I have the following query which works great in Postgres 9.1:
SELECT users.id, GREATEST(
COALESCE(MAX(messages.created_at), '2012-07-25 16:05:41.870117'),
COALESCE(MAX(phone_calls.created_at), '2012-07-25 16:05:41.870117')
) AS latest_interaction
FROM users LEFT JOIN messages ON users.id = messages.user_id
LEFT JOIN phone_calls ON users.id = phone_calls.user_id
GROUP BY users.id
ORDER BY latest_interaction DESC
LIMIT 5;
But what I want to do is something like this:
SELECT users.id, GREATEST(
COALESCE(MAX(messages.created_at), '2012-07-25 16:05:41.870117'),
COALESCE(MAX(phone_calls.created_at), '2012-07-25 16:05:41.870117')
) AS latest_interaction
FROM users LEFT JOIN messages ON users.id = messages.user_id
LEFT JOIN phone_calls ON users.id = phone_calls.user_id
GROUP BY users.id
ORDER BY
CASE WHEN(
latest_interaction > '2012-09-05 16:05:41.870117')
THEN 0
WHEN(latest_interaction > '2012-09-04 16:05:41.870117')
THEN 2
WHEN(latest_interaction > '2012-09-04 16:05:41.870117')
THEN 3
ELSE 4
END
LIMIT 5;
And I get the following error:
ERROR: column "latest_interaction" does not exist
It seems like I cannot use the alias for the aggregate latest_interaction in the order by clause with a CASE statement.
Are there any workarounds for this?
Try to wrap it as a subquery:
SELECT *
FROM
(
SELECT users.id,
GREATEST(
COALESCE(MAX(messages.created_at), '2012-07-25 16:05:41.870117'),
COALESCE(MAX(phone_calls.created_at), '2012-07-25 16:05:41.870117')
) AS latest_interaction
FROM users LEFT JOIN messages ON users.id = messages.user_id
LEFT JOIN phone_calls ON users.id = phone_calls.user_id
GROUP BY users.id
) Sub
ORDER BY
CASE WHEN(
latest_interaction > '2012-09-05 16:05:41.870117')
THEN 0
WHEN(latest_interaction > '2012-09-04 16:05:41.870117')
THEN 2
WHEN(latest_interaction > '2012-09-04 16:05:41.870117')
THEN 3
ELSE 4
END
LIMIT 5;
The PG manual says the ORDER BY expression:
Each expression can be the name or ordinal number of an output column (SELECT list item), or it can be an arbitrary expression formed from input-column values.
The sub-query solution from #Mahmoud will work, or you can create the ORDER BY using the original columns messages.created_at or phone_calls.created_at
Related
I have the query with subquery for result output as yesterday_sum column. I need to filter only rows where yesterday_sum > 1 but can't add HAVING sum(p.profit_percent) / :workDays > 1 AND yesterday_sum > 1 because yesterday_sum is not part of the GROUP BY. And I can't add a condition for positions p because yesterday_sum is not the positions column.
SELECT u.id AS id,
u.nickname AS title,
sum(p.profit_percent) / :workDays AS middle,
(
SELECT sum(ps.profit_percent)
FROM positions ps
WHERE ps.user_id = u.id
AND ps.open_at BETWEEN
:dateYesterday AND
:dateYesterday + INTERVAL '1 day'
GROUP BY (ps.user_id)
) AS yesterday_sum
FROM positions p
INNER JOIN users u ON u.id = p.user_id
AND p.profit_percent IS NOT NULL
AND p.parent_ticket IS NULL
AND p.close_at IS NOT NULL
AND p.open_at BETWEEN :dateFrom AND :dateTo
GROUP BY (u.id, u.nickname)
HAVING sum(p.profit_percent) / :workDays > 1
ORDER BY middle DESC;
How can I get rid of rows with yesterday_sum column less than 1 and NULL?
To use the column yesterday_sum in a WHERE clause you can produce the named column by placing the query as a subquery of the main one. Then, filtering is trivial.
