Question

我有这个查询，它连接了几个表，然后是很多字段的groupbys。

SELECT a.SVC_ORD_NBR
        ,a.PO_NBR
        ,f.CUST_NBR as SHIPTO_NBR
        ,f.CUST_NAME
        ,f.CUST_ID
        ,f.CUST_LVL_2_NBR as Fleet_num
        ,f.CUST_LVL_2_NAME as Fleet
        ,f.CITY_NAME as SHIPTO_CITY
        ,f.STATE_CD as SHIPTO_STATE
        ,a.CREATE_DTM
        ,e.START_DTM as ACTVTY_START
        ,e.END_DTM as ACTVTY_END
        ,e.OPER_NBR
        ,o.DEBIT_MEMO_DTM
        ,o.CLOSE_DTM
        ,c.EQUIP_NBR as SAP_EQUIP_NBR
        ,c.TECH_NBR as Door_ID
        ,c.EQUIP_TYP_CD
        ,c.EQUIP_TYP_DESC
        ,c.BEGIN_WRNTY_DTE
        ,c.END_WRNTY_DTE
        ,b.REPLCMT_DTE
        ,h.PRIORITY_CD
        ,h.PRIORITY_DESC
        ,b.REPLCMT_CYCLE_CNT
        ,b.PREV_REPLCMT_DTE
        ,b.PREV_REPLCMT_CYCLE_CNT
        ,d.PROD_CD
        ,d.PRIM_PROD_DESC
        ,d.CUST_PROD_CTGY_CD
        ,b.ACTVTY_TYP_CD
        ,b.ACTVTY_TYP_DESC
        ,b.QTY
        ,b.UOM
        ,b.GRS_AMT_DCRNCY
        ,b.DISC_AMT_DCRNCY
        ,b.NET_AMT_DCRNCY
        ,a.STD_LABR_NET_AMT_DCRNCY
        ,a.OVTM_LABR_NET_AMT_DCRNCY
        ,a.TRAVEL_NET_AMT_DCRNCY
        ,a.MATL_NET_AMT_DCRNCY
        ,a.TAX_NET_AMT_DCRNCY
        ,a.FRT_NET_AMT_DCRNCY
        ,h.PM_ACTVTY_TYP_CD
        ,h.PM_ACTVTY_TYP_DESC
        ,b.CAUSE_CD
        ,b.CAUSE_DESC
        ,a.BRNCH_ID
        ,br.BRNCH_CD
        ,br.BRNCH_NAME
        ,SAP_SLS_ORG_NAME   
  FROM DMT_MAS_SVC_ORD_FACT a 
  INNER JOIN DMT_MAS_SVC_ACTVTY_FACT b on a.SVC_ORD_ID=b.SVC_ORD_ID
  INNER JOIN DMT_MAS_SVC_ACTVTY_DEGEN_DIM e on e.SVC_ACTVTY_ID=b.SVC_ACTVTY_ID
  INNER JOIN DMT_MAS_EQUIP_STORE_DIM c on b.EQUIP_STORE_ID=c.EQUIP_STORE_ID 
  INNER JOIN DMT_MAS_PROD_DIM d on b.PROD_ID=d.PROD_ID 
  INNER JOIN DMT_MAS_CUST_DIM f on a.SHIPTO_CUST_ID=f.CUST_ID
  INNER JOIN DMT_MAS_SVC_ORD_MISC_DIM h on a.SVC_ORD_MISC_DIM_ID = h.SVC_ORD_MISC_DIM_ID
  INNER JOIN DMT_MAS_SVC_ORD_DEGEN_DIM o on o.SVC_ORD_ID = a.SVC_ORD_ID
  INNER JOIN DMT_MAS_BRNCH_DIM br on br.BRNCH_ID = a.BRNCH_ID
WHERE CUST_LVL_2_NBR IN ('0080000268')
    AND e.START_DTM>='2016-01-01'
    AND f.SRC_SYS_DEL_FLAG != 'Y'
    AND CUST_NAME NOT LIKE '%USE%'
