MySQL loop and multiple LEFT joins - mysql

I got the following code:
SELECT
COALESCE(rv.views, 0) as views
FROM
( select 0 as n
union all select 1
union all select 2
union all select 3 ) n
LEFT JOIN restaurant_views rv
on rv.date = date_add("2015-02-24", interval - n.n day)
and restaurant_id = 192
This code is giving me the amount of views a restaurant had the last 4 days.
I am looking for a similar query to get the amount of likes a restaurant had the last 4 days.
This is what I got so far:
SELECT
( COUNT( DISTINCT a.restaurant_id)
+ COUNT( DISTINCT d.restaurant_id)) as num_likes
FROM
( select 0 as n
union all select 1
union all select 2
union all select 3 ) n
LEFT JOIN apple_likes a
on a.vote_date = date_add("2015-02-24", interval - n.n day)
and a.restaurant_id = 192
LEFT JOIN android_likes d
on d.vote_date = date_add("2015-02-24", interval - n.n day)
and d.restaurant_id = 192
And here is the output, which is as you can see not what I'm looking for:
What do I have to change to get the number of likes in the last query?
(I have checked that the restaurant has likes on all days, so I am positive it's something wrong with the query)

Try this one:
SELECT
( a.likes)
+ d.likes) as num_likes
FROM
( select 0 as n
union all select 1
union all select 2
union all select 3 ) n
LEFT JOIN (
SELECT vote_date,COUNT(*) as likes
FROM apple_likes
WHERE restaurant_id = 192
GROUP BY restaurant_id, vote_date
) as a
on a.vote_date = date_add("2015-02-24", interval - n.n day)
LEFT JOIN (
SELECT vote_date, COUNT(*) as likes
FROM android_likes
WHERE restaurant_id = 192
GROUP BY restaurant_id, vote_date
) as d
on d.vote_date = date_add("2015-02-24", interval - n.n day)

I can think of a couple items that might be what you are encountering...
Just because somebody VIEWS a restaurant, does that mean they actually VOTED??? And if Voted, are the only two devices that of apple or android? What if viewing from a browser and they are on a Windows machine browser-based?
Date Equality. In the restaurant views table, is the date field ALWAYS that of a time = 12:00:00 (ie: midnight/morning of the day). If the time-stamps of the votes are anything other than 12:00:00, and you are trying to compare for a date = date + time is probably failing. What you may need is a comparison of the date( vote_date ) = date( date_add( ... )) so this way BOTH are ignoring the time component... Now, that being said, a function on a date column is not going to be optimized, even if the restaurant ID is numeric and part of the index key... it would be PARTIALLY optimized. You may want to just add a generic date of AND vote_date >= '2015-02-20' so it can optimize the restaurant and date, then apply the DATE( vote_date ) for the actual qualfying of records.

Related

Getting all previous records of table by date MySQL

My table currently has 21000 records, it's daily updated and almost 300 entries are inserted. Now, what I want is to have a query which will fetch the counts of elements that my table had for the previous 10 days, so it returns:
26000
21300
21000
etc
Right now, I wrote this:
"SELECT COUNT(*) from tbl_task where `task_start_time` < '2020-12-01'"
And it returns 21000 but only for 1 day. I want by query to return records according to 10 days.
However, this does it for only 1 day.
edit : database flavor is mysql and date column is date not datetime
The most efficient method may be aggregation and cumulative sums:
select date(task_start_time) as dte, count(*) as cnt_on_day,
sum(count(*)) over (order by date(task_start_time)) as running_cnt
from tbl_task
group by dte
order by dte desc
limit 10;
This returns the last 10 days in the data. You can easily adjust to more days if you like -- in fact all of them -- without much trouble.
I don't know if I'm wrong, but could you not simple add a GROUP BY - statement? Like:
"SELECT COUNT(*) from tbl_task where `task_start_time` < '2020-12-01' GROUP
BY task_start_time"
EDIT:
This should only work if task_start_time is a date, not if it is a datetime
EDIT2:
If it is a datetime you could use the date function:
SELECT COUNT(*) from tbl_task where `task_start_time` < '2020-12-01' GROUP
BY DATE(task_start_time)
You can use UNION ALL and date arithmetic.
SELECT count(*)
FROM tbl_task
WHERE task_start_time < current_date
UNION ALL
SELECT count(*)
FROM tbl_task
WHERE task_start_time < date_sub(current_date, INTERVAL 1 DAY)
...
UNION ALL
SELECT count(*)
FROM tbl_task
WHERE task_start_time < date_sub(current_date, INTERVAL 9 DAY);
Edit:
You might also join a derived table that uses FROM-less SELECTs and UNION ALL to get the days to look back and then aggregate. This might be a little easier to construct dynamically. (But it may be slower I suspect.)
SELECT count(*)
FROM (SELECT 0 x
UNION ALL
SELECT 1
...
UNION ALL
SELECT 9)
INNER JOIN tbl_task t
ON t.task_start_time < date_sub(current_date, INTERVAL x.x DAY)
GROUP BY x.x;
In MySQL version 8+ you can even use a recursive CTE to construct the table with the days.
WITH RECURSIVE x
AS
(
SELECT 0 x
UNION ALL
SELECT x + 1
FROM x
WHERE x + 1 < 10
)
SELECT count(*)
FROM x
INNER JOIN tbl_task t
ON t.task_start_time < date_sub(current_date, INTERVAL x.x DAY)
GROUP BY x.x;

