索引覆蓋分析

時間 2019-12-17

標籤索引覆蓋分析简体版

原文原文鏈接

系統優化期間，須要找到優化後最能帶來效益的點進行優化，這樣才能從根本上提高系統運行效率。sql

如下內容以某個保單增量處理程序邏輯進行分析。當前增量保單事實提取過程分爲如下兩步：
a. 經過upd_time增量查詢出各個關聯表的增量cntr_id主鍵並插入tmp_f_cntr_v8_cntrrange臨時表;
b. 經過tmp_f_cntr_v8_cntrrange表與std_contract等表JOIN操做，獲取增量保單數據。

1. 增量範圍肯定
1.1 業務行爲共用增量控制表：
當前現狀描述：互不交叉的業務行爲（例如新契約個單、合同團單、新契約團單等）共用同一張tmp_f_cntr_v8_cntrrange增量控制表，以下：

問題：
不一樣業務行爲共用tmp_f_cntr_v8_cntrrange增量表，同時都採用tmp_f_cntr_v8_cntrrange表提取增量數據，無故成倍地增長了增量數據範圍。

解決方案：
應該爲各業務行爲建立單獨的增量表（例如tmp_f_cntr_v8_cntrrange_appl_psn、tmp_f_cntr_v8_cntrrange_appl_grp、tmp_f_cntr_v8_cntrrange_cntr_grp等），並經過這些增量表提取保單數據。

1.2 upd_time肯定增量保單範圍：
當前現狀描述：
代碼以下：app

select cntr_idas CNTR_NO,'999'as POL_CODE,'2'as APP_STD_FLAG from std_contract where upd_time>=?and upd_time<?
union
select distinctcntr_id as CNTR_NO,'999'as POL_CODE,'2'as APP_STD_FLAG from cntr_basic_state whereupd_time>=?and upd_time<?
union
select distinctcntr_id as CNTR_NO,'999'as POL_CODE,'2'as APP_STD_FLAG from cntr_sub_state where upd_time>=?and upd_time<?
union
select distinctcntr_id as CNTR_NO,'999'as POL_CODE,'2'as APP_STD_FLAG from insured whereupd_time>=? and upd_time<?
union
select distinctcntr_id as CNTR_NO,'999'as POL_CODE,'2'as APP_STD_FLAG from psn_cntr_holder whereupd_time>=?and upd_time<?
union
select distinctcntr_id as CNTR_NO,'999'as POL_CODE,'2'as APP_STD_FLAG from if_of_alt whereupd_time>=? and upd_time<?
union
select distinctstd_contract.cntr_idas CNTR_NO,'999'as POL_CODE,'2'as APP_STD_FLAG from grant_tranlst,std_contractwhere std_contract.cntr_no=grant_tranlst.cntr_noand grant_tranlst.upd_time>=?and grant_tranlst.upd_time<?
union
select distinctcntr_id as CNTR_NO,'999'as POL_CODE,'2'as APP_STD_FLAG from insur_appl_cv_task,std_contractwhere insur_appl_cv_task.rel_key_no=std_contract.appl_noand insur_appl_cv_task.upd_time>=?and insur_appl_cv_task.upd_time<?

