计算序列中上升值的边界

Question

我有连续 id 和波动压力值 druck 的记录。我的目标是找出压力增加发生在哪些值之间，并相应地获得该范围的最低和最高 id。我已经有一个使用经典 SQL 的解决方案，但我对更有效的解决方案感兴趣。

以下是一些典型条目：

create table produktion ( id int, druck numeric (5, 2 ) );
insert into produktion
values (1, 1.35), (2, 1.37), (3, 1.45), ( 4, 1.48), ( 5, 1.51), ( 6, 1.39),
       (7, 1.53), (8, 1.55), (9, 1.62), (10, 1.39), (11, 1.32), (12, 1.28);

我期望这样的结果：

========================
| erste_id | letzte_id |
|=======================
|        1 |         5 |
|        6 |         9 |
========================

这是当前使用的查询：

SELECT p1.id AS erste_id,  -- first id
       p2.id AS letzte_id  -- last id
FROM   produktion AS p1,
       produktion AS p2
WHERE  p1.id < p2.id AND
       NOT EXISTS( SELECT *
                   FROM   produktion AS p3,
                          produktion AS p4
                   WHERE  p3.druck <= p4.druck AND
                          p4.id = p3.id - 1 AND
                          p3.id BETWEEN p1.id + 1 AND p2.id OR
                          p3.id = p1.id - 1 AND p3.druck < p1.druck OR
                          p3.id = p2.id + 1 AND p3.druck > p2.druck )

更新

我忘了说在相等的值下，一个序列被认为是中断的。

更新 2

我对 FatFreddy 的查询做了一些小改动，使其符合我的要求。

WITH
     find_boundaries AS (
   SELECT id,
          CASE WHEN lag(  druck, 1, druck ) over ( ORDER BY id ) <  druck AND
                    druck <  lead( druck, 1, druck ) OVER ( ORDER BY id ) THEN NULL
               WHEN lag(  druck, 1, druck ) OVER ( ORDER BY id ) <  druck AND
                    druck >= lead( druck, 1, druck ) OVER ( ORDER BY id ) THEN 'end_run'
               WHEN lag(  druck, 1, druck ) OVER ( ORDER BY id ) >= druck AND 
                    druck <  lead( druck, 1, druck ) OVER ( ORDER BY id ) THEN 'start_run'
          END AS row_type
   FROM produktion ),

     start_boundary AS (
   SELECT id,
          row_number() OVER ( ORDER BY id) AS correlated_start_row
   FROM   find_boundaries
   WHERE  row_type = 'start_run' ),

     end_boundary AS (
   SELECT id,
          row_number() OVER ( ORDER BY id ) AS correlated_end_row
   FROM   find_boundaries
   WHERE  row_type = 'end_run' )

SELECT s.id AS anfang,
       e.id AS ende
FROM   start_boundary AS s
       JOIN end_boundary AS e
         ON s.correlated_start_row = e.correlated_end_row
ORDER  BY 1

Answer 1

对我提出的查询进行 EXPLAIN ANALYZE（12 个条目）

Planning time: 0.184 ms
Execution time: 3.525 ms

我可以给你一个计划时间更多但执行时间更少的查询

Planning time: 0.290 ms
Execution time: 0.269 ms

100 个条目

yours
    Planning time: 0.193 ms
    Execution time: 10457.269 ms
mine
    Planning time: 0.342 ms
    Execution time: 1.175 ms

1000 个条目

yours
    no result after 5 minutes
mine
    Planning time: 0.343 ms
    Execution time: 5.866 ms

1000000 个条目

yours
    no result after 5 minutes
mine
    Planning time: 0.348 ms
    Execution time: 5217.038 ms

用散列连接替换一些嵌套循环。使用一些 window 函数 https://www.postgresql.org/docs/9.5/static/functions-window.html 我的查询看起来像这样：

with temp_flow as
(
select
id,
case when 
    id = 1 and lead(produktion.druck,1, 0::numeric) over(order by id)   <=  produktion.druck then 'fall' 
when 
    id = 1 and lead(produktion.druck,1, 0::numeric) over(order by id)  >  produktion.druck then 'start raise' 
when 
    lag(produktion.druck,1, 0::numeric) over(order by id)  < produktion.druck  and    lead(produktion.druck,1, 0::numeric) over(order by id)   > produktion.druck then 'raise'
when 
   lag(produktion.druck,1, 0::numeric) over(order by id)  < produktion.druck  and    lead(produktion.druck,1, 0::numeric) over(order by id)   <=  produktion.druck then 'end raise'
when 
   lag(produktion.druck,1, 0::numeric) over(order by id)  = produktion.druck  and    lead(produktion.druck,1, 0::numeric) over(order by id)   <=  produktion.druck then 'fall'
when 
   lag(produktion.druck,1, 0::numeric) over(order by id)  >=  produktion.druck  and    lead(produktion.druck,1, 0::numeric) over(order by id)   >  produktion.druck then 'start raise'
else 'fall'
end as way,

lag(produktion.druck,1, 0::numeric) over(order by id) as beforelag,
produktion.druck,
lead(produktion.druck,1, 0::numeric) over(order by id)  as afterlead
from produktion
order by id
),

temp_start as  
(
select
id,
row_number() over () as xrow
 from temp_flow
where temp_flow.way = 'start raise'
order by id
),

temp_end as 
(
select
id,
row_number() over () as xrow
 from temp_flow
where temp_flow.way = 'end raise'
order by id
)

select
temp_start.id as first,
temp_end.id as last
from 
temp_start
join temp_end on temp_end.xrow = temp_start.xrow

在你的设置中没有声明用 equal druck 应该做什么，所以你可以在某个地方放一个 >= 而不是 > 你可以使用数字来代替，例如'start raise' 我用它来提高可读性，它应该比文本比较更快。

祝您在系统上使用大数据测试愉快。

编辑：正确的结果做了更多的案例，获取了一些奇怪的案例

计算序列中上升值的边界

Calculating the boundaries of rising values in a sequence

postgresql

gaps-and-islands

更新

更新 2