我有一个SQL查询，每天都会使用过程将结果集插入meter_read_alert表中。
此过程每天大约需要3个小时。解释计划如下所示。
如何优化查询？通过查看解释计划，有哪些问题领域？
该表具有范围间隔分区和本地索引。每天大约发生900万行插入。
查询：
select * from (
SELECT
'SBILL' Scenario,
mfg_serial_num,
device_id,
channel_id,
read_value,
read_time,
date_val,
reading_hourly,
reading_daywise,
CASE
WHEN read_value IS NULL THEN'04'--(Meter goes stale)'
WHEN event_name = 'BACKFLOW' THEN'01'
WHEN event_name = 'TAMPER' THEN'02'
WHEN event_name = 'Tamper' THEN'02'
WHEN event_name = 'EMPTYPIPE' THEN '03' END AS CodeAlert_HOUR,
CASE WHEN reading_daywise BETWEEN CON.min_value AND CON.max_value THEN CON.alert_code END AS CodeAlert_DAY,
CASE
WHEN read_value IS NULL THEN'4'
WHEN event_name = 'BACKFLOW' THEN'4'
WHEN event_name = 'TAMPER' THEN'4'
WHEN event_name = 'Tamper' THEN'4'
WHEN event_name = 'EMPTYPIPE' THEN'4' END AS Priority_HOUR,
CASE WHEN reading_daywise BETWEEN CON.min_value AND CON.max_value THEN CON.priority END AS Priority_DAY,
CASE WHEN val_status='VAL' THEN 'A' WHEN val_status='EST' THEN 'E' ELSE val_status END Val_status,
desc_text,
CASE WHEN reading_hourly BETWEEN CONH.min_value AND CONH.max_value THEN CONH.alert_code END AS ConAlert_per_hour
FROM
(
SELECT d.mfg_serial_num,
d.id device_id,
mdr.channel_id,
mdr.read_value,
mdr.read_time,
To_date(To_char(mdr.read_time, 'dd/mm/yyyy')) date_val,
mdr.read_value - Lag(mdr.read_value, 1, 0) over ( PARTITION BY mdr.channel_id ORDER BY mdr.read_time) reading_hourly,
Max(mdr.read_value) keep (dense_rank last ORDER BY To_char (mdr.read_time, 'HH24:MI:SS')) over ( PARTITION BY mdr.channel_id,To_date(To_char(mdr.read_time, 'dd/mm/yyyy'))) -
Min(mdr.read_value) keep (dense_rank first ORDER BY To_char (mdr.read_time, 'HH24:MI:SS') ) over ( PARTITION BY mdr.channel_id,To_date(To_char(mdr.read_time, 'dd/mm/yyyy'))) as reading_daywise,
det.event_desc desc_text,
det.event_name,
mdr.val_status
FROM (select channel_id,read_value,read_time,val_status,device_id from mudr.register_read
WHERE read_time between (select TO_TIMESTAMP(sysdate-7) from dual) and (select TO_TIMESTAMP(sysdate-1) - INTERVAL '1' SECOND from dual)
--ORDER BY channel_id,read_time
)Mdr
left join (select channel_id, device_id from EIP.device_channel_rel --where eff_end_time is null
) DCR ON dcr.channel_id = mdr.channel_id
left join (select id, mfg_serial_num from EIP.device) d ON d.id = DCR.device_id
left join (select device_id, device_event_type_id,event_time ,
case when to_char(event_time,'mi') between '01' and '59' then trunc(event_time,'HH24') + interVal '60' minute else event_time end New_Event_Time
from MUDR.device_event
where event_time between (select TO_TIMESTAMP(sysdate-7) from dual) and (select TO_TIMESTAMP(sysdate-1) - INTERVAL '1' SECOND from dual)
) de ON mdr.device_id = de.device_id
and mdr.read_time= de.New_Event_Time
left join (select device_event_type_id,event_desc,event_name from MUDR.device_event_type) det ON det.device_event_type_id= de.device_event_type_id
--ORDER BY channel_id,read_time
)t1
left join (SELECT * FROM condition WHERE scenario = 'SBILL' AND unit_of_period_check='DAY' )CON ON t1.reading_daywise BETWEEN CON.min_value AND CON.max_value
left join (SELECT * FROM condition WHERE scenario = 'SBILL' AND unit_of_period_check='HOUR' )CONH ON t1.reading_hourly BETWEEN CONH.min_value AND CONH.max_value
ORDER BY channel_id,read_time) a
where not exists
(
select 1
from meter_read_alert b
WHERE a.scenario=b.scenario
AND a.mfg_serial_num= b.mfg_serial_num
AND a.device_id = b.device_id
and a.channel_id=b.channel_id
AND a.read_time=b.read_time
and b.read_time between (select TO_TIMESTAMP(sysdate-7) from dual) and (select TO_TIMESTAMP(sysdate-1) - INTERVAL '1' SECOND from dual)
);
计划哈希值：618507408
------------------------------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | Pstart| Pstop |
------------------------------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 106K| 46M| | 153K (1)| 00:00:07 | | |
| 1 | SORT ORDER BY | | 106K| 46M| 48M| 153K (1)| 00:00:07 | | |
| 2 | MERGE JOIN OUTER | | 106K| 46M| | 153K (1)| 00:00:07 | | |
| 3 | SORT JOIN | | 106K| 44M| 92M| 153K (1)| 00:00:07 | | |
| 4 | MERGE JOIN OUTER | | 106K| 44M| | 153K (1)| 00:00:07 | | |
| 5 | SORT JOIN | | 106K| 41M| 87M| 153K (1)| 00:00:07 | | |
|* 6 | HASH JOIN RIGHT ANTI | | 106K| 41M| 3432K| 145K (1)| 00:00:06 | | |
| 7 | PARTITION RANGE ITERATOR | | 87775 | 2400K| | 20495 (1)| 00:00:01 | KEY | KEY |
| 8 | VIEW | VW_SQ_1 | 87775 | 2400K| | 20495 (1)| 00:00:01 | | |
|* 9 | TABLE ACCESS BY LOCAL INDEX ROWID BATCHED | METER_READ_ALERT | 87775 | 2914K| | 20495 (1)| 00:00:01 | KEY | KEY |
|* 10 | INDEX RANGE SCAN | READTIME_ALL_IDX | 87775 | | | 81 (0)| 00:00:01 | KEY | KEY |
| 11 | FAST DUAL | | 1 | | | 2 (0)| 00:00:01 | | |
| 12 | FAST DUAL | | 1 | | | 2 (0)| 00:00:01 | | |
| 13 | VIEW | | 155K| 56M| | 124K (1)| 00:00:05 | | |
| 14 | WINDOW SORT | | 155K| 24M| 26M| 124K (1)| 00:00:05 | | |
| 15 | WINDOW SORT | | 155K| 24M| 26M| 124K (1)| 00:00:05 | | |
|* 16 | HASH JOIN RIGHT OUTER | | 155K| 24M| | 119K (1)| 00:00:05 | | |
| 17 | TABLE ACCESS FULL | DEVICE_EVENT_TYPE | 1513 | 124K| | 8 (0)| 00:00:01 | | |
|* 18 | HASH JOIN RIGHT OUTER | | 52840 | 4076K| | 119K (1)| 00:00:05 | | |
| 19 | PARTITION RANGE ITERATOR | | 3 | 54 | | 1 (0)| 00:00:01 | KEY | KEY |
| 20 | PARTITION HASH ALL | | 3 | 54 | | 1 (0)| 00:00:01 | 1 | LAST |
| 21 | VIEW | | 3 | 54 | | 1 (0)| 00:00:01 | | |
| 22 | TABLE ACCESS BY LOCAL INDEX ROWID BATCHED | DEVICE_EVENT | 3 | 48 | | 1 (0)| 00:00:01 | KEY | KEY |
|* 23 | INDEX RANGE SCAN | IX_DEVICE_EVENT_1 | 1 | | | 1 (0)| 00:00:01 | KEY | KEY |
| 24 | FAST DUAL | | 1 | | | 2 (0)| 00:00:01 | | |
| 25 | FAST DUAL | | 1 | | | 2 (0)| 00:00:01 | | |
|* 26 | HASH JOIN OUTER | | 52840 | 3147K| 2944K| 119K (1)| 00:00:05 | | |
|* 27 | HASH JOIN OUTER | | 52840 | 2322K| 2312K| 115K (1)| 00:00:05 | | |
| 28 | VIEW | | 51466 | 1708K| | 113K (1)| 00:00:05 | | |
| 29 | SORT ORDER BY | | 51466 | 1357K| 2240K| 113K (1)| 00:00:05 | | |
| 30 | PARTITION RANGE ITERATOR | | 51466 | 1357K| | 112K (1)| 00:00:05 | KEY | KEY |
| 31 | PARTITION HASH ALL | | 51466 | 1357K| | 112K (1)| 00:00:05 | 1 | LAST |
| 32 | TABLE ACCESS BY LOCAL INDEX ROWID BATCHED| REGISTER_READ | 51466 | 1357K| | 112K (1)| 00:00:05 | KEY | KEY |
|* 33 | INDEX SKIP SCAN | UK_REGISTER_READ | 1 | | | 112K (1)| 00:00:05 | KEY | KEY |
| 34 | FAST DUAL | | 1 | | | 2 (0)| 00:00:01 | | |
| 35 | FAST DUAL | | 1 | | | 2 (0)| 00:00:01 | | |
| 36 | INDEX FAST FULL SCAN | UK_DEVICE_CHANNEL_REL | 1482K| 15M| | 1443 (3)| 00:00:01 | | |
| 37 | TABLE ACCESS FULL | DEVICE | 1096K| 16M| | 4063 (2)| 00:00:01 | | |
|* 38 | FILTER | | | | | | | | |
|* 39 | SORT JOIN | | 1 | 26 | | 3 (34)| 00:00:01 | | |
|* 40 | EXTERNAL TABLE ACCESS FULL | CONDITION | 1 | 26 | | 2 (0)| 00:00:01 | | |
|* 41 | FILTER | | | | | | | | |
|* 42 | SORT JOIN | | 2 | 46 | | 3 (34)| 00:00:01 | | |
|* 43 | EXTERNAL TABLE ACCESS FULL | CONDITION | 2 | 46 | | 2 (0)| 00:00:01 | | |
------------------------------------------------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
6 - access("T1"."MFG_SERIAL_NUM"="ITEM_1" AND "T1"."DEVICE_ID"="ITEM_2" AND "T1"."CHANNEL_ID"="ITEM_3" AND "T1"."READ_TIME"="ITEM_4")
9 - filter("B"."SCENARIO"='SBILL')
10 - access("B"."READ_TIME">= (SELECT TO_TIMESTAMP(TO_CHAR(SYSDATE@!-7)) FROM "SYS"."DUAL" "DUAL") AND "B"."READ_TIME"<= (SELECT
TO_TIMESTAMP(TO_CHAR(SYSDATE@!-1))-INTERVAL'+00 00:00:01.000000' DAY(2) TO SECOND(6) FROM "SYS"."DUAL" "DUAL"))
16 - access("DEVICE_EVENT_TYPE_ID"(+)="DE"."DEVICE_EVENT_TYPE_ID")
18 - access("MDR"."DEVICE_ID"="DE"."DEVICE_ID"(+) AND "MDR"."READ_TIME"="DE"."NEW_EVENT_TIME"(+))
23 - access("EVENT_TIME">= (SELECT TO_TIMESTAMP(TO_CHAR(SYSDATE@!-7)) FROM "SYS"."DUAL" "DUAL") AND "EVENT_TIME"<= (SELECT
TO_TIMESTAMP(TO_CHAR(SYSDATE@!-1))-INTERVAL'+00 00:00:01.000000' DAY(2) TO SECOND(6) FROM "SYS"."DUAL" "DUAL"))
26 - access("ID"(+)="DEVICE_ID")
27 - access("CHANNEL_ID"(+)="MDR"."CHANNEL_ID")
33 - access("READ_TIME">= (SELECT TO_TIMESTAMP(TO_CHAR(SYSDATE@!-7)) FROM "SYS"."DUAL" "DUAL") AND "READ_TIME"<= (SELECT
TO_TIMESTAMP(TO_CHAR(SYSDATE@!-1))-INTERVAL'+00 00:00:01.000000' DAY(2) TO SECOND(6) FROM "SYS"."DUAL" "DUAL"))
filter("READ_TIME">= (SELECT TO_TIMESTAMP(TO_CHAR(SYSDATE@!-7)) FROM "SYS"."DUAL" "DUAL") AND "READ_TIME"<= (SELECT
TO_TIMESTAMP(TO_CHAR(SYSDATE@!-1))-INTERVAL'+00 00:00:01.000000' DAY(2) TO SECOND(6) FROM "SYS"."DUAL" "DUAL"))
38 - filter("T1"."READING_DAYWISE"<="CONDITION"."MAX_VALUE"(+))
39 - access(INTERNAL_FUNCTION("T1"."READING_DAYWISE")>=INTERNAL_FUNCTION("CONDITION"."MIN_VALUE"(+)))
filter(INTERNAL_FUNCTION("T1"."READING_DAYWISE")>=INTERNAL_FUNCTION("CONDITION"."MIN_VALUE"(+)))
40 - filter("SCENARIO"(+)='SBILL' AND "UNIT_OF_PERIOD_CHECK"(+)='DAY')
41 - filter("T1"."READING_HOURLY"<="CONDITION"."MAX_VALUE"(+))
42 - access(INTERNAL_FUNCTION("T1"."READING_HOURLY")>=INTERNAL_FUNCTION("CONDITION"."MIN_VALUE"(+)))
filter(INTERNAL_FUNCTION("T1"."READING_HOURLY")>=INTERNAL_FUNCTION("CONDITION"."MIN_VALUE"(+)))
43 - filter("SCENARIO"(+)='SBILL' AND "UNIT_OF_PERIOD_CHECK"(+)='HOUR')
答案 0 :(得分：2)
要选择的数据只有两个限制。主要的是时间跨度。另一个是您要排除meter_read_alert中已经存在的数据。您可以这样开始查询，并尽早应用限制：
with main as
(
select
s.scenario,
mdr.channel_id,
mdr.read_value,
mdr.read_time,
mdr.val_status,
mdr.device_id as mdr_device_id,
dcr.device_id as dcr_device_id,
d.mfg_serial_num as dcr_mfg_serial_num
from (select 'SBILL' as scenario from dual) s
cross join mudr.register_read mdr
left join from eip.device_channel_rel dcr on dcr.channel_id = mdr.channel_id
left join eip.device d on d.id = dcr.device_id
where mdr.read_time >= systimestamp - interval '7' day and mdr.read_time < systimestamp
and not exists
(
select null
from meter_read_alert mra
where mra.scenario = s.scenario
and mra.mfg_serial_num = d.mfg_serial_num
and mra.device_id = mdr.device_id
and mra.channel_id = mdr.channel_id
and mra.read_time = mdr.read_time
)
)
select ...
（您确定要在这里使用外部联接吗？如果内部联接足够，请进行这些内部联接。）
您的时间范围情况看起来很复杂。我希望我的查询正确无误。如果您希望它稍有不同，请进行更改。对于日期/时间，我们通常不使用BETWEEN，但是>=和<不受时间精度的影响（例如somedate < date '2019-12-01'将获得完整的11月，直到最后一个毫秒） ，十亿分之一秒或不包括12月的任何时间。
加速查询主要是关于提供适当的索引。我建议：
create index idx1 on mudr.register_read (read_time, channel_id);
create index idx2 on eip.device_channel_rel (channel_id, device_id);
create index idx3 on eip.device (device_id);
create index idx4 on meter_read_alert (read_time, channel_id, device_id, mfg_serial_num, scenario);
或者甚至使用覆盖索引：
create index idx1 on mudr.register_read (read_time, channel_id, read_value, val_status, device_id);
create index idx2 on eip.device_channel_rel (channel_id, device_id);
create index idx3 on eip.device (device_id);
create index idx4 on meter_read_alert (read_time, channel_id, device_id, mfg_serial_num, scenario);