 2MB | [58,58] | 19 | 49578.19 18 | -> CStore Scan on dwrdim_dw1.dwr_dim_material_code_d mate | [19.779, 33.206] | 15442566 | 15442566 | | [1MB, 2MB] | 1MB | | 19 | 35704.02 19 | -> CStore Scan on dwrdim_dw1.dwr_dim_inventory_org_d inven | [0.383, 0.739] | 135072 | 135072 | 2814 | [1MB, 1MB] | 1MB | [53,53] | 14 | 2823.85 SQL Diagnostic Information -------------------------------------------------------------------------------------------- Execute diagnostic information PlanNode[4] Large Table in Broadcast "Vector Streaming(type: BROADCAST ng: LC_DL1->LC_DW1)" Predicate Information (identified by plan id) ------------------------------------------------------------------------------------------------------------------------------ 3 --Vector Hash Right Join (4, 6) Hash Cond: (((numeric_out(s.ap_invoice_regstn_id))::character varying)::text = ('6600'::text || (s.attribute1)::text)) 6 --Vector Hash Left Join (7, 19) Hash Cond: (tmp4.po_shipment_target_inv_org_key = inven.inventory_org_key) 7 --Vector Hash Left Join (8, 17) Hash Cond: (tmp4.item_id = mate.item_id) Skew Join Optimized by Statistic 8 --Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) Skew Filter(type: ROUNDROBIN): ((tmp4.item_id = (-999999)::numeric) OR (tmp4.item_id IS NULL)) 9 --Vector Hash Left Join (10, 14) Hash Cond: (tmp4.ap_invoice_id = s.ap_invoice_id) Skew Join Optimized by Statistic 10 --Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) Skew Filter(type: ROUNDROBIN): (tmp4.ap_invoice_id = 1001113812002::numeric) 11 --Vector Hash Left Join (12, 13) Hash Cond: (tmp4.item_category_key = cate.pur_item_catg_key) 13 --CStore Scan on dwrdim_dw1.dwr_dim_pur_item_category_d cate Filter: ((cate.del_flag)::text = 'N'::text) Pushdown Predicate Filter: ((cate.del_flag)::text = 'N'::text) 14 --Vector Streaming(type: PART REDISTRIBUTE PART BROADCAST ng: LC_DL1->LC_DW1) Skew Filter(type: BROADCAST): (s.ap_invoice_id = 1001113812002::numeric) 15 --Vector Partition Iterator Iterations: 147 16 --Partitioned CStore Scan on dwifin.dwi_ap_invoice s Partitions Selected by Static Prune: 1..147 17 --Vector Streaming(type: PART LOCAL PART BROADCAST) Skew Filter(type: BROADCAST): (mate.item_id = (-999999)::numeric) 18 --CStore Scan on dwrdim_dw1.dwr_dim_material_code_d mate Filter: ((mate.del_flag)::text = 'N'::text) Pushdown Predicate Filter: ((mate.del_flag)::text = 'N'::text) 19 --CStore Scan on dwrdim_dw1.dwr_dim_inventory_org_d inven Filter: ((inven.del_flag)::text = 'N'::text) Pushdown Predicate Filter: ((inven.del_flag)::text = 'N'::text) 2.禁止大表广播 如上小节显示确实是id=4的这一步是一个大的结果集（2879w条）做了broadcast，并且紧接着的id=5的HashJoin耗时很长。因此通过增加hint方式禁止dwifin.dwi_ap_invoice_regstn走广播。分析发现表dwifin.dwi_ap_invoice_regstn是视图apr展开出现的，因此增加如下hint信息，其中 1. no merge (apr)是防止视图apr中的语句提升，导致的hint信息失效 2. no broadcast(apr)表示禁止apr走broadcast EXPLAIN performance SELECT /*+ no merge (apr) no broadcast(apr) */ TMP4.TAX_AMT, CATE.L1_PUR_ITEM_CATG_CN_NAME || '-' || CATE.L2_PUR_ITEM_CATG_CN_NAME || '-' || CATE.L3_PUR_ITEM_CATG_CN_NAME AS PRODUCT_CATEGORY, MATE.ITEM_CODE AS PRODUCT_CODE, INVEN.INVENTORY_ORG_NAME, TMP4.INVOICE_WITHHOLDING_TAX_GROUP, TMP4.PAYMENT_WITHHOLDING_TAX_GROUP, TMP4.PO_CHARGE_ACCOUNT_CODE, TMP4.CFS_INVOICE_NUMBER, APR.TAX_INVOICE_DATE FROM DWLTAX.DWL_TAX_TAXDP_ERP_AP_INVOICE_TMP5 TMP4, DWRDIM_DW1.DWR_DIM_PUR_ITEM_CATEGORY_D CATE, DWRDIM_DW1.DWR_DIM_MATERIAL_CODE_D MATE, DWRDIM_DW1.DWR_DIM_INVENTORY_ORG_D INVEN, DWTAXDI.DWI_AP_INVOICE_I AP, DWTAXDI.DWI_AP_INVOICE_REGSTN_I APR WHERE 1 = 1 AND TMP4.ITEM_CATEGORY_KEY = CATE.PUR_ITEM_CATG_KEY(+) AND CATE.DEL_FLAG(+) = 'N' AND TMP4.ITEM_ID = MATE.ITEM_ID(+) AND MATE.DEL_FLAG(+) = 'N' AND TMP4.PO_SHIPMENT_TARGET_INV_ORG_KEY = INVEN.INVENTORY_ORG_KEY(+) AND INVEN.DEL_FLAG(+) = 'N' AND TMP4.AP_INVOICE_ID = AP.AP_INVOICE_ID(+) AND 6600 || AP.ATTRIBUTE1 = TO_CHAR(APR.AP_INVOICE_REGSTN_ID(+)) 获取如上语句的performance信息(详细见附件 case-step2-禁止大表广播.txt) id | operation | A-time | A-rows | E-rows | E-distinct | Peak Memory | E-memory | A-width | E-width | E-costs ----+---------------------------------------------------------------------------------------------------------+------------------------+-----------+-----------+------------+----------------+----------------+-----------+---------+------------- 1 | -> Row Adapter | 15685.781 | 69237018 | 69237018 | | 87KB | | | 573 | 33341721.22 2 | -> Vector Streaming (type: GATHER) | 11361.740 | 69237018 | 69237018 | | 536KB | | | 573 | 33341721.22 3 | -> Vector Hash Left Join (4, 19) | [15269.267, 18985.791] | 69237018 | 69237018 | | [74MB, 74MB] | 101MB(9984MB) | | 573 | 33340295.43 4 | -> Vector Streaming(type: REDISTRIBUTE) | [4743.867, 18632.182] | 69237018 | 69237018 | 79721 | [1MB, 2MB] | 2MB | | 578 | 29821930.76 5 | -> Vector Hash Left Join (6, 18) | [1473.990, 15359.055] | 69237018 | 69237018 | | [866KB, 898KB] | 16MB | | 578 | 1832322.90 6 | -> Vector Hash Left Join (7, 16) | [1130.814, 15223.646] | 69237018 | 69237018 | 179 | [32MB, 32MB] | 28MB(9923MB) | | 576 | 1817105.07 7 | -> Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) | [681.709, 14909.424] | 69237018 | 69237018 | 4167 | [1MB, 1MB] | 2MB | | 570 | 1788113.85 8 | -> Vector Hash Left Join (9, 13) | [1049.201, 12602.796] | 69237018 | 69237018 | | [173MB, 174MB] | 227MB(10089MB) | | 570 | 1304897.12 9 | -> Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) | [128.704, 11737.099] | 69237018 | 69237018 | 20271 | [1MB, 1MB] | 2MB | | 567 | 847160.16 10 | -> Vector Hash Left Join (11, 12) | [368.537, 443.623] | 69237018 | 69237018 | | [30MB, 30MB] | 22MB(9918MB) | | 567 | 363943.43 11 | -> CStore Scan on dwltax.dwl_tax_taxdp_erp_ap_invoice_tmp5 tmp4 | [148.366, 175.347] | 69237018 | 69237018 | 526 | [4MB, 4MB] | 1MB | | 553 | 340168.44 12 | -> CStore Scan on dwrdim_dw1.dwr_dim_pur_item_category_d cate | [13.319, 24.442] | 8228448 | 8228448 | 171426 | [2MB, 2MB] | 1MB | [104,104] | 26 | 9056.99 13 | -> Vector Streaming(type: PART REDISTRIBUTE PART BROADCAST ng: LC_DL1->LC_DW1) | [242.053, 294.233] | 117191217 | 117191170 | 2441483 | [1MB, 1MB] | 3MB | [47,47] | 22 | 406136.10 14 | -> Vector Partition Iterator | [118.124, 154.954] | 117191170 | 117191170 | | [41KB, 41KB] | 1MB | | 22 | 300641.93 15 | -> Partitioned CStore Scan on dwifin.dwi_ap_invoice s | [86.942, 105.441] | 117191170 | 117191170 | | [6MB, 6MB] | 1MB | | 22 | 300641.93 16 | -> Vector Streaming(type: PART LOCAL PART BROADCAST) | [83.793, 117.853] | 15442613 | 15442566 | 321720 | [584KB, 584KB] | 2MB | [58,58] | 19 | 49578.19 17 | -> CStore Scan on dwrdim_dw1.dwr_dim_material_code_d mate | [21.898, 35.895] | 15442566 | 15442566 | | [1MB, 2MB] | 1MB | | 19 | 35704.02 18 | -> CStore Scan on dwrdim_dw1.dwr_dim_inventory_org_d inven | [0.389, 0.661] | 135072 | 135072 | 2814 | [1MB, 