性能下降返回光标与联合所有
Posted
技术标签:
【中文标题】性能下降返回光标与联合所有【英文标题】：Performance drop returning cursor with union all 【发布时间】：2021-11-23 09:05:48 【问题描述】：在一个游标中使用带有两个子查询的 UNION ALL 时，我面临着无法解决和不可能的性能下降（至少我认为这是问题所在）。在测试窗口中打开游标结果时，PL/SQL Developer 只是冻结。
如果我关闭任何子查询 - 一切正常。
如果我将整个查询从光标移到常规 SQL 查询窗口 - 一切都很好，无需关闭某些部分。
程序结构如下，期待任何帮助：
procedure p_proc(p_param varchar2,
outcur out sys_refcursor) is
begin
open outcur for
select *
from (select -- visible cols
si.item_full_name
, si.final_price
, si.full_price
, si.receipt_num
, si.receipt_date
, si.vendor_code
, case when det.br_summary is null and mr.motiv_rate_value is not null then mr.motiv_rate_value
when det.br_summary is not null then det.br_summary
end personal_bonus_amount
, case when det.br_summary is null and mr.motiv_rate_value is not null then 1
when det.br_summary is not null then det.cross_sale_kt
end personal_bonus_koeff
-- service cols
, case when det.br_summary is null and mr.motiv_rate_value is not null then 'approximate'
when det.br_summary is not null then 'definite'
end personal_bonus_type
, coalesce(det.sale_stream, mr.sale_stream, 'Not defined') item_group_name
, si.operation_type
, si.src
-- pagination
, row_number() over (order by si.receipt_date desc) rn
from (-- curr day
select b.cost final_price
, case when b.discount = 0 then null else b.price
end full_price
, b.doc_number receipt_num
, b.receipt_date receipt_date
, i.item_code vendor_code
, i.full_name item_full_name
, b.subsite code_op
, b.operator_id
, to_char(b.businessday, 'yyyymm') sale_period
, b.oper_type operation_type
, 'bill' src
from scheme.bills b
join scheme.items i on i.item_code = b.item
where b.businessday = trunc(p_date_to)
and b.subsite = p_office_id
and b.operator_id = p_emp_id
union all
-- prev days
select l.txn_amount final_price
, case when l.disc = 0 then null else l.price
end full_price
, t.receipt_num receipt_num
, t.ts receipt_date
, i.item_code vendor_code
, i.full_name item_full_name
, s.office_code code_op
, e.emp_code operator_id
, to_char(l.dt,'yyyymm') sale_period
, l.txn_type operation_type
, 'txn' src
from scheme.txn t
join scheme.txn_lines l on t.rtl_txn_id = l.rtl_txn_id
join scheme.items i on l.item_id = i.item_id
join scheme.offices s on t.subsite_id = s.subsite_id
join scheme.employees e on t.employee_id = e.employee_id
where t.ts between trunc(p_date_from) and trunc(p_date_to)
and t.subsite_id = v_op_id
and t.employee_id = v_emp_id
) si
/* fact */
left join scheme.sales_details det on si.sale_period = det.period
and si.code_op = det.op_code
and ltrim(si.operator_id,'0') = ltrim(det.tab_num,'0')
and si.receipt_num = det.rcpt_num
and si.vendor_code = det.item_article
/* prognosis */
left join scheme.rates mr on si.sale_period = mr.motiv_rate_period
