今天mentor給了一個sql語句優化的任務。（環境是sql developer）有一個語句執行很慢，查詢出來的結果有17544條記錄，但需970秒，速度很慢。語句是這樣的：
SELECT DISTINCT 'AMEND_NEW',
reporttitle,
reportsubtitle,
cab_cab_transactions.branchcode,
cab_cab_transactions.prtfo_cd,
cab_cab_transactions.sstm_scrty_id,
cab_cab_transactions.sstm_trx_id,
cab_cab_transactions.trde_dttm,
cab_cab_transactions.efcte_dttm,
cab_cab_transactions.due_stlmnt_dt,
cab_cab_transactions.cncl_efcte_dttm,
cab_cab_transactions.trde_sstm_id,
cab_cab_transactions.trx_type_cd,
cab_cab_transactions.trx_type_dscrn,
cab_cab_transactions.trx_subtype_cd,
cab_cab_transactions.trde_stat_flg,
cab_cab_transactions.csh_cr_dr_indcr,
cab_cab_transactions.long_shrt_indcr,
cab_cab_transactions.lcl_crncy,
cab_cab_transactions.stlmt_crncy,
cab_cab_transactions.nomin_qty,
cab_cab_transactions.price,
cab_cab_transactions.lcl_cst,
cab_cab_transactions.prtfo_cst,
cab_cab_transactions.lcl_book_cst,
cab_cab_transactions.prtfo_book_cst,
cab_cab_transactions.lcl_sell_prcds,
cab_cab_transactions.prtfo_sell_prcds,
cab_cab_transactions.lcl_gnls,
cab_cab_transactions.prtfo_gnls,
cab_cab_transactions.lcl_acrd_intrt,
cab_cab_transactions.prtfo_acrd_intrt,
cab_cab_transactions.stlmt_crncy_stlmt_amt,
cab_cab_transactions.lcl_net_amt,
cab_cab_transactions.prtfo_net_amt,
cab_cab_transactions.fx_bght_amt,
cab_cab_transactions.fx_sold_amt,
cab_cab_transactions.prtfo_crncy_stlmt_amt,
cab_cab_transactions.prtfo_net_incme,
cab_cab_transactions.dvnd_crncy_net_incme,
cab_cab_transactions.dvnd_type_cd,
cab_cab_transactions.lcl_intrt_pd_rec,
cab_cab_transactions.prtfo_intrt_pd_rec,
cab_cab_transactions.lcl_dvdnd_pd_rec,
cab_cab_transactions.prtfo_dvdnd_pd_rec,
cab_cab_transactions.lcl_sundry_inc_pd_rec,
cab_cab_transactions.prtfo_sundry_inc_pd_rec,
cab_cab_transactions.bnk_csh_cptl_secid,
cab_cab_transactions.bnk_csh_inc_secid,
cab_cab_transactions.reportdate,
cab_cab_transactions.filename,
sysdate,
'e483448'
FROM cab_cfg_trx_type_mapping RIGHT JOIN(cab_cab_tran_adjustments
INNER JOIN cab_cab_transactions ON(cab_cab_transactions.branchcode = cab_cab_tran_adjustments.branchcode )
AND(cab_cab_tran_adjustments.sstm_trx_id = cab_cab_transactions.sstm_trx_id)) ON(cab_cfg_trx_type_mapping.cab_trx_type_cd = cab_cab_transactions.trx_type_cd)
AND(nvl(cab_cfg_trx_type_mapping.cab_trx_subtype_cd,' ') = nvl(cab_cab_transactions.trx_subtype_cd,' ')
AND (cab_cfg_trx_type_mapping.branchcode=cab_cab_transactions.branchcode))
WHERE cab_cab_transactions.prtfo_cd IN
(SELECT DISTINCT prtfo_cd
FROM cab_cab_valuations_working
WHERE created_by = 'e483448'
AND branchcode='ISA')
AND cab_cab_tran_adjustments.efcte_dttm > '2011-07-31'