問題：
目前在相關表上的upd_time字段（也只此字段）均加了索引，可是要得到cntr_id時，仍然須要經過lookup rid到原始表中取出cntr_id字段（索引葉子節點沒有cntr_id）。這裏存在三點問題:
1). lookup rid以獲得cntr_id的操做，是upd_time索引查找範圍語句資源消耗的大頭，lookup rid只適合於小數據量；
2). lookup rid查找原表（由於有不少其餘任務訪問原表）會增長系統阻塞的機率及下降I/O量；
3). 可能會由於返回數據量大，使得upd_time索引失效。
優化方法：
每張表修改upd_time索引，經過在索引中包含相關字段（例如cntr_id），消除lookup rid操做，將整個增量查找過程轉換爲索引查找，能夠完全解決以上三個問題
2. 保單增量事實獲取：
當前現狀描述：
獲取保單增量事實的代碼可參見後文《優化前.sql》。
問題：
全部的相關表都是全表（這些表都是大表。固然，語句中有where語句，但仍然是從全表中過濾數據）JOIN，並經過tmp_f_cntr_v8_cntrrange增量範圍表提取出增量數據。這裏面存在如下問題：
1). 全部表（例如std_contract、cntr_sub_state等）均經過where語句，以及與tmp_f_cntr_v8_cntrrange表關聯獲得增量數據。這依賴於where語句中的索引，以及和tmp_f_cntr_v8_cntrrange表的關聯索引狀況，不然極可能引發全表掃描（這種狀況很是可怕）；
2). 在上文1.2肯定增量範圍時，upd_time索引以及lookup rid已經獲得了各個表的增量數據，獲取保單事實的時候對同一張表又作了一次重複的訪問，沒有必要；
3). 通過分析，對絕大部分表（都是大表）使用的字段都很是的少，進行全表操做很是的不划算。
解決方案：
整合「肯定保單增量範圍」、「提取保單事實」兩個步驟：
絕大部分表的「保單事實使用到的字段」上建立索引（若是包含upd_time，可同時解決增量範圍的問題），能夠將表掃描轉換爲索引掃描（字段少因此索引會很小），徹底能夠下降系統阻塞（由於只須要訪問索引，不須要和其它任務爭搶原表），下降I/O量；優化