For example, you can do:
select *
from (
SELECT u.id AS id,
u.nickname AS title,
sum(p.profit_percent) / :workDays AS middle,
(
SELECT sum(ps.profit_percent)
FROM positions ps
WHERE ps.user_id = u.id
AND ps.open_at BETWEEN
:dateYesterday AND
:dateYesterday + INTERVAL '1 day'
GROUP BY (ps.user_id)
) AS yesterday_sum
FROM positions p
INNER JOIN users u ON u.id = p.user_id
AND p.profit_percent IS NOT NULL
AND p.parent_ticket IS NULL
AND p.close_at IS NOT NULL
AND p.open_at BETWEEN :dateFrom AND :dateTo
GROUP BY (u.id, u.nickname)
HAVING sum(p.profit_percent) / :workDays > 1
) x
where yesterday_sum >= 1 -- yesterday_sum is available here
ORDER BY middle DESC;
select coalesce(ratings.positive,0) as positive,coalesce(ratings.negative,0) as negative,articles.id,x.username,commentnumb,
articles.category,
articles."createdAt",
articles.id,
articles.title,
articles."updatedAt"
FROM articles
LEFT JOIN (SELECT id AS userId,username,about FROM users) x ON articles.user_id = x.userId
LEFT JOIN (SELECT id,
article_id,
sum(case when rating = '1' then 1 else 0 end) as positive,
sum(case when rating = '0' then 1 else 0 end) as negative
from article_ratings
GROUP by id
) as ratings ON ratings.article_id = articles.id
LEFT JOIN (SELECT article_id,id,
count(article_id) as commentNumb
from article_comments
GROUP by id
) as comments ON comments.article_id = articles.id
WHERE articles."createdAt" <= :date
group by ratings.positive,ratings.negative,articles.id,x.username,commentnumb
order by articles."createdAt" desc
LIMIT 10
The code is working, however I have many more comments and many more ratings than what is counted in both SUM and COUNT functions.
How do I fix this query?
This is using postgres.
I've done some experimentation and it seems that the third join for comments is the one causing issues.
In the derived tables, you should ideally be grouping using article_id. But, you are grouping based on id. Due to this, you are getting more than the necessary rows in the derived tables. I have modified the query to suit your needs.
SELECT COALESCE(ratings.positive,0) AS positive,COALESCE(ratings.negative,0) AS negative,articles.id,x.username,commentnumb,
articles.category,
articles."createdAt",
articles.id,
articles.title,
articles."updatedAt"
FROM articles
LEFT OUTER JOIN (SELECT id AS userId,username,about FROM users) x ON articles.user_id = x.userId
LEFT OUTER JOIN (SELECT article_id,
SUM(case when rating = '1' then 1 else 0 end) as positive,
SUM(case when rating = '0' then 1 else 0 end) as negative
FROM article_ratings
GROUP by article_id
) AS ratings ON ratings.article_id = articles.id
LEFT OUTER JOIN (SELECT article_id,
count(article_id) as commentNumb
FROM article_comments
GROUP by article_id
) AS comments ON comments.article_id = articles.id
WHERE articles."createdAt" <= :date
ORDER BY articles."createdAt" desc
LIMIT 10;
I am trying to rank users on my system based on the user's totalArticleViews and the user's totalArticles on my system. The ranking should be based on the formula (totalArticleViews + ( totalArticles * 500 )) / 100
I have a system that allows users to post articles, a record is created every time any of these articles are read by anyone. My database has the following tables. users, articles, reads.
I have tried to get the views to insert into the formula, but i'm having issues getting all the users articles and multiplying it by 500 to insert into the formula to rank them all
with article_views AS (
SELECT article_id, COUNT(reads.id) AS views, 1 * 500 AS points
FROM reads
WHERE article_id IN (
SELECT id FROM articles WHERE articles.published_on IS NOT NULL AND
articles.deleted_at IS NULL
)
GROUP BY article_id
),
published AS (
SELECT COUNT(articles.id) AS TotalArticle, COUNT(articles.id) * 500 AS
points
FROM articles
WHERE published_on IS NOT NULL AND deleted_at IS NULL
GROUP BY articles.user_id
)
SELECT
users.id AS user_id,
ROUND((SUM(article_views.views) + () ) / 100.0, 2) AS points,
ROW_NUMBER() OVER (ORDER BY ROUND((SUM(article_views.views) + ()) /
100.0, 2) DESC)
FROM users
LEFT JOIN articles ON users.id = articles.user_id
LEFT JOIN reads ON articles.id = reads.article_id
LEFT JOIN article_views ON reads.article_id = article_views.article_id
WHERE
users.id IN (SELECT user_id FROM role_user WHERE role_id = 2)
AND status = 'ACTIVE'
GROUP BY users.id
ORDER BY points DESC NULLS LAST
I'm stuck at this point
(SUM(article_views.views) + () ) / 100.0, 2)
Simply use the published CTE by including the GROUP BY column user_id in SELECT and then joining published to users by this field in main level query.
