I've got a problem where my count(*) will return the number of rows before distinct rows are filtered.
This is a simplified version of my query. Note that I'll extract a lot of other data from the tables, so group by won't return the same result, as I'd have to group by maybe 10 columns. The way it works is that m is a map mapping q, c and kl, so there can be several references to q.id. I only want one.
SELECT distinct on (q.id) count(*) over() as full_count
from q, c, kl, m
where c.id = m.chapter_id
and q.id = m.question_id
and q.active = 1
and c.class_id = m.class_id
and kl.id = m.class_id
order by q.id asc
If I run this i get full_count = 11210 while it only returns 9137 rows. If I run it without the distinct on (q.id), distinct on (q.id) is indeed the number of rows.
So it seems that the count function doesn't have access to the filtered rows. How can I solve this? Do I need to rethink my approach?
I'm not entirely sure what exactly you are trying to count, but this might get you started:
select id,
full_count,
id_count
from (
SELECT q.id,
count(*) over() as full_count,
count(*) over (partition by q.id) as id_count,
row_number() over (partition by q.id order by q.id) as rn
from q
join m on q.id = m.question_id
join c on c.id = m.chapter_id and c.class_id = m.class_id
join kl on kl.id = m.class_id
where q.active = 1
) t
where rn = 1
order by q.id asc
If you need the count per id, then the column id_count would be what you need. If you need the overall count, but just on row per id, then the full_count is probably what you want.
(note that I re-wrote your implicit join syntax to use explicit JOINs)
Can you use a subquery:
select qid, count(*) over () as full_count
from (SELECT distinct q.id
from q, c, kl, m
where c.id = m.chapter_id
and q.id = m.question_id
and q.active = 1
and c.class_id = m.class_id
and kl.id = m.class_id
) t
order by q.id asc
But the group by is the right approach. The key word distinct in selectis really just syntactic sugar for doing a group by on all non-aggregate-functioned columns.
Related
We have many old selects like this:
SELECT
tm."ID",tm."R_PERSONES",tm."R_DATASOURCE", ,tm."MATCHCODE",
d.NAME AS DATASOURCE,
p.PDID
FROM TABLE_MAPPINGS tm,
PERSONES p,
DATASOURCES d,
(select ID
from TABLE_MAPPINGS
where (R_PERSONES, MATCHCODE)
in (select
R_PERSONES, MATCHCODE
from TABLE_MAPPINGS
where
id in (select max(id)
from TABLE_MAPPINGS
group by MATCHCODE)
)
) tm2
WHERE tm.R_PERSONES = p.ID
AND tm.R_DATASOURCE=d.ID
and tm2.id = tm.id;
These are large tables, and queries take a long time.
How to rebuild them?
Thank you
You can query the table only once using something like (untested as you have not provided a minimal example of your create table statements or sample data):
SELECT *
FROM (
SELECT m.*,
COUNT(CASE WHEN rnk = 1 THEN 1 END)
OVER (PARTITION BY r_persones, matchcode) AS has_max_id
FROM (
SELECT tm.ID,
tm.R_PERSONES,
tm.R_DATASOURCE,
tm.MATCHCODE,
d.NAME AS DATASOURCE,
p.PDID,
RANK() OVER (PARTITION BY tm.matchcode ORDER BY tm.id DESC) As rnk
FROM TABLE_MAPPINGS tm
INNER JOIN PERSONES p ON tm.R_PERSONES = p.ID
INNER JOIN DATASOURCES d ON tm.R_DATASOURCE = d.ID
) m
)
WHERE has_max_id > 0;
First finding the maximum ID using the RANK analytic function and then finding all the relevant r_persones, matchcode pairs using conditional aggregation in a COUNT analytic function.
Note: you want to use the RANK or DENSE_RANK analytic functions to match the maximums as it can match multiple rows per partition; whereas ROW_NUMBER will only ever put a single row per partition first.
You're querying table_mappings 3 times; how about doing it only once?