優化前：spa

SELECT e.PROV_BRANCH_CODE AS PROV_BRANCH_CODE,
a.appl_no AS APPL_NO,
a.cntr_no AS CNTR_NO,
CASE
WHEN a.mr_type = 'M' THEN a.cntr_no
ELSE (SELECT TOP 1 cntr_no
FROM std_contractwith(nolock)
WHERE a.master_cntr_id = std_contract.cntr_id
AND std_contract.incr_flag <> '2')
END ASMASTER_CNTR_NO,
a.sg_no ASSG_NO,
Ltrim(Rtrim(CONVERT(VARCHAR(6), a.mgr_branch_no))) AS TOWN_BRANCH_CODE,
dbo.Pub_salesno(a.n_sales_branch_no, a.n_sales_code, 8)AS KEY_SALER_ID,
Isnull(a.n_sales_branch_no, '!') ASBRANCH_NO,
Isnull(a.n_sales_code, '!') AS AGENT_NO,
a.sales_channel AS SALES_CHANNEL_CODE,
'!' ASMANAGE_CHANNEL_CODE,
'!' ASCENTER_CODE,
a.pol_code ASPOL_CODE,
CASE
WHEN a.moneyin_itrvl = 'W' THEN 1
ELSE b.moneyin_dur
END ASMONEY_DUR,
a.moneyin_itrvl AS ITRVL_CODE,
d.ipsn_cust_no AS KEY_IPSN_CUST_ID,
(SELECT TOP1 hldr_cust_no
FROM psn_cntr_holderwith(nolock)
WHERE a.cntr_id = psn_cntr_holder.cntr_id
AND psn_cntr_holder.incr_flag <> '2'
ORDER BY psn_cntr_holder.upd_timeDESC) ASKEY_HLDR_CUST_ID,
CASE
WHEN b.insur_dur_unit = 'W' THEN 999
ELSE b.insur_dur
END ASINSUR_DUR,
a.in_force_date AS IN_FORCE_DATE,
a.sign_date AS SIGN_DATE,
a.cntr_term_date AS TERM_DATE,
CASE
WHEN a.cg_no IS NULL THEN (SELECT TOP 1 Cast(ext_key12 AS DATETIME)
FROM insur_appl_cv_task h WITH(nolock)
WHERE a.appl_no = h.rel_key_no
AND a.cntr_no = h.ext_key9
AND Isdate(h.ext_key12) = 1
AND h.incr_flag <> '2'
ORDER BY task_seq DESC)
ELSE (SELECT TOP 1 Cast(ext_key12 AS DATETIME)
FROM insur_appl_cv_taskh WITH(nolock)
WHERE a.appl_no = h.rel_key_no
AND a.cg_no = h.ext_key9
AND Isdate(h.ext_key12) = 1
AND h.incr_flag <> '2'
ORDER BY task_seq DESC)
END ASRESP_DATE,
a.mr_type AS MR_TYPE_CODE,
a.cntr_stat AS CNTR_STAT,
a.cntr_type AS CNTR_TYPE_CODE,
8 ASDATA_SRC_ID,
z.std_premium AS FACT_STD_PREM,
z.face_amnt AS FACT_FACE_AMNT,
d.ipsn_num AS FACT_IPSN_NUM,
a.cntr_term_cause AS CNTR_STOP_CODE,
z.stop_moneyin_date AS STOP_MONEYIN_DATE,
a.cntr_expiry_date AS EXPIRY_DATE,
f.out_force_date AS OUT_FORCE_DATE,
f.re_in_force_date AS RE_IN_FORCE_DATE,
Substring(a.appl_no, 1, 4) AS CARD_CODE,
CASE
WHEN a.Sales_channelin('OA', 'SP') THEN dbo.Pub_salesno(a.n_sales_branch_no, a.n_sales_code, 8)
ELSE '-1'
END ASKEY_SITE_ID,
Dateadd(day, -1, auto_in_force_date) ASENROLL_DATE,
(SELECT TOP1 contact_seq
FROM psn_cntr_holderwith(nolock)
WHERE a.cntr_id = psn_cntr_holder.cntr_id
AND psn_cntr_holder.incr_flag <> '2'
ORDER BY psn_cntr_holder.upd_timeDESC) ASCONTACT_SEQ,
CASE