WITH article_views AS (
SELECT r.article_id,
COUNT(r.id) AS views,
1 * 500 AS points
FROM reads r
WHERE r.article_id IN (
SELECT id
FROM articles a
WHERE a.published_on IS NOT NULL
AND a.deleted_at IS NULL
)
GROUP BY r.article_id
),
published AS (
SELECT a.user_id,
COUNT(a.id) AS TotalArticle,
COUNT(a.id) * 500 AS points
FROM articles a
WHERE a.published_on IS NOT NULL
AND a.deleted_at IS NULL
GROUP BY a.user_id
)
SELECT u.id AS user_id,
ROUND((SUM(av.views) + (p.TotalArticle)) / 100.0, 2) AS points,
ROW_NUMBER() OVER (ORDER BY ROUND((SUM(av.views) + (p.points))
/ 100.0, 2) DESC) AS rn
FROM users u
LEFT JOIN articles a ON u.id = a.user_id
LEFT JOIN reads r ON a.id = r.article_id
LEFT JOIN article_views av ON r.article_id = av.article_id
LEFT JOIN published p ON u.id = p.user_id
WHERE u.id IN (
SELECT user_id FROM role_user WHERE role_id = 2
)
AND u.status = 'ACTIVE'
GROUP BY u.id
ORDER BY points DESC NULLS LAST
Imagine I have a query like the following one:
SELECT
u.ID,
( SELECT
COUNT(*)
FROM
POSTS p
WHERE
p.USER_ID = u.ID
AND p.TYPE = 1
) AS interesting_posts,
( SELECT
COUNT(*)
FROM
POSTS p
WHERE
p.USER_ID = u.ID
AND p.TYPE = 2
) AS boring_posts,
( SELECT
COUNT(*)
FROM
COMMENTS c
WHERE
c.USER_ID = u.ID
AND c.TYPE = 1
) AS interesting_comments,
( SELECT
COUNT(*)
FROM
COMMENTS c
WHERE
c.USER_ID = u.ID
AND c.TYPE = 2
) AS boring_comments
FROM
USERS u;
( Hopefully it's correct because I just came up with it and didn't test it )
where I try to calculate the number of interesting and boring posts and comments that the user has.
Now, the problem with this query is that we have 2 sequential scans on both the posts and comments table and I wonder if there is a way to avoid that?
I could probably LEFT JOIN both posts and comments to the users table and do some aggregation but it's gonna generate a lot of rows before aggregation and I am not sure if that's a good way to go.
Aggregate posts and comments and outer join them to the users table.
select
u.id as user_id,
coaleasce(p.interesting, 0) as interesting_posts,
coaleasce(p.boring, 0) as boring_posts,
coaleasce(c.interesting, 0) as interesting_comments,
coaleasce(c.boring, 0) as boring_comments
from users u
left join
(
select
user_id,
count(case when type = 1 then 1 end) as interesting,
count(case when type = 2 then 1 end) as boring
from posts
group by user_id
) p on p.user_id = u.id
left join
(
select
user_id,
count(case when type = 1 then 1 end) as interesting,
count(case when type = 2 then 1 end) as boring
from comments
group by user_id
) c on c.user_id = u.id;
compare results and execution plan (here you scan posts once):
with c as (
select distinct
count(1) filter (where TYPE = 1) over (partition by USER_ID) interesting_posts
, count(1) filter (where TYPE = 2) over (partition by USER_ID) boring_posts
, USER_ID
)
, p as (select USER_ID,max(interesting_posts) interesting_posts, max(boring_posts) boring_posts from c)
SELECT
u.ID, interesting_posts,boring_posts
, ( SELECT
COUNT(*)
FROM
COMMENTS c
WHERE
c.USER_ID = u.ID
) AS comments
FROM
USERS u
JOIN p on p.USER_ID = u.ID
I'm generating a data set that looks like this
category user total
1 jonesa 0
2 jonesa 0
3 jonesa 0
1 smithb 0
2 smithb 0
3 smithb 5
1 brownc 2
2 brownc 3
3 brownc 4
Where a particular user has 0 records in all categories is it possible to remove their rows form the set? If a user has some activity like smithb does, I'd like to keep all of their records. Even the zeroes rows. Not sure how to go about that, I thought a CASE statement may be of some help but I'm not sure, this is pretty complicated for me. Here is my query
SELECT DISTINCT c.category,
u.user_name,
CASE WHEN (
SELECT COUNT(e.entry_id)
FROM category c1
INNER JOIN entry e1
ON c1.category_id = e1.category_id
WHERE c1.category_id = c.category_id
AND e.user_name = u.user_name
AND e1.entered_date >= TO_DATE ('20140625','YYYYMMDD')
AND e1.entered_date <= TO_DATE ('20140731', 'YYYYMMDD')) > 0 -- I know this won't work
THEN 'Yes'
ELSE NULL
END AS TOTAL
FROM user u
INNER JOIN role r
ON u.id = r.user_id
AND r.id IN (1,2),
category c
LEFT JOIN entry e
ON c.category_id = e.category_id
WHERE c.category_id NOT IN (19,20)
I realise the case statement won't work, but it was an attempt on how this might be possible. I'm really not sure if it's possible or the best direction. Appreciate any guidance.