AND cab_cab_tran_adjustments.efcte_dttm <= '2011-08-31'
AND eff_trde_stat_flg <> 'X'
AND cab_cab_transactions.branchcode = 'ISA'
AND cab_cab_tran_adjustments.branchcode = 'ISA'
AND(cab_cfg_trx_type_mapping.cab_reportgroup = 'CABValuation' OR cab_cfg_trx_type_mapping.cab_reportgroup IS NULL)
問題在distinct上面，它會導致對全表掃描，而且會導致排序，然後刪除重復的記錄，所以速度很慢，因此需要優化distinct。查了不少資料，並逐一嘗試，最後發現了一個非常可觀的優化結果，用group by。語句如下：
SELECT 'AMEND_NEW',
reporttitle,
reportsubtitle,
cab_cab_transactions.branchcode,
cab_cab_transactions.prtfo_cd,
cab_cab_transactions.sstm_scrty_id,
cab_cab_transactions.sstm_trx_id,
cab_cab_transactions.trde_dttm,
cab_cab_transactions.efcte_dttm,
cab_cab_transactions.due_stlmnt_dt,
cab_cab_transactions.cncl_efcte_dttm,
cab_cab_transactions.trde_sstm_id,
cab_cab_transactions.trx_type_cd,
cab_cab_transactions.trx_type_dscrn,
cab_cab_transactions.trx_subtype_cd,
cab_cab_transactions.trde_stat_flg,
cab_cab_transactions.csh_cr_dr_indcr,
cab_cab_transactions.long_shrt_indcr,
cab_cab_transactions.lcl_crncy,
cab_cab_transactions.stlmt_crncy,
cab_cab_transactions.nomin_qty,
cab_cab_transactions.price,
cab_cab_transactions.lcl_cst,
cab_cab_transactions.prtfo_cst,
cab_cab_transactions.lcl_book_cst,
cab_cab_transactions.prtfo_book_cst,
cab_cab_transactions.lcl_sell_prcds,
cab_cab_transactions.prtfo_sell_prcds,
cab_cab_transactions.lcl_gnls,
cab_cab_transactions.prtfo_gnls,
cab_cab_transactions.lcl_acrd_intrt,
cab_cab_transactions.prtfo_acrd_intrt,
cab_cab_transactions.stlmt_crncy_stlmt_amt,
cab_cab_transactions.lcl_net_amt,
cab_cab_transactions.prtfo_net_amt,
cab_cab_transactions.fx_bght_amt,
cab_cab_transactions.fx_sold_amt,
cab_cab_transactions.prtfo_crncy_stlmt_amt,
cab_cab_transactions.prtfo_net_incme,
cab_cab_transactions.dvnd_crncy_net_incme,
cab_cab_transactions.dvnd_type_cd,
cab_cab_transactions.lcl_intrt_pd_rec,
cab_cab_transactions.prtfo_intrt_pd_rec,
cab_cab_transactions.lcl_dvdnd_pd_rec,
cab_cab_transactions.prtfo_dvdnd_pd_rec,
cab_cab_transactions.lcl_sundry_inc_pd_rec,
cab_cab_transactions.prtfo_sundry_inc_pd_rec,
cab_cab_transactions.bnk_csh_cptl_secid,
cab_cab_transactions.bnk_csh_inc_secid,
cab_cab_transactions.reportdate,
cab_cab_transactions.filename,
sysdate,
'e483448'
FROM cab_cfg_trx_type_mapping RIGHT JOIN(cab_cab_tran_adjustments
INNER JOIN cab_cab_transactions ON(cab_cab_transactions.branchcode = cab_cab_tran_adjustments.branchcode )
AND(cab_cab_tran_adjustments.sstm_trx_id = cab_cab_transactions.sstm_trx_id)) ON(cab_cfg_trx_type_mapping.cab_trx_type_cd = cab_cab_transactions.trx_type_cd)
AND(nvl(cab_cfg_trx_type_mapping.cab_trx_subtype_cd,' ') = nvl(cab_cab_transactions.trx_subtype_cd,' ')