WITH
tab_map
AS
(SELECT a.id,
a.r_persones,
a.matchcode,
a.datasource,
ROW_NUMBER ()
OVER (PARTITION BY a.matchcode ORDER BY a.id DESC) rn
FROM table_mappings a)
SELECT tm.id,
tm.r_persones,
tm.matchcode,
d.name AS datasource,
p.pdid
FROM tab_map tm
JOIN persones p ON p.id = tm.r_persones
JOIN datasources d ON d.id = tm.r_datasource
WHERE tm.rn = 1
I am having the following query and running it on postgress
select
p.id as p_id,
p.name as p_name,
p.tags,
p.creator,
p.value
p.creation_date,
cp.id as c_part_id,
fr.distance
count(*) OVER() AS total_item
from t_p p
left join t_c_part cp on p.id = cp.p_id
left join t_fl fr on p.id = fr.p_id
where p.name = 'test'
ORDER BY p.id ASC, p.name ASC
OFFSET 0 FETCH NEXT 25 ROWS only
What is missing here is that I also need to get max(p.value) and min(p.value) not affected by the "where" clause - so calculated from total (all) values.
I am dreaming that I can do it within one query and reduce the number of transactions.
Honestly not sure if it is possible!
What I tried is something like this ->
SELECT
(SELECT COUNT(*) from t_p) as count,
(SELECT json_agg(t.*) FROM (
SELECT * FROM t_p
where ***
) AS t) AS rows
But this one did not look really nice as it require additional JSON manipulation at the backend.
I discovered that I might try to use the "with" statement to create a temporary view so the where condition is only evaluated once, but did not succeed to make it works...
You can add the extra columns as scalar subqueries in the form (select min(value) from t_p). Their values are not related to the main query so they should be totally independent.
Your original query has some minor syntax issues (missing commas). I fixed those and the result is:
select
p.id as p_id,
p.name as p_name,
p.tags,
p.creator,
p.value,
p.creation_date,
cp.p_id as c_part_id,
fr.distance,
count(*) OVER() AS total_item,
(select min(value) from t_p) as min_value,
(select max(value) from t_p) as max_value
from t_p p
left join t_c_part cp on p.id = cp.p_id
left join t_fl fr on p.id = fr.p_id
where p.name = 'test'
ORDER BY p.id ASC, p.name ASC
OFFSET 0 FETCH NEXT 25 ROWS only
See running query (without any data) at DB Fiddle.
You can join to a sub-query that calculates both MIN & MAX.
...
from t_p p
left join t_c_part cp on p.id = cp.p_id
left join t_fl fr on p.id = fr.p_id
cross join (
select
min(value) as min_value
, max(value) as max_value
, avg(value) as avg_value
from t_p
) as v
...
Then use v.min_value and v.max_value in the select.
Doesn't even have to be a LATERAL.
You could get the minimum and maximum "on the side" like this:
select
p.id as p_id,
p.name as p_name,
p.tags,
p.creator,
p.value
p.creation_date,
cp.id as c_part_id,
fr.distance,
count(*) OVER() AS total_item,
p.min_value,
p.max_value
from (SELECT id,
name,
tags,
creator,
value,
creation_date,
min(value) OVER () AS min_value,
max(value) OVER () AS max_value,
FROM t_p) AS p
left join t_c_part cp on p.id = cp.p_id
left join t_fl fr on p.id = fr.p_id
where p.name = 'test'
ORDER BY p.id ASC, p.name ASC
OFFSET 0 FETCH NEXT 25 ROWS only;
Now that I got the Select all forums and get latest post too.. how? question answered, I am trying to write a query to select all threads in one particular forum and order them by the date of the latest post (column "updated_at").
This is my structure again:
forums forum_threads forum_posts
---------- ------------- -----------
id id id
parent_forum (NULLABLE) forum_id content
name user_id thread_id
description title user_id
icon views updated_at
created_at created_at
updated_at
last_post_id (NULLABLE)
I tried writing this query, and it works.. but not as expected: It doesn't order the threads by their last post date:
SELECT DISTINCT ON(t.id) t.id, u.username, p.updated_at, t.title
FROM forum_threads t
LEFT JOIN forum_posts p ON p.thread_id = t.id
LEFT JOIN users u ON u.id = p.user_id
WHERE t.forum_id = 3
ORDER BY t.id, p.updated_at DESC;
How can I solve this one?
Assuming you want a single row per thread and not all rows for all posts.
DISTINCT ON is still the most convenient tool. But the leading ORDER BY items have to match the expressions of the DISTINCT ON clause. If you want to order the result some other way, you need to wrap it into a subquery and add another ORDER BY to the outer query:
SELECT *
FROM (
SELECT DISTINCT ON (t.id)
t.id, u.username, p.updated_at, t.title
FROM forum_threads t
LEFT JOIN forum_posts p ON p.thread_id = t.id
LEFT JOIN users u ON u.id = p.user_id
WHERE t.forum_id = 3
ORDER BY t.id, p.updated_at DESC
) sub
ORDER BY updated_at DESC;
If you are looking for a query without subquery for some unknown reason, this should work, too:
SELECT DISTINCT
t.id
, first_value(u.username) OVER w AS username
, first_value(p.updated_at) OVER w AS updated_at
, t.title
FROM forum_threads t
LEFT JOIN forum_posts p ON p.thread_id = t.id
LEFT JOIN users u ON u.id = p.user_id
WHERE t.forum_id = 3
WINDOW w AS (PARTITION BY t.id ORDER BY p.updated_at DESC)
ORDER BY updated_at DESC;
There is quite a bit going on here:
The tables are joined and rows are selected according to JOIN and WHERE clauses.
The two instances of the window function first_value() are run (on the same window definition) to retrieve username and updated_at from the latest post per thread. This results in as many identical rows as there are posts in the thread.
The DISTINCT step is executed after the window functions and reduces each set to a single instance.
ORDER BY is applied last and updated_at references the OUT column (SELECT list), not one of the two IN columns (FROM list) of the same name.
Yet another variant, a subquery with the window function row_number():
SELECT id, username, updated_at, title
FROM (
SELECT t.id
, u.username
, p.updated_at
, t.title
, row_number() OVER (PARTITION BY t.id
ORDER BY p.updated_at DESC) AS rn
FROM forum_threads t
LEFT JOIN forum_posts p ON p.thread_id = t.id
LEFT JOIN users u ON u.id = p.user_id
WHERE t.forum_id = 3
) sub
WHERE rn = 1
ORDER BY updated_at DESC;
Similar case:
Return records distinct on one column but order by another column
You'll have to test which is faster. Depends on a couple of circumstances.
Forget the distinct on:
SELECT t.id, u.username, p.updated_at, t.title
FROM forum_threads t
LEFT JOIN forum_posts p ON p.thread_id = t.id
LEFT JOIN users u ON u.id = p.user_id
WHERE t.forum_id = 3
ORDER BY p.updated_at DESC;
I'm having an absolute brain fade
SELECT p.ProductCategory, f.ProductSubCategory, COUNT(*) AS Cnt
FROM Sales f
JOIN Products p ON f.ProductSubCategory = p.ProductSubCategory
GROUP BY p.ProductCategory, f.ProductSubCategory
ORDER BY 1,3 DESC
This shows me the count for each ProductSubCategory, I would like to see only the highest ProductSubCategory per ProductCategory.
I wish to see (I don't care about the Count value)
There are a couple of different ways to do this. One involves joining the results back to themselves and using the max aggregate. But since you are using SQL Server, you can use ROW_NUMBER to achieve the same result:
with cte as (
select p.productcategory, p.ProductSubCategory, COUNT(*) cnt,
ROW_NUMBER() over (partition by p.productcategory order by count(*) desc) rn
from products p
join sales s on p.ProductSubCategory = s.ProductSubCategory
group by p.productcategory, p.ProductSubCategory
)
select *
from cte
where rn = 1
You already got the answer, Please see the following code to. It may help you.
SELECT p.ProductCategory,
f.ProductSubCategory,
COUNT(*) AS Cnt
FROM Sales f
JOIN Products p ON f.ProductSubCategory = p.ProductSubCategory
JOIN (
SELECT p.ProductCategory,
f.ProductSubCategory,
ROW_NUMBER() OVER ( PARTITION BY p.ProductCategory,
f.ProductSubCategory
ORDER BY COUNT(*) DESC) [Row]
FROM Sales f
JOIN Products p ON f.ProductSubCategory = p.ProductSubCategory) Lu
ON P.ProductCategory = Lu.ProductCategory
AND f.ProductSubCategory = Lu.ProductSubCategory
WHERE Lu.Row = 1
GROUP By p.ProductCategory,
f.ProductSubCategory
Having some problems while trying to optimize my SQL.
I got 2 tables like this:
Names
id, analyseid, name
Analyses
id, date, analyseid.
I want to get the newest analyse from Analyses (ordered by date) for every name (they are unique) in Names. I can't really see how to do this without using 2 x nested selects.
My try (Dont get confused about the names. It's the same principle):
SELECT
B.id,
B.chosendatetime,
vStockNames.name
FROM
vStockNames
INNER JOIN
(
SELECT TOP 1
vAnalysesHistory.id,
vAnalysesHistory.chosendatetime,
vAnalysesHistory.companyid
FROM
vAnalysesHistory
ORDER BY
vAnalysesHistory.chosendatetime DESC
) AS B
ON
B.companyid = vStockNames.stockid
In my example the problem is that i only get 1 row returned (because of top 1). But if I exclude this, I can get multiple analyses of the same name.
Can you help me ? - THanks in advance.
SQL Server 2000+:
SELECT (SELECT TOP 1
a.id
FROM vAnalysesHistory AS a
WHERE a.companyid = n.stockid
ORDER BY a.chosendatetime DESC) AS id,
n.name,
(SELECT TOP 1
a.chosendatetime
FROM vAnalysesHistory AS a
WHERE a.companyid = n.stockid
ORDER BY a.chosendatetime DESC) AS chosendatetime
FROM vStockNames AS n
SQL Server 2005+, using CTE:
WITH cte AS (
SELECT a.id,
a.date,
a.analyseid,
ROW_NUMBER() OVER(PARTITION BY a.analyseid
ORDER BY a.date DESC) AS rk
FROM ANALYSES a)
SELECT n.id,
n.name,
c.date
FROM NAMES n
JOIN cte c ON c.analyseid = n.analyseid
AND c.rk = 1
...without CTE:
SELECT n.id,
n.name,
c.date
FROM NAMES n
JOIN (SELECT a.id,
a.date,
a.analyseid,
ROW_NUMBER() OVER(PARTITION BY a.analyseid
ORDER BY a.date DESC) AS rk
FROM ANALYSES a) c ON c.analyseid = n.analyseid
AND c.rk = 1
You're only asking for the TOP 1, so that's all you're getting. If you want one per companyId, you need to specify that in the SELECT on vAnalysesHistory. Of course, JOINs must be constant and do not allow this. Fortunately, CROSS APPLY comes to the rescue in cases like this.
SELECT
B.id,
B.chosendatetime,
vStockNames.name
FROM
vStockNames
CROSS APPLY
(
SELECT TOP 1
vAnalysesHistory.id,
vAnalysesHistory.chosendatetime,
vAnalysesHistory.companyid
FROM
vAnalysesHistory
WHERE companyid = vStockNames.stockid
ORDER BY
vAnalysesHistory.chosendatetime DESC
) AS B
You could also use ROW_NUMBER() to do the same:
SELECT
B.id,
B.chosendatetime,
vStockNames.name
FROM
vStockNames
INNER JOIN
(
SELECT
vAnalysesHistory.id,
vAnalysesHistory.chosendatetime,
vAnalysesHistory.companyid,
ROW_NUMBER() OVER (PARTITION BY companyid ORDER BY chosendatetime DESC) AS row
FROM
vAnalysesHistory
) AS B
ON
B.companyid = vStockNames.stockid AND b.row = 1
Personally I'm a fan of the first approach. It will likely be faster and is easier to read IMO.
Will something like this work for you?
;with RankedAnalysesHistory as
(
SELECT
vah.id,
vah.chosendatetime,
vah.companyid
,rank() over (partition by vah.companyid order by vah.chosendatetime desc) rnk
FROM
vAnalysesHistory vah
)
SELECT
B.id,
B.chosendatetime,
vsn.name
FROM
vStockNames vsn
join RankedAnalysesHistory as rah on rah.companyid = vsn.stockid and vah.rnk = 1
It seems to me that you only need SQL-92 for this. Of course, explicit documentation of the joining columns between the tables would help.
Simple names
SELECT B.ID, C.ChosenDate, N.Name
FROM (SELECT A.AnalyseID, MAX(A.Date) AS ChosenDate
FROM Analyses AS A
GROUP BY A.AnalyseID) AS C
JOIN Analyses AS B ON C.AnalyseID = B.AnalyseID AND C.ChosenDate = B.Date
JOIN Names AS N ON N.AnalyseID = C.AnalyseID
The sub-select generates the latest analysis for each company; the join with Analyses picks up the Analyse.ID value corresponding to that latest analysis, and the join with Names picks up the company name. (The C.ChosenDate in the select-list could be replaced by B.Date AS ChosenDate, of course.)
Complicated names
SELECT B.ID, C.ChosenDateTime, N.Name
FROM (SELECT A.CompanyID, MAX(A.ChosenDateTime) AS ChosenDateTime
FROM vAnalysesHistory AS A
GROUP BY A.CompanyID) AS C
JOIN vAnalysesHistory AS B ON C.CompanyID = B.CompanyID
AND C.ChosenDateTime = B.ChosenDateTime
JOIN vStockNames AS N ON N.AnalyseID = C.AnalyseID
Same query with systematic renaming (and slightly different layout to avoid horizontal scrollbars).