1MB] | 1MB | [53,53] | 14 | 2823.85 19 | -> Vector Streaming(type: REDISTRIBUTE ng: LC_DL1->LC_DW1) | [30.667, 49.474] | 28791678 | 28782758 | 599641 | [2MB, 2MB] | 3MB | [75,75] | 16 | 56030.49 20 | -> Vector Subquery Scan on apr | [42.087, 61.734] | 28791678 | 28782758 | | [376KB, 376KB] | 1MB | | 16 | 30826.02 21 | -> CStore Scan on dwifin.dwi_ap_invoice_regstn s | [5.177, 8.049] | 28791678 | 28782758 | | [1MB, 1MB] | 1MB | | 16 | 28004.18 SQL Diagnostic Information ---------------------------------------------------------------------------------------------------------- Execute diagnostic information PlanNode[4] DataSkew:"Vector Streaming(type: REDISTRIBUTE)", min_dn_tuples:257082, max_dn_tuples:47206637 Predicate Information (identified by plan id) ---------------------------------------------------------------------------------------------------------------------------------- 3 --Vector Hash Left Join (4, 19) Hash Cond: ((('6600'::text || (s.attribute1)::text)) = ((numeric_out(apr.ap_invoice_regstn_id))::character varying)::text) 5 --Vector Hash Left Join (6, 18) Hash Cond: (tmp4.po_shipment_target_inv_org_key = inven.inventory_org_key) 6 --Vector Hash Left Join (7, 16) Hash Cond: (tmp4.item_id = mate.item_id) Skew Join Optimized by Statistic 7 --Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) Skew Filter(type: ROUNDROBIN): ((tmp4.item_id = (-999999)::numeric) OR (tmp4.item_id IS NULL)) 8 --Vector Hash Left Join (9, 13) Hash Cond: (tmp4.ap_invoice_id = s.ap_invoice_id) Skew Join Optimized by Statistic 9 --Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) Skew Filter(type: ROUNDROBIN): (tmp4.ap_invoice_id = 1001113812002::numeric) 10 --Vector Hash Left Join (11, 12) Hash Cond: (tmp4.item_category_key = cate.pur_item_catg_key) 12 --CStore Scan on dwrdim_dw1.dwr_dim_pur_item_category_d cate Filter: ((cate.del_flag)::text = 'N'::text) Pushdown Predicate Filter: ((cate.del_flag)::text = 'N'::text) 13 --Vector Streaming(type: PART REDISTRIBUTE PART BROADCAST ng: LC_DL1->LC_DW1) Skew Filter(type: BROADCAST): (s.ap_invoice_id = 1001113812002::numeric) 14 --Vector Partition Iterator Iterations: 147 15 --Partitioned CStore Scan on dwifin.dwi_ap_invoice s Partitions Selected by Static Prune: 1..147 16 --Vector Streaming(type: PART LOCAL PART BROADCAST) Skew Filter(type: BROADCAST): (mate.item_id = (-999999)::numeric) 17 --CStore Scan on dwrdim_dw1.dwr_dim_material_code_d mate Filter: ((mate.del_flag)::text = 'N'::text) Pushdown Predicate Filter: ((mate.del_flag)::text = 'N'::text) 18 --CStore Scan on dwrdim_dw1.dwr_dim_inventory_org_d inven Filter: ((inven.del_flag)::text = 'N'::text) Pushdown Predicate Filter: ((inven.del_flag)::text = 'N'::text) 3.表达式倾斜的hint 发现自诊断信息中倾斜告警 而Plan ID为4的算子是 其中是s是视图dwtaxdi.dwi_ap_invoice_i展开后的表dwifin.dwi_ap_invoice，查询此表的列attribute1的统计信息如下，发现在NULL值上存在严重倾斜 因为重分布列是一个表达式6600 || AP.ATTRIBUTE1，当前DWS的倾斜的hint不支持表达式，因为我们做如下变通实现表达式的值倾斜的hint SELECT /*+ no merge (apr) no broadcast(apr) no merge(ap) skew(ap (attr1) ('6600')) */ TMP4.TAX_AMT, CATE.L1_PUR_ITEM_CATG_CN_NAME || '-' || CATE.L2_PUR_ITEM_CATG_CN_NAME || '-' || CATE.L3_PUR_ITEM_CATG_CN_NAME AS PRODUCT_CATEGORY, MATE.ITEM_CODE AS PRODUCT_CODE, INVEN.INVENTORY_ORG_NAME, TMP4.INVOICE_WITHHOLDING_TAX_GROUP, TMP4.PAYMENT_WITHHOLDING_TAX_GROUP, TMP4.PO_CHARGE_ACCOUNT_CODE, TMP4.CFS_INVOICE_NUMBER, APR.TAX_INVOICE_DATE FROM