    AND CUST_NAME NOT LIKE '%DNU%'
    AND CUST_NAME NOT LIKE '%CLOSED%'
    AND CUST_NAME NOT LIKE '%use%'
    AND CUST_NAME NOT LIKE '%RELOCATED%'
    AND CUST_NAME NOT LIKE '%OLD%'
    AND CREATE_DTM >= NOW() - INTERVAL '1.5 years'
GROUP BY a.SVC_ORD_NBR
        ,a.PO_NBR
        ,f.CUST_NBR
        ,f.CUST_NAME
        ,f.CUST_ID
        ,f.CUST_LVL_2_NBR
        ,f.CUST_LVL_2_NAME
        ,f.CITY_NAME
        ,f.STATE_CD
        ,e.START_DTM
        ,e.END_DTM
        ,e.OPER_NBR
        ,c.EQUIP_NBR
        ,c.TECH_NBR
        ,c.EQUIP_TYP_CD
        ,c.EQUIP_TYP_DESC
        ,c.BEGIN_WRNTY_DTE
        ,c.END_WRNTY_DTE
        ,b.REPLCMT_DTE
        ,h.PRIORITY_CD
        ,h.PRIORITY_DESC
        ,b.REPLCMT_CYCLE_CNT
        ,b.PREV_REPLCMT_DTE
        ,b.PREV_REPLCMT_CYCLE_CNT
        ,d.PROD_CD
        ,d.PRIM_PROD_DESC
        ,d.CUST_PROD_CTGY_CD
        ,b.ACTVTY_TYP_CD
        ,b.ACTVTY_TYP_DESC
        ,b.QTY
        ,b.UOM
        ,b.GRS_AMT_DCRNCY
        ,b.DISC_AMT_DCRNCY
        ,b.NET_AMT_DCRNCY
        ,a.STD_LABR_NET_AMT_DCRNCY
        ,a.OVTM_LABR_NET_AMT_DCRNCY
        ,a.TRAVEL_NET_AMT_DCRNCY
        ,a.MATL_NET_AMT_DCRNCY
        ,a.TAX_NET_AMT_DCRNCY
        ,a.FRT_NET_AMT_DCRNCY   
        ,h.PM_ACTVTY_TYP_CD
        ,h.PM_ACTVTY_TYP_DESC
        ,b.CAUSE_CD
        ,b.CAUSE_DESC
        ,a.CREATE_DTM
        ,a.BRNCH_ID
        ,br.BRNCH_CD
        ,br.BRNCH_NAME
        ,SAP_SLS_ORG_NAME
        ,o.DEBIT_MEMO_DTM
        ,o.CLOSE_DTM

完成返回33080行的查询大约需要20分钟。我在查询上运行了explain analyze，结果发现大部分时间都在seq scan上进行。第二个耗时的任务是hash join。

确切地说，它是表DMT_MAS_SVC_ACTVTY_FACT，其中有6米行，执行seq扫描。

Seq Scan on dmt_mas_svc_actvty_fact b (cost=0.00..397,938.32 rows=6,851,932 width=730) (actual time=0.029..2,427.990 rows=6,851,932 loops=1)

就hash join而言，它与已经编入索引的表svc_ord_id的{{1}}字段有关。所以我想那里没什么可做的。

dmt_mas_svc_actvty_fact

我不明白为这么大的桌子做了昂贵的Hash Join (cost=189,428.58..613,158.11 rows=9,647 width=2,916) (actual time=1,021.686..5,234.921 rows=36,276 loops=1) Hash Cond: (b.svc_ord_id = a.svc_ord_id)。我该怎么做才能提高性能？ seq scan的完整输出低于

explain (analyze, verbose, buffers)

为什么Postgres选择在大桌面上执行seq扫描？

0 个答案: