我有一张表，其中包含系统的到达和退出及其时间段。我也有到达和退出的预测。我想计算不完整的预测期间的头部的起始和结束计数（理想情况下使用合并和连接）。
要重现的数据：
create table HEAD_COUNT
(
PERIOD_START DATE,
HEAD_COUNT_START NUMBER ,
HEAD_COUNT_END NUMBER ,
ARRIVAL NUMBER ,
EXITS NUMBER
);
Insert into HEAD_COUNT values ('01-DEC-18',0,0,13,275);
Insert into HEAD_COUNT values ('01-NOV-18',0,0,0,46);
Insert into HEAD_COUNT values ('01-OCT-18',0,0,6,61);
Insert into HEAD_COUNT values ('01-SEP-18',0,0,275,1292);
Insert into HEAD_COUNT values ('01-AUG-18',0,0,46,1790);
Insert into HEAD_COUNT values ('01-JUL-18',0,0,61,17);
Insert into HEAD_COUNT values ('01-JUN-18',0,0,1292,3);
Insert into HEAD_COUNT values ('01-MAY-18',0,0,1790,15);
Insert into HEAD_COUNT values ('01-APR-18',0,0,17,158);
Insert into HEAD_COUNT values ('01-MAR-18',0,0,3,9);
Insert into HEAD_COUNT values ('01-FEB-18',0,0,15,0);
Insert into HEAD_COUNT values ('01-JAN-18',0,0,158,4);
Insert into HEAD_COUNT values ('01-DEC-17',0,0,9,179);
Insert into HEAD_COUNT values ('01-NOV-17',250,186,0,64);
Insert into HEAD_COUNT values ('01-OCT-17',276,250,4,30);
Insert into HEAD_COUNT values ('01-SEP-17',638,276,179,541);
我最初的尝试：
merge INTO head_count h USING
(SELECT period_start,
head_count_start,
head_count_end,
arrival,
exits
FROM head_count
CONNECT BY prior period_start = add_months(period_start, -1)
START WITH period_start = TRUNC(sysdate, 'Month')
)
src ON (h.period_start = src.period_start)
WHEN matched THEN
UPDATE
SET h.head_count_start = src.head_count_end,
h.head_count_end = h.head_count_start + h.arrival - h.exits
但是，它没有产生正确的结果。感谢任何帮助！
答案 0 :(得分：2)
尝试使用Oracle的recursive subquery factoring clause代替CONNECT BY。它需要一些额外的输入，但根据我的经验，语法更有意义。
例如：
--Update the HEAD_COUNT using the last arrival and exits.
merge into head_count using
(
--Summarize head count start and end.
with head_count_cte(period_start, head_count_start, head_count_end, arrival, exits) as
(
--First month.
select period_start, head_count_start, head_count_start + arrival - exits head_count_end, arrival, exits
from head_count
where period_start = (select min(period_start) from head_count)
union all
--All months after the first month.
select
head_count.period_start,
head_count_cte.head_count_end as head_count_start,
head_count_cte.head_count_end + head_count.arrival - head_count.exits as head_count_end,
head_count.arrival, head_count.exits
from head_count_cte
join head_count
on add_months(head_count_cte.period_start, 1) = head_count.period_start
)
select *
from head_count_cte
order by period_start desc
) recursive_head_count
on (head_count.period_start = recursive_head_count.period_start)
when matched then update set
head_count.head_count_start = recursive_head_count.head_count_start,
head_count.head_count_end = recursive_head_count.head_count_end;
这是一个包含所有数据和结果的完整SQL Fiddle。
答案 1 :(得分：2)
递归可以用分析函数代替，这应该更快一点：
演示：http://sqlfiddle.com/#!4/87500c/2
select t.* ,
first_value( HEAD_COUNT_START ) over (order by PERIOD_START )
+