DWLTAX.DWL_TAX_TAXDP_ERP_AP_INVOICE_TMP5 TMP4, DWRDIM_DW1.DWR_DIM_PUR_ITEM_CATEGORY_D CATE, DWRDIM_DW1.DWR_DIM_MATERIAL_CODE_D MATE, DWRDIM_DW1.DWR_DIM_INVENTORY_ORG_D INVEN, (SELECT *, 6600 || AP.ATTRIBUTE1 AS ATTR1 FROM DWTAXDI.DWI_AP_INVOICE_I AP) AP, DWTAXDI.DWI_AP_INVOICE_REGSTN_I APR WHERE 1 = 1 AND TMP4.ITEM_CATEGORY_KEY = CATE.PUR_ITEM_CATG_KEY(+) AND CATE.DEL_FLAG(+) = 'N' AND TMP4.ITEM_ID = MATE.ITEM_ID(+) AND MATE.DEL_FLAG(+) = 'N' AND TMP4.PO_SHIPMENT_TARGET_INV_ORG_KEY = INVEN.INVENTORY_ORG_KEY(+) AND INVEN.DEL_FLAG(+) = 'N' AND TMP4.AP_INVOICE_ID = AP.AP_INVOICE_ID(+) AND ATTR1 = TO_CHAR(APR.AP_INVOICE_REGSTN_ID(+)) 其中构建了子查询 AP SELECT *, 6600 || AP.ATTRIBUTE1 AS ATTR1 FROM DWTAXDI.DWI_AP_INVOICE_I AP 在把原始的关联列表达式放到子查询里面，然后把 6600 || AP.ATTRIBUTE1 命名为attr1。 在父查询中首先禁止AP这个子查询提升。然后在父查询中通过hint 子查询AP这个结果集的列attr1存在倾斜值'6600' 。这个倾斜值是计算出来的（NULL || 6600 = ‘6600’），并且在原始关联计算中关联表达式是如下，即 6600 || AP.ATTRIBUTE1的结果被转换为text类型（字符串类型） 获取新的语句的performance如下(详细见附件 case-step3-倾斜优化.txt) id | operation | A-time | A-rows | E-rows | E-distinct | Peak Memory | E-memory | A-width | E-width | E-costs ----+------------------------------------------------------------------------------------------------------+-----------------------+-----------+-----------+------------+----------------+----------------+-----------+---------+------------ 1 | -> Row Adapter | 9045.793 | 69237018 | 69237018 | | 87KB | | | 573 | 2040755.71 2 | -> Vector Streaming (type: GATHER) | 4842.656 | 69237018 | 69237018 | | 520KB | | | 573 | 2040755.71 3 | -> Vector Hash Left Join (4, 21) | [2673.707, 11389.688] | 69237018 | 69237018 | | [1MB, 1MB] | 16MB | | 573 | 2039329.92 4 | -> Vector Hash Left Join (5, 19) | [1951.482, 10931.220] | 69237018 | 69237018 | 179 | [32MB, 32MB] | 28MB(10018MB) | | 571 | 2009687.71 5 | -> Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) | [1541.777, 10591.702] | 69237018 | 69237018 | 4167 | [1MB, 1MB] | 2MB | | 565 | 1980696.49 6 | -> Vector Hash Left Join (7, 18) | [1703.438, 1980.655] | 69237018 | 69237018 | | [30MB, 30MB] | 22MB(10010MB) | | 565 | 1497479.76 7 | -> Vector Hash Left Join (8, 10) | [1523.277, 1708.622] | 69237018 | 69237018 | 526 | [165MB, 166MB] | 191MB(10151MB) | | 551 | 1473704.77 8 | -> Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) | [94.501, 203.619] | 69237018 | 69237018 | 20271 | [1MB, 1MB] | 2MB | | 553 | 823385.17 9 | -> CStore Scan on dwltax.dwl_tax_taxdp_erp_ap_invoice_tmp5 tmp4 | [142.734, 171.486] | 69237018 | 69237018 | | [4MB, 4MB] | 1MB | | 553 | 340168.44 10 | -> Vector Streaming(type: PART REDISTRIBUTE PART BROADCAST ng: LC_DL1->LC_DW1) | [811.192, 853.583] | 117191217 | 117191170 | 2441483 | [2MB, 2MB] | 3MB | [44,44] | 17 | 598718.74 11 | -> Vector Hash Left Join (12, 15) | [340.998, 790.399] | 117191170 | 117191170 | | [39MB, 39MB] | 27MB(10015MB) | | 17 | 493224.57 12 | -> Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) | [53.170, 79.836] | 117191170 | 117191170 | 79721 | [2MB, 2MB] | 3MB | | 41 | 412662.90 13 | -> Vector Partition Iterator | [145.450, 171.527] | 117191170 | 117191170 | | [41KB, 41KB] | 1MB | | 22 | 303514.27 14 | -> Partitioned CStore Scan on dwifin.dwi_ap_invoice s | [112.099, 134.193] | 117191170 | 117191170 | | [6MB, 6MB] | 1MB | | 22 | 300641.93 15 | -> Vector Streaming(type: PART REDISTRIBUTE PART BROADCAST) | [48.632, 99.230] | 28791678 | 28782758 | 282184 | [2MB, 2MB] | 3MB | [75,75] | 16 | 56928.04 16 | -> Vector Subquery Scan on apr | [41.916, 78.189] | 28791678 | 28782758 | | [376KB, 376KB] | 1MB | | 16 | 30826.02 17 | -> CStore Scan on dwifin.dwi_ap_invoice_regstn s | [5.233, 10.667] | 28791678 | 28782758 | | [1MB, 1MB] | 1MB | | 16 | 28004.18 18 | -> CStore Scan on dwrdim_dw1.dwr_dim_pur_item_category_d cate | [12.065, 20.667] | 8228448 | 8228448 | 171426 | [2MB, 2MB] | 1MB | [104,104] | 26 | 9056.99 19 | -> Vector Streaming(type: PART LOCAL PART BROADCAST) | [67.272, 97.378] | 15442613 | 15442566 | 321720 | [584KB, 584KB] | 2MB | [58,58] | 19 | 49578.19 20 | -> CStore Scan on dwrdim_dw1.dwr_dim_material_code_d mate | [18.605, 31.713] | 15442566 | 15442566 | | [1MB, 2MB] | 1MB | | 19 | 35704.02 21 | -> CStore Scan on dwrdim_dw1.dwr_dim_inventory_org_d inven | [0.378, 0.647] | 135072 | 135072 | 2814 | [1MB, 1MB] | 1MB | [53,53] | 14 | 2823.85 Predicate Information (identified by plan id) ---------------------------------------------------------------------------------------------------------------------------------- 3 --Vector Hash Left Join (4, 21) Hash Cond: (tmp4.po_shipment_target_inv_org_key = inven.inventory_org_key) 4 --Vector Hash Left Join (5, 19) Hash Cond: (tmp4.item_id = mate.item_id) Skew Join Optimized by Statistic 5 --Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) Skew Filter(type: ROUNDROBIN): ((tmp4.item_id = (-999999)::numeric) OR (tmp4.item_id IS NULL)) 6 --Vector Hash Left Join (7, 18) Hash Cond: (tmp4.item_category_key = cate.pur_item_catg_key) 7 --Vector Hash Left Join (8, 10) Hash Cond: (tmp4.ap_invoice_id = s.ap_invoice_id) Skew Join Optimized by Statistic 8 --Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) Skew Filter(type: ROUNDROBIN): (tmp4.ap_invoice_id = 1001113812002::numeric) 10 --Vector Streaming(type: PART REDISTRIBUTE PART BROADCAST ng: LC_DL1->LC_DW1) Skew Filter(type: BROADCAST): (s.ap_invoice_id = 1001113812002::numeric) 11 --Vector Hash Left Join (12, 15) Hash Cond: ((('6600'::text || (s.attribute1)::text)) = ((numeric_out(apr.ap_invoice_regstn_id))::character varying)::text) Skew Join Optimized by Hint 12 --Vector Streaming(type: PART REDISTRIBUTE PART ROUNDROBIN) Skew Filter(type: ROUNDROBIN): ((('6600'::text || (s.attribute1)::text)) = '6600'::text) 13 --Vector Partition Iterator Iterations: 147 14 --Partitioned CStore Scan on dwifin.dwi_ap_invoice s Partitions Selected by Static Prune: 1..147 15 --Vector Streaming(type: PART REDISTRIBUTE PART BROADCAST) Skew Filter(type: BROADCAST): ((((numeric_out(apr.ap_invoice_regstn_id))::character varying)::text) = '6600'::text) 18 --CStore Scan on dwrdim_dw1.dwr_dim_pur_item_category_d cate Filter: ((cate.del_flag)::text = 'N'::text) Pushdown Predicate Filter: ((cate.del_flag)::text = 'N'::text) 19 --Vector Streaming(type: PART LOCAL PART BROADCAST) Skew Filter(type: BROADCAST): (mate.item_id = (-999999)::numeric) 20 --CStore Scan on dwrdim_dw1.dwr_dim_material_code_d mate Filter: ((mate.del_flag)::text = 'N'::text) Pushdown Predicate Filter: ((mate.del_flag)::text = 'N'::text) 21 --CStore Scan on dwrdim_dw1.dwr_dim_inventory_org_d inven Filter: ((inven.del_flag)::text = 'N'::text) Pushdown Predicate Filter: ((inven.del_flag)::text = 'N'::text) 附件：case-step1-原始执行信息.txt0B 附件：case-step3-倾斜优化.txt862.61KB 附件：case-step2-禁止大表广播.txt0B 点击关注，第一时间了解华为云新鲜技术~
2023-08-01