and si.code_op = mr.code_op
and si.vendor_code = mr.code_1c
where 1 = 1
and si.final_price between nvl(p_price_from, si.final_price) and nvl(p_price_to, si.final_price)
/* if no filters */
and (item_group_cnt = 0 or coalesce(det.sale_stream, mr.sale_stream, 'Not defined') in (select * from table(p_item_group)))
and si.receipt_num = nvl(p_receipt_num, si.receipt_num)
)
where rn between p_page_num * p_page_size + 1 and (p_page_num + 1) * p_page_size;
end;
UPD解释游标中使用的整个查询的计划：
----------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost | Time |
----------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 10 | 32810 | 62 | 00:00:01 |
| * 1 | VIEW | | 10 | 32810 | 62 | 00:00:01 |
| * 2 | WINDOW SORT PUSHED RANK | | 2 | 2956 | 62 | 00:00:01 |
| 3 | NESTED LOOPS OUTER | | 2 | 2956 | 61 | 00:00:01 |
| 4 | NESTED LOOPS OUTER | | 2 | 2826 | 53 | 00:00:01 |
| 5 | VIEW | | 2 | 2728 | 46 | 00:00:01 |
| 6 | UNION-ALL | | | | | |
| 7 | NESTED LOOPS | | 1 | 138 | 32 | 00:00:01 |
| 8 | NESTED LOOPS | | 1 | 138 | 32 | 00:00:01 |
| 9 | PARTITION RANGE SINGLE | | 1 | 66 | 29 | 00:00:01 |
| * 10 | TABLE ACCESS BY LOCAL INDEX ROWID BATCHED | F003_BILL | 1 | 66 | 29 | 00:00:01 |
| * 11 | INDEX RANGE SCAN | IX_SUBSITE_DOCNUM_BUSINDAY_SEQ | 1 | | 5 | 00:00:01 |
| * 12 | INDEX RANGE SCAN | IX_D001_CODE_1C_ITEM_ID | 1 | | 2 | 00:00:01 |
| 13 | TABLE ACCESS BY INDEX ROWID | D001_ITEM | 1 | 72 | 3 | 00:00:01 |
| 14 | NESTED LOOPS | | 1 | 183 | 14 | 00:00:01 |
| 15 | NESTED LOOPS | | 1 | 183 | 14 | 00:00:01 |
| 16 | NESTED LOOPS | | 1 | 104 | 12 | 00:00:01 |
| 17 | NESTED LOOPS | | 1 | 70 | 7 | 00:00:01 |
| 18 | NESTED LOOPS | | 1 | 30 | 4 | 00:00:01 |
| 19 | TABLE ACCESS BY INDEX ROWID | D005_EMPLOYEE | 1 | 18 | 3 | 00:00:01 |
| * 20 | INDEX UNIQUE SCAN | PK_D005 | 1 | | 2 | 00:00:01 |
| 21 | TABLE ACCESS BY INDEX ROWID | D018_SUBSITE | 1 | 12 | 1 | 00:00:01 |
| * 22 | INDEX UNIQUE SCAN | PK_D018 | 1 | | 0 | 00:00:01 |
| 23 | PARTITION RANGE ITERATOR | | 1 | 40 | 3 | 00:00:01 |
| 24 | PARTITION HASH SINGLE | | 1 | 40 | 3 | 00:00:01 |
| * 25 | TABLE ACCESS FULL | F007_RTL_TXN | 1 | 40 | 3 | 00:00:01 |
| * 26 | TABLE ACCESS BY GLOBAL INDEX ROWID BATCHED | F008_RTL_TXN_LI | 1 | 34 | 5 | 00:00:01 |
| * 27 | INDEX RANGE SCAN | IX_F008_RTL_TXN_ID | 7 | | 3 | 00:00:01 |
| * 28 | INDEX UNIQUE SCAN | PK_D001 | 1 | | 1 | 00:00:01 |
| 29 | TABLE ACCESS BY INDEX ROWID | D001_ITEM | 1 | 79 | 2 | 00:00:01 |
| * 30 | TABLE ACCESS BY INDEX ROWID BATCHED | T_OP_MOTIVATION_RATE_MYRTK | 1 | 49 | 7 | 00:00:01 |
| * 31 | INDEX RANGE SCAN | IDX02_CODE_OP_1C | 3 | | 3 | 00:00:01 |
| * 32 | TABLE ACCESS BY INDEX ROWID BATCHED | DET_SALES_PPT_DWH | 1 | 65 | 4 | 00:00:01 |
| * 33 | INDEX RANGE SCAN | IDX_03_RCPT_NUM | 3 | | 2 | 00:00:01 |