coalesce(sum( ARRIVAL - EXITS) over (order by PERIOD_START
ROWS between unbounded preceding and 1 preceding ),0) as Starting_headcount,
first_value( HEAD_COUNT_START ) over (order by PERIOD_START )
+
sum( ARRIVAL - EXITS) over (order by PERIOD_START ) as Ending_Headcount
from HEAD_COUNT t
order by period_start;
| PERIOD_START | HEAD_COUNT_START | HEAD_COUNT_END | ARRIVAL | EXITS | STARTING_HEADCOUNT | ENDING_HEADCOUNT |
|----------------------|------------------|----------------|---------|-------|--------------------|------------------|
| 2017-09-01T00:00:00Z | 638 | 276 | 179 | 541 | 638 | 276 |
| 2017-10-01T00:00:00Z | 276 | 250 | 4 | 30 | 276 | 250 |
| 2017-11-01T00:00:00Z | 250 | 186 | 0 | 64 | 250 | 186 |
| 2017-12-01T00:00:00Z | 0 | 0 | 9 | 179 | 186 | 16 |
| 2018-01-01T00:00:00Z | 0 | 0 | 158 | 4 | 16 | 170 |
| 2018-02-01T00:00:00Z | 0 | 0 | 15 | 0 | 170 | 185 |
| 2018-03-01T00:00:00Z | 0 | 0 | 3 | 9 | 185 | 179 |
| 2018-04-01T00:00:00Z | 0 | 0 | 17 | 158 | 179 | 38 |
| 2018-05-01T00:00:00Z | 0 | 0 | 1790 | 15 | 38 | 1813 |
| 2018-06-01T00:00:00Z | 0 | 0 | 1292 | 3 | 1813 | 3102 |
| 2018-07-01T00:00:00Z | 0 | 0 | 61 | 17 | 3102 | 3146 |
| 2018-08-01T00:00:00Z | 0 | 0 | 46 | 1790 | 3146 | 1402 |
| 2018-09-01T00:00:00Z | 0 | 0 | 275 | 1292 | 1402 | 385 |
| 2018-10-01T00:00:00Z | 0 | 0 | 6 | 61 | 385 | 330 |
| 2018-11-01T00:00:00Z | 0 | 0 | 0 | 46 | 330 | 284 |
| 2018-12-01T00:00:00Z | 0 | 0 | 13 | 275 | 284 | 22 |
然后上面的查询可以用作MERGE语句中的子查询：
MERGE INTO HEAD_COUNT h
USING(
select t.* ,
first_value( HEAD_COUNT_START ) over (order by PERIOD_START )
+
coalesce(sum( ARRIVAL - EXITS) over (order by PERIOD_START
ROWS between unbounded preceding and 1 preceding ),0) as Starting_headcount,
first_value( HEAD_COUNT_START ) over (order by PERIOD_START )
+
sum( ARRIVAL - EXITS) over (order by PERIOD_START ) as Ending_Headcount
from HEAD_COUNT t
) t
ON (t.PERIOD_START = h.PERIOD_START )
WHEN MATCHED THEN UPDATE
SET h.HEAD_COUNT_START = t.Starting_headcount,
h.HEAD_COUNT_END = t.Ending_Headcount
演示：http://sqlfiddle.com/#!4/34268/1
| PERIOD_START | HEAD_COUNT_START | HEAD_COUNT_END | ARRIVAL | EXITS |
|----------------------|------------------|----------------|---------|-------|
| 2017-09-01T00:00:00Z | 638 | 276 | 179 | 541 |
| 2017-10-01T00:00:00Z | 276 | 250 | 4 | 30 |
| 2017-11-01T00:00:00Z | 250 | 186 | 0 | 64 |
| 2017-12-01T00:00:00Z | 186 | 16 | 9 | 179 |
| 2018-01-01T00:00:00Z | 16 | 170 | 158 | 4 |
| 2018-02-01T00:00:00Z | 170 | 185 | 15 | 0 |
| 2018-03-01T00:00:00Z | 185 | 179 | 3 | 9 |
| 2018-04-01T00:00:00Z | 179 | 38 | 17 | 158 |
| 2018-05-01T00:00:00Z | 38 | 1813 | 1790 | 15 |
| 2018-06-01T00:00:00Z | 1813 | 3102 | 1292 | 3 |
| 2018-07-01T00:00:00Z | 3102 | 3146 | 61 | 17 |
| 2018-08-01T00:00:00Z | 3146 | 1402 | 46 | 1790 |
| 2018-09-01T00:00:00Z | 1402 | 385 | 275 | 1292 |
| 2018-10-01T00:00:00Z | 385 | 330 | 6 | 61 |
| 2018-11-01T00:00:00Z | 330 | 284 | 0 | 46 |
| 2018-12-01T00:00:00Z | 284 | 22 | 13 | 275 |