WHEN g.cntr_no IS NOT NULL THEN 1
ELSE 0
END ASBANK_FLAG,
a.renew_flag AS RENEW_FLAG,
a.renew_times AS RENEW_TIMES,
'!' ASKEY_MERG_CUST_ID,
1 ASAPPL_STD_FLAG,
a.incr_flag AS INCR_FLAG,
a.cntr_id AS EXT_KEY,
a.moneyin_type ASMONEYIN_TYPE,
a.bank_code AS BANK_CODE,
a.bank_acc_no AS BANK_ACC_NO,
a.acc_cust_name AS ACC_CUST_NAME,
a.appl_date AS APPL_DATE,
a.lose_reg_num AS LOSE_NUM
FROM tmp_f_cntr_v8_cntrrangeewith(nolock)
INNER JOIN std_contract a WITH(nolock)
ON a.cntr_id = e.cntr_no
LEFT JOIN cntr_basic_statebwith(nolock)
ON a.cntr_id = b.cntr_id
AND b.incr_flag <> '2'
LEFT JOIN(SELECT cntr_id,
Min(ipsn_cust_no)AS ipsn_cust_no,
Count(cntr_id) AS ipsn_num,
e.PROV_BRANCH_CODE,
e.BATCH_NO
FROM tmp_f_cntr_v8_cntrrangee WITH(nolock)
INNER JOIN insured WITH(nolock)
ON cntr_id = e.cntr_no
WHERE incr_flag <> '2'
AND e.APP_STD_FLAG = '2'
AND ipsn_cust_no IS NOT NULL
GROUP BY cntr_id,
e.PROV_BRANCH_CODE,
e.BATCH_NO)AS d
ON a.cntr_id = d.cntr_id
AND d.BATCH_NO = e.BATCH_NO
AND d.PROV_BRANCH_CODE = e.PROV_BRANCH_CODE
LEFT JOIN(SELECT cntr_id,
Sum(std_premium) AS std_premium,
Sum(face_amnt) AS face_amnt,
Max(stop_moneyin_date)AS stop_moneyin_date,
e.PROV_BRANCH_CODE,
e.BATCH_NO
FROM tmp_f_cntr_v8_cntrrangee WITH(nolock)
INNER JOIN cntr_sub_state WITH(nolock)
ON cntr_id = e.cntr_no
WHERE incr_flag <> '2'
AND e.APP_STD_FLAG = '2'
GROUP BY cntr_id,
e.PROV_BRANCH_CODE,
e.BATCH_NO)AS z
ON a.cntr_id = z.cntr_id
AND z.BATCH_NO = e.BATCH_NO
AND z.PROV_BRANCH_CODE = e.PROV_BRANCH_CODE
LEFT JOIN(SELECT cntr_id,
Max(out_force_date) AS out_force_date,
Max(re_in_force_date)AS re_in_force_date,
e.PROV_BRANCH_CODE,
e.BATCH_NO
FROM tmp_f_cntr_v8_cntrrangee WITH(nolock)
INNER JOIN if_of_alt WITH(nolock)
ON cntr_id = e.cntr_no
WHERE incr_flag <> '2'
AND e.APP_STD_FLAG = '2'
GROUP BY cntr_id,
e.PROV_BRANCH_CODE,
e.BATCH_NO)AS f
ON a.cntr_id = f.cntr_id
AND f.BATCH_NO = e.BATCH_NO
AND f.PROV_BRANCH_CODE = e.PROV_BRANCH_CODE
LEFT JOIN(SELECT DISTINCT grant_tranlst.cntr_no,
e.PROV_BRANCH_CODE,
e.BATCH_NO
FROM tmp_f_cntr_v8_cntrrangee WITH(nolock)
INNER JOIN grant_tranlst WITH(nolock)
ON grant_tranlst.cntr_no = e.cntr_no
WHERE incr_flag <> '2'
AND e.APP_STD_FLAG = '2')AS g
ON a.cntr_no = g.cntr_no
AND g.BATCH_NO = e.BATCH_NO
AND g.PROV_BRANCH_CODE = e.PROV_BRANCH_CODE
WHERE a.cntr_no IS NOT NULL
AND a.cntr_type <> 'M'
AND a.pol_code <> '263'
AND e.APP_STD_FLAG = '2'
AND Isnumeric(a.renew_flag) = 1
AND e.PROV_BRANCH_CODE = ?
AND e.BATCH_NO = ?