资源下载

更多资源
Mario

Mario

马里奥是站在游戏界顶峰的超人气多面角色。马里奥靠吃蘑菇成长，特征是大鼻子、头戴帽子、身穿背带裤，还留着胡子。与他的双胞胎兄弟路易基一起，长年担任任天堂的招牌角色。

时间: 2026-04-11 大小: 211.28KB 下载: 56次
腾讯云软件源

腾讯云软件源

为解决软件依赖安装时官方源访问速度慢的问题，腾讯云为一些软件搭建了缓存服务。您可以通过使用腾讯云软件源站来提升依赖包的安装速度。为了方便用户自由搭建服务架构，目前腾讯云软件源站支持公网访问和内网访问。

时间: 2026-08-21 大小: 1MB 下载: 10次
Sublime Text

Sublime Text

Sublime Text具有漂亮的用户界面和强大的功能，例如代码缩略图，Python的插件，代码段等。还可自定义键绑定，菜单和工具栏。Sublime Text 的主要功能包括：拼写检查，书签，完整的 Python API ， Goto 功能，即时项目切换，多选择，多窗口等等。Sublime Text 是一个跨平台的编辑器，同时支持Windows、Linux、Mac OS X等操作系统。

时间: 2026-08-21 大小: 9.84MB 下载: 40次
WebStorm

WebStorm

WebStorm 是jetbrains公司旗下一款JavaScript 开发工具。目前已经被广大中国JS开发者誉为“Web前端开发神器”、“最强大的HTML5编辑器”、“最智能的JavaScript IDE”等。与IntelliJ IDEA同源，继承了IntelliJ IDEA强大的JS部分的功能。

时间: 2025-12-13 大小: 249.54MB 下载: 22次
热门标签
DeepSeek Jdk25 OpenAi HarmonyOS6 Nacos3 SpringBoot4 SpringCloud Docker Kubernetes Service Mesh Redis Gemini3 Rocky

欢迎您来访！

登录后，您将获得更多功能！
热门文章
联系我们

有任何问题或合作意向欢迎联系我们

Email: 99873273@qq.com

QQ: 99873273

用户登录
用户注册