AND (cab_cfg_trx_type_mapping.branchcode=cab_cab_transactions.branchcode))
WHERE cab_cab_transactions.prtfo_cd IN
(SELECT DISTINCT prtfo_cd
FROM cab_cab_valuations_working
WHERE created_by = 'e483448'
AND branchcode='ISA')
AND cab_cab_tran_adjustments.efcte_dttm > '2011-07-31'
AND cab_cab_tran_adjustments.efcte_dttm <= '2011-08-31'
AND eff_trde_stat_flg <> 'X'
AND cab_cab_transactions.branchcode = 'ISA'
AND cab_cab_tran_adjustments.branchcode = 'ISA'
AND(cab_cfg_trx_type_mapping.cab_reportgroup = 'CABValuation' OR cab_cfg_trx_type_mapping.cab_reportgroup IS NULL)
GROUP BY reporttitle,
reportsubtitle,
cab_cab_transactions.branchcode,
cab_cab_transactions.prtfo_cd,
cab_cab_transactions.sstm_scrty_id,
cab_cab_transactions.sstm_trx_id,
cab_cab_transactions.trde_dttm,
cab_cab_transactions.efcte_dttm,
cab_cab_transactions.due_stlmnt_dt,
cab_cab_transactions.cncl_efcte_dttm,
cab_cab_transactions.trde_sstm_id,
cab_cab_transactions.trx_type_cd,
cab_cab_transactions.trx_type_dscrn,
cab_cab_transactions.trx_subtype_cd,
cab_cab_transactions.trde_stat_flg,
cab_cab_transactions.csh_cr_dr_indcr,
cab_cab_transactions.long_shrt_indcr,
cab_cab_transactions.lcl_crncy,
cab_cab_transactions.stlmt_crncy,
cab_cab_transactions.nomin_qty,
cab_cab_transactions.price,
cab_cab_transactions.lcl_cst,
cab_cab_transactions.prtfo_cst,
cab_cab_transactions.lcl_book_cst,
cab_cab_transactions.prtfo_book_cst,
cab_cab_transactions.lcl_sell_prcds,
cab_cab_transactions.prtfo_sell_prcds,
cab_cab_transactions.lcl_gnls,
cab_cab_transactions.prtfo_gnls,
cab_cab_transactions.lcl_acrd_intrt,
cab_cab_transactions.prtfo_acrd_intrt,
cab_cab_transactions.stlmt_crncy_stlmt_amt,
cab_cab_transactions.lcl_net_amt,
cab_cab_transactions.prtfo_net_amt,
cab_cab_transactions.fx_bght_amt,
cab_cab_transactions.fx_sold_amt,
cab_cab_transactions.prtfo_crncy_stlmt_amt,
cab_cab_transactions.prtfo_net_incme,
cab_cab_transactions.dvnd_crncy_net_incme,
cab_cab_transactions.dvnd_type_cd,
cab_cab_transactions.lcl_intrt_pd_rec,
cab_cab_transactions.prtfo_intrt_pd_rec,
cab_cab_transactions.lcl_dvdnd_pd_rec,
cab_cab_transactions.prtfo_dvdnd_pd_rec,
cab_cab_transactions.lcl_sundry_inc_pd_rec,
cab_cab_transactions.prtfo_sundry_inc_pd_rec,
cab_cab_transactions.bnk_csh_cptl_secid,
cab_cab_transactions.bnk_csh_inc_secid,
cab_cab_transactions.reportdate,
cab_cab_transactions.filename
最後執行時間只有15.1秒，快了60多倍，不得不說這優化效果還是很可觀的。不過查了很多資料，仍然沒有發現合理地解釋：為什麼distinct 和group by的效率會有這麼大差別。查的很多資料，講的基本都是兩者相差不大，實現也差不多。有待解決。
DISTINCT和GROUP BY這兩者本質上應該沒有可比性，distinct 取出唯一列，group by 是分組，但有時候在優化的時候，在沒有聚合函數的時候，他們查出來的結果也一樣。