優化後：code

SELECT a.appl_no AS APPL_NO,
a.cntr_no AS CNTR_NO,
CASE
WHEN a.mr_type = 'M' THEN a.cntr_no
ELSE (SELECT TOP 1 cntr_no
FROM std_contract WITH(nolock)
WHERE a.master_cntr_id = std_contract.cntr_id)
END ASMASTER_CNTR_NO,
a.sg_no ASSG_NO,
Ltrim(Rtrim(CONVERT(VARCHAR(6), a.mgr_branch_no))) AS TOWN_BRANCH_CODE,
dbo.Pub_salesno(a.n_sales_branch_no, a.n_sales_code, 8)AS KEY_SALER_ID,
Isnull(a.n_sales_branch_no, '!') AS BRANCH_NO,
Isnull(a.n_sales_code, '!') AS AGENT_NO,
a.sales_channel AS SALES_CHANNEL_CODE,
'!' ASMANAGE_CHANNEL_CODE,
'!' ASCENTER_CODE,
a.pol_code AS POL_CODE,
CASE
WHEN a.moneyin_itrvl = 'W' THEN 1
ELSE b.moneyin_dur
END ASMONEY_DUR,
a.moneyin_itrvl AS ITRVL_CODE,
d.ipsn_cust_no AS KEY_IPSN_CUST_ID,
(SELECT TOP1 hldr_cust_no
FROM psn_cntr_holderWITH(nolock)
WHERE a.cntr_id = psn_cntr_holder.cntr_id
AND psn_cntr_holder.incr_flag <> '2'
ORDER BY psn_cntr_holder.upd_time DESC) ASKEY_HLDR_CUST_ID,
CASE
WHEN b.insur_dur_unit = 'W' THEN 999
ELSE b.insur_dur
END ASINSUR_DUR,
a.in_force_date AS IN_FORCE_DATE,
a.sign_date ASSIGN_DATE,
a.cntr_term_date AS TERM_DATE,
CASE
WHEN a.cg_no IS NULL THEN (SELECT TOP 1 Cast(ext_key12 AS DATETIME)
FROM insur_appl_cv_task hWITH(nolock)
WHERE a.appl_no = h.rel_key_no
AND a.cntr_no = h.ext_key9
AND Isdate(h.ext_key12) = 1
AND h.incr_flag <> '2'
ORDER BY task_seq DESC)
ELSE (SELECT TOP 1 Cast(ext_key12 AS DATETIME)
FROM insur_appl_cv_taskh WITH(nolock)
WHERE a.appl_no = h.rel_key_no
AND a.cg_no = h.ext_key9
AND Isdate(h.ext_key12) = 1
AND h.incr_flag <> '2'
ORDER BY task_seq DESC)
END ASRESP_DATE,
a.mr_type AS MR_TYPE_CODE,
a.cntr_stat AS CNTR_STAT,
a.cntr_type ASCNTR_TYPE_CODE,
8 ASDATA_SRC_ID,
z.std_premium AS FACT_STD_PREM,
z.face_amnt ASFACT_FACE_AMNT,
d.ipsn_num AS FACT_IPSN_NUM,
a.cntr_term_cause AS CNTR_STOP_CODE,
z.stop_moneyin_date AS STOP_MONEYIN_DATE,
a.cntr_expiry_date AS EXPIRY_DATE,
f.out_force_date AS OUT_FORCE_DATE,
f.re_in_force_date ASRE_IN_FORCE_DATE,
Substring(a.appl_no, 1, 4) AS CARD_CODE,
CASE
WHEN a.Sales_channelin('OA', 'SP') THEN dbo.Pub_salesno(a.n_sales_branch_no, a.n_sales_code, 8)
ELSE '-1'
END ASKEY_SITE_ID,
Dateadd(day, -1, auto_in_force_date) ASENROLL_DATE,
(SELECT TOP1 contact_seq
FROM psn_cntr_holderWITH(nolock)
WHERE a.cntr_id = psn_cntr_holder.cntr_id
AND psn_cntr_holder.incr_flag <> '2'
ORDER BY psn_cntr_holder.upd_timeDESC) ASCONTACT_SEQ,
CASE
WHEN g.cntr_no IS NOT NULL THEN 1
ELSE 0
END ASBANK_FLAG,
a.renew_flag AS RENEW_FLAG,
a.renew_times AS RENEW_TIMES,
'!' ASKEY_MERG_CUST_ID,
1 ASAPPL_STD_FLAG,
a.incr_flag AS INCR_FLAG,
a.cntr_id AS EXT_KEY,
a.moneyin_type AS MONEYIN_TYPE,
a.bank_code AS BANK_CODE,
a.bank_acc_no ASBANK_ACC_NO,
a.acc_cust_name AS ACC_CUST_NAME,
a.appl_date AS APPL_DATE,
a.lose_reg_num ASLOSE_NUM
FROM csipdb.std_contract_tmpa WITH(nolock)
LEFT JOIN csipdb.cntr_basic_state_tmpbWITH(nolock)
ON a.cntr_id = b.cntr_id
AND b.incr_flag <> '2'
LEFT JOIN(SELECT cntr_id,
Min(ipsn_cust_no)AS ipsn_cust_no,
Count(cntr_id) AS ipsn_num
FROM csipdb.insured_tmpWITH(nolock)
WHERE incr_flag <> '2'
AND ipsn_cust_no IS NOT NULL
GROUP BY cntr_id)AS d
ON a.cntr_id = d.cntr_id
LEFT JOIN(SELECT cntr_id,
Sum(std_premium) AS std_premium,
Sum(face_amnt) AS face_amnt,
Max(stop_moneyin_date)AS stop_moneyin_date
FROM csipdb.cntr_sub_state_tmpWITH(nolock)
WHERE incr_flag <> '2'
GROUP BY cntr_id)AS z
ON a.cntr_id = z.cntr_id
LEFT JOIN(SELECT cntr_id,
Max(out_force_date) AS out_force_date,
Max(re_in_force_date)AS re_in_force_date
FROM csipdb.if_of_alt_tmpWITH(nolock)
WHERE incr_flag <> '2'
GROUP BY cntr_id)AS f
ON a.cntr_id = f.cntr_id
LEFT JOIN(SELECT DISTINCT cntr_no
FROM csipdb.grant_tranlst_tmp WITH(nolock)
WHERE incr_flag <> '2')AS g
ON a.cntr_no = g.cntr_no
WHERE a.cntr_no IS NOT NULL
AND a.cntr_type <> 'M'
AND a.pol_code <> '263'
AND Isnumeric(a.renew_flag) = 1

相關標籤/搜索

每日一句

每一个你不满意的现在，都有一个你没有努力的曾经。