问题
Ok, I have experience in various databases, but not Oracle specifically, so still trying to wrap my head around how it handles things (i.e. the lack of 'Top' drives me mad!) Anyway, here's what we're trying to do...
Note: We're using Oracle 11. I understand a new feature was added to Oracle 12 (i.e. FETCH NEXT n ROWS ONLY) but we can't use that unfortunately.
To start, we have an articles table (technically it's a view, but I'm simplifying it here). Here are the relevant fields...
- ID (Integer, Primary Key)
- Title (String)
- Body (String)
- IsFavorite ('Y' or 'N')
- IsFeatured ('Y' or 'N')
- DatePublished (Date)
Multiple articles can be marked as favorites and/or featured.
What we want to return is a result set that has the following, in order...
- The most-recently published article with
IsFeatured = 'Y', if any. - The most-recently published article with
IsFavorite = 'Y'that's not the row from #1 (this avoids duplicates if the most recent featured is also the most recent favorited and instead selects the next favorited row) if any - The three most-recently published articles that are not #1 or #2, if any
Here's what I've come up with so far (cut/pasted/edited here so there may be some typos), but this just feels so 'clunky' to me... like I'm wasting a lot of unnecessary processing and iterations.
WITH mostRecentlyFeatured as (
Select * from (
Select 2 as MAJOR_SORT, A.*
From Articles A
where IsFeatured = 'Y'
order by DatePublished DESC
)
where ROWNUM = 1
),
mostRecentlyFavorited as (
Select * from (
Select 1 as MAJOR_SORT, A.*
From Articles A
minus Select * From mostRecentlyFeatured
where IsFavorite = 'Y'
order by DatePublished DESC
)
where ROWNUM = 1
),
topThreeOthers as (
Select * from (
select 0 as MAJOR_SORT, A.*
from Articles
minus
SELECT * from mostRecentlyFeatured
minus
SELECT * from mostRecentlyFavorited
order by DatePublished desc
)
where ROWNUM <= 3
),
finalRows as (
Select * from mostRecentlyFeatured
union all
Select * from mostRecentlyFavorited
union all
select * from topThreeOthers
)
Select * from finalRows
Order By MAJOR_SORT DESC,
DatePublished DESC;
This isn't the most uncommon of queries so I can't imagine there isn't a better way to do this, but I'm just not yet seeing it. So is there?
回答1:
the lack of 'Top' drives me mad
Since you are on Oracle 11g, TOP-n is not supported. So ROWNUM is the only way to go. See How ROWNUM works in pagination query.
For example, an OFFSET of 4 and FETCH next 4 using ROWNUM:
SQL> SELECT val
FROM (SELECT val, rownum AS rnum
FROM (SELECT val
FROM order_test
ORDER BY val)
WHERE rownum <= 8)
WHERE rnum >= 5;
VAL
----------
3
3
4
4
From Oracle 12c on ward, you could use the Top-n row limiting feature. Since you haven't provided any sample data, here is a simple demo:
SQL> select * from order_test order by val;
VAL
----------
1
1
2
2
3
3
4
4
5
5
6
6
7
7
8
8
9
9
10
10
20 rows selected.
First 4 rows :
SQL> SELECT val
2 FROM order_test
3 ORDER BY VAL
4 FETCH FIRST 4 ROWS ONLY;
VAL
----------
1
1
2
2
Next 4 rows(look at OFFSET) :
SQL> SELECT val
2 FROM order_test
3 ORDER BY VAL
4 OFFSET 4 ROWS FETCH NEXT 4 ROWS ONLY;
VAL
----------
3
3
4
4
Finally, next 4 rows with OFFSET 8 rows :
SQL> SELECT val
2 FROM order_test
3 ORDER BY VAL
4 OFFSET 8 ROWS FETCH NEXT 4 ROWS ONLY;
VAL
----------
5
5
6
6
Using the above approach, you could get the most-recently published article marked with 'IsFeatured' as follows:
SELECT * FROM articles
WHERE isfeatured = true
ORDER BY datepublished DESC
FETCH FIRST ONE ROW ONLY;
回答2:
Here's a possible solution to your query. It uses a set of CTEs to select the first n featured articles, the first m favourite articles (that are not in the first n featured articles) and a total of k articles:
- the first generates row numbers for featured and favourite articles;
- the second counts how many of the first
nfeatured articles are also int the firstmfavourites; and - the third generates a row number for articles sorted based on first
nfeatured, firstmfavourites (unless they were also amongst the firstnfeatured articles, in which case the next favourites are chosen) and date published.
Finally the top k rows are selected from the last CTE.
Here's a query where n = 2, m = 3 and k = 8:
-- first 2 (n) featured, first 3 (m) favourites, then 3 others (8 (k) total)
WITH favfeatrank AS (
-- rank articles by date and whether they are favourite or featured
SELECT ID, Title, IsFavorite, IsFeatured, DatePublished,
ROW_NUMBER() OVER (ORDER BY IsFeatured DESC, DatePublished DESC) feat,
ROW_NUMBER() OVER (ORDER BY IsFavorite DESC, DatePublished DESC) fav
FROM Articles
),
numfavfeat AS (
-- number of favourites that are also in the first n featured
-- use m in the next line
SELECT COUNT(CASE WHEN fav <= 3 THEN 1 END) AS featfav
FROM favfeatrank
WHERE feat <= 2 -- use n here
),
articlerank AS (
-- articles ranked according to featured, favourite and date published
SELECT ID, Title, IsFavorite, IsFeatured, DatePublished,
ROW_NUMBER() OVER (ORDER BY CASE WHEN feat <= 2 THEN 0 -- use n here
WHEN fav <= 3 + featfav THEN 1 -- use m here
ELSE 2
END,
DatePublished DESC) AS rn
FROM favfeatrank
CROSS JOIN numfavfeat
)
SELECT ID, Title, IsFavorite, IsFeatured, DatePublished
FROM articlerank
WHERE rn <= 8 -- use k here
Demo on dbfiddle
来源：https://stackoverflow.com/questions/61546104/is-there-a-more-efficient-way-to-aggregate-the-top-row-from-multiple-select-stat