Try this:
delete from t1
where user in (
select user
from t1
group by user
having count(distinct category) = sum(case when total=0 then 1 else 0 end) )
The sub query can get all the users fit your removal requirement.
count(distinct category) get how many category a user have.
sum(case when total=0 then 1 else 0 end) get how many rows with activities a user have.
There are a number of ways to do this, but the less verbose the SQL is, the harder it may be for you to follow along with the logic. For that reason, I think that using multiple Common Table Expressions will avoid the need to use redundant joins, while being the most readable.
-- assuming user_name and category_name are unique on [user] and [category] respectively.
WITH valid_categories (category_id, category_name) AS
(
-- get set of valid categories
SELECT c.category_id, c.category AS category_name
FROM category c
WHERE c.category_id NOT IN (19,20)
),
valid_users ([user_name]) AS
(
-- get set of users who belong to valid roles
SELECT u.[user_name]
FROM [user] u
WHERE EXISTS (
SELECT *
FROM [role] r
WHERE u.id = r.[user_id] AND r.id IN (1,2)
)
),
valid_entries (entry_id, [user_name], category_id, entry_count) AS
(
-- provides a flag of 1 for easier aggregation
SELECT e.[entry_id], e.[user_name], e.category_id, CAST( 1 AS INT) AS entry_count
FROM [entry] e
WHERE e.entered_date BETWEEN TO_DATE('20140625','YYYYMMDD') AND TO_DATE('20140731', 'YYYYMMDD')
-- determines if entry is within date range
),
user_categories ([user_name], category_id, category_name) AS
( SELECT u.[user_name], c.category_id, c.category_name
FROM valid_users u
-- get the cartesian product of users and categories
CROSS JOIN valid_categories c
-- get only users with a valid entry
WHERE EXISTS (
SELECT *
FROM valid_entries e
WHERE e.[user_name] = u.[user_name]
)
)
/*
You can use these for testing.
SELECT COUNT(*) AS valid_categories_count
FROM valid_categories
SELECT COUNT(*) AS valid_users_count
FROM valid_users
SELECT COUNT(*) AS valid_entries_count
FROM valid_entries
SELECT COUNT(*) AS users_with_entries_count
FROM valid_users u
WHERE EXISTS (
SELECT *
FROM user_categories uc
WHERE uc.user_name = u.user_name
)
SELECT COUNT(*) AS users_without_entries_count
FROM valid_users u
WHERE NOT EXISTS (
SELECT *
FROM user_categories uc
WHERE uc.user_name = u.user_name
)
SELECT uc.[user_name], uc.[category_name], e.[entry_count]
FROM user_categories uc
INNER JOIN valid_entries e ON (uc.[user_name] = e.[user_name] AND uc.[category_id] = e.[category_id])
*/
-- Finally, the results:
SELECT uc.[user_name], uc.[category_name], SUM(NVL(e.[entry_count],0)) AS [entry_count]
FROM user_categories uc
LEFT OUTER JOIN valid_entries e ON (uc.[user_name] = e.[user_name] AND uc.[category_id] = e.[category_id])
Here's another method:
WITH totals AS (
SELECT
c.category,
u.user_name,
COUNT(e.entry_id) AS total,
SUM(COUNT(e.entry_id)) OVER (PARTITION BY u.user_name) AS user_total
FROM
user u
INNER JOIN
role r ON u.id = r.user_id
CROSS JOIN
category c
LEFT JOIN
entry e ON c.category_id = e.category_id
AND u.user_name = e.user_name
AND e1.entered_date >= TO_DATE ('20140625', 'YYYYMMDD')
AND e1.entered_date <= TO_DATE ('20140731', 'YYYYMMDD')
WHERE
r.id IN (1, 2)
AND c.category_id IN (19, 20)
GROUP BY
c.category,
u.user_name
)
SELECT
category,
user_name,
total
FROM
totals
WHERE
user_total > 0
;
The totals derived table calculates the totals per user and category as well as totals across all categories per user (using SUM() OVER ...). The main query returns only rows where the user total is greater than zero.