----------------------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
------------------------------------------
* 1 - filter("RN">=1 AND "RN"<=10)
* 2 - filter(ROW_NUMBER() OVER ( ORDER BY INTERNAL_FUNCTION("SI"."RECEIPT_DATE") DESC )<=10)
* 10 - filter("F003"."OPERATOR_ID"='000189513' AND "F003"."COST">=TO_NUMBER(TO_CHAR("F003"."COST")) AND "F003"."COST"<=TO_NUMBER(TO_CHAR("F003"."COST")))
* 11 - access("F003"."SUBSITE"='S165' AND "F003"."BUSINESSDAY"=TO_DATE(' 2021-11-23 00:00:00', 'syyyy-mm-dd hh24:mi:ss'))
* 11 - filter("F003"."BUSINESSDAY"=TO_DATE(' 2021-11-23 00:00:00', 'syyyy-mm-dd hh24:mi:ss') AND "F003"."DOC_NUMBER" IS NOT NULL)
* 12 - access("I"."D001_CODE_1C"="F003"."ITEM")
* 12 - filter("I"."D001_CODE_1C" IS NOT NULL)
* 20 - access("E"."EMPLOYEE_ID"=3561503543)
* 22 - access("S"."SUBSITE_ID"=29260)
* 25 - filter("T"."EMPLOYEE_ID"=3561503543 AND "T"."SUBSITE_ID"=29260 AND "T"."F007_TS"<=TO_DATE(' 2021-11-23 00:00:00', 'syyyy-mm-dd hh24:mi:ss') AND "T"."F007_RCPT_NUM_1C" IS NOT NULL)
* 26 - filter("L"."F008_AMOUNT">=TO_NUMBER(TO_CHAR("L"."F008_AMOUNT")) AND "L"."F008_AMOUNT"<=TO_NUMBER(TO_CHAR("L"."F008_AMOUNT")))
* 27 - access("T"."RTL_TXN_ID"="L"."RTL_TXN_ID")
* 28 - access("L"."ITEM_ID"="I"."ITEM_ID")
* 30 - filter("SI"."SALE_PERIOD"="MR"."MOTIV_RATE_PERIOD"(+))
* 31 - access("SI"."CODE_OP"="MR"."CODE_OP"(+) AND "SI"."VENDOR_CODE"="MR"."CODE_1C"(+))
* 32 - filter("SI"."CODE_OP"="DET"."OP_CODE"(+) AND "SI"."VENDOR_CODE"="DET"."ITEM_ARTICLE"(+) AND "DET"."ITEM_ARTICLE"(+) IS NOT NULL AND "DET"."PERIOD"(+)=TO_NUMBER("SI"."SALE_PERIOD") AND
LTRIM("SI"."OPERATOR_ID",'0')=LTRIM("DET"."TAB_NUM_RTK"(+),'0'))
* 33 - access("SI"."RECEIPT_NUM"="DET"."RCPT_NUM"(+))
* 33 - filter("DET"."RCPT_NUM"(+) IS NOT NULL)
【问题讨论】：
请对选择执行解释计划并发布结果 添加了解释计划，无论使用什么“页面大小”，性能都保持良好。 【参考方案1】：实际解决方案
设法从 DBA 获得过程执行计划。问题是优化器在过程中执行查询时选择了另一个索引来加入 scheme.sales_details 表。添加了 INDEX HINT 与常规查询中使用的相同索引，一切正常。
下方已弃用的想法
据我了解，问题出在 Oracle 优化器中，它“认为”首先执行 UNION ALL 比将谓词推入子查询要好。将这个 union 分成两个单独的查询让他毫不犹豫地推动 pred。
也许这可以通过玩一些提示来解决，这暂时是 wip。
临时解决方法是重新组合查询，从此结构开始
select *
from (select row_number() rn
, u.*
from (select *
from first_query
union all
select *
from second_query) u
-- some joins
join first_table ft
join second_table st
-- predicate block
where 1=1
and a = b
)
where rn between c and d;
到这里
select *
from (select row_number() rn
, u.*
from (select *
from first_query) u
-- some joins
join first_table ft
join second_table st
-- predicate block
where 1=1
and a = b
union all
select row_number() rn
, u.*
from (select *
from second_query) u
-- some joins
join first_table ft
join second_table st
-- predicate block
where 1=1
and a = b
)
where rn between c and d;
这不是完美的解决方案，因为它使 JOIN 部分加倍，但至少它有效。
【讨论】：
以上是关于性能下降返回光标与联合所有的主要内容，如果未能解决你的问题，请参考以下文章