MYSQL SUM until last day of Each month for last 12 months

I have a table like this two
Table A
date amount B_id
'2020-1-01' 3000000 1
'2019-8-01' 15012 1
'2019-6-21' 90909 1
'2020-1-15' 84562 1
--------
Table B
id type
1 7
2 5
I have to show sum of amount until the last date of each month for the last 12 month.
The query i have prepared is like this..
SELECT num2.last_dates,
(SELECT SUM(amount) FROM A
INNER JOIN B ON A.B_id = B.id
WHERE B.type = 7 AND A.date<=num2.last_dates
),
(SELECT SUM(amount) FROM A
INNER JOIN B ON A.B_id = B.id
WHERE B.type = 5 AND A.date<=num2.last_dates)
FROM
(SELECT last_dates
FROM (
SELECT LAST_DAY(CURDATE() - INTERVAL CUSTOM_MONTH MONTH) last_dates
FROM(
SELECT 1 CUSTOM_MONTH UNION
SELECT 0 UNION
SELECT 2 UNION
SELECT 3 UNION
SELECT 4 UNION
SELECT 5 UNION
SELECT 6 UNION
SELECT 7 UNION
SELECT 8 UNION
SELECT 9 UNION
SELECT 10 UNION
SELECT 11 UNION
SELECT 12 )num
) num1
)num2
ORDER BY num2.last_dates
This gives me the result like this which is exactly how i need it. I need this query to execute faster. Is there any better way to do what i am trying to do?
2019-05-31 33488.69 109.127800
2019-06-30 263.690 1248932.227800
2019-07-31 274.690 131.827800
2019-08-31 627.690 13.687800
2019-09-30 1533.370000 08.347800
2019-10-31 1444.370000 01.327800
2019-11-30 5448.370000 247.227800
2019-12-31 61971.370000 016.990450
2020-01-31 19550.370000 2535.185450
2020-02-29 986.370000 405.123300
2020-03-31 1152.370000 26.793300
2020-04-30 9404.370000 11894.683300
2020-05-31 3404.370000 17894.683300
I'd use conditional aggregation, and pre-aggregate the monthly totals in one pass, instead of doing twenty-six individual passes repeatedly through the same data.
I'd start with something like this:
SELECT CASE WHEN A.date < DATE(NOW()) + INTERVAL -14 MONTH
THEN LAST_DAY( DATE(NOW()) + INTERVAL -14 MONTH )
ELSE LAST_DAY( A.date )
END AS _month_end
, SUM(IF( B.type = 5 , B.amount , NULL)) AS tot_type_5
, SUM(IF( B.type = 7 , B.amount , NULL)) AS tot_type_7
FROM A
JOIN B
ON B.id = A.B_id
WHERE B.type IN (5,7)
GROUP
BY _month_end
(column amount isn't qualified in original query, so just guessing here which table that is from. adjust as necessary. best practice is to qualify all column references.
That gets us the subtotals for each month, in a single pass through A and B.
We can get that query tested and tuned.
Then we can incorporate that as an inline view in an outer query which adds up those monthly totals. (I'd do an outer join, just in case rows are missing, sow we don't wind up omitting rows.)
Something like this:
SELECT d.dt + INTERVAL -i.n MONTH + INTERVAL -1 DAY AS last_date
, SUM(IFNULL(t.tot_type_5,0)) AS rt_type_5
, SUM(IFNULL(t.tot_type_7,0)) AS rt_type_7
FROM ( -- first day of next month
SELECT DATE(NOW()) + INTERVAL -DAY(DATE(NOW()))+1 DAY + INTERVAL 1 MONTH AS dt
) d
CROSS
JOIN ( -- thirteen integers, integers 0 thru 12
SELECT 0 AS n
UNION ALL SELECT 1 UNION ALL SELECT 2 UNION ALL SELECT 3 UNION ALL SELECT 4
UNION ALL SELECT 5 UNION ALL SELECT 6 UNION ALL SELECT 7 UNION ALL SELECT 8
UNION ALL SELECT 9 UNION ALL SELECT 10 UNION ALL SELECT 11 UNION ALL SELECT 12
) i
LEFT
JOIN ( -- totals by month
SELECT CASE WHEN A.date < DATE(NOW()) + INTERVAL -14 MONTH
THEN LAST_DAY( DATE(NOW()) + INTERVAL -14 MONTH )
ELSE LAST_DAY( A.date )
END AS _month_end
, SUM(IF( B.type = 5 , B.amount , NULL)) AS tot_type_5
, SUM(IF( B.type = 7 , B.amount , NULL)) AS tot_type_7
FROM A
JOIN B
ON B.id = A.B_id
WHERE B.type IN (5,7)
GROUP
BY _month_end
) t
ON t._month_end < d.dt
GROUP BY d.dt + INTERVAL -i.n MONTH + INTERVAL -1 DAY
ORDER BY d.dt + INTERVAL -i.n MONTH + INTERVAL -1 DAY DESC
The design is meant to do one swoop through the A JOIN B set. We're expecting to get about 14 rows back. And we're doing a semi-join, duplicating the oldest months multiple times, so approx . 14 x 13 / 2 = 91 rows, that get collapsed into 13 rows.
The big rock in terms of performance is going to be materializing that inline view query.
This is how I'd probably approach this in MySQL 8 with SUM OVER:
Get the last 12 months.
Use these months to add empty month rows to the original data, as MySQL doesn't support full outer joins.
Get the running totals for all months.
Show only the last twelve months.
The query:
with months (date) as
(
select last_day(current_date - interval 1 month) union all
select last_day(current_date - interval 2 month) union all
select last_day(current_date - interval 3 month) union all
select last_day(current_date - interval 4 month) union all
select last_day(current_date - interval 5 month) union all
select last_day(current_date - interval 6 month) union all
select last_day(current_date - interval 7 month) union all
select last_day(current_date - interval 8 month) union all
select last_day(current_date - interval 9 month) union all
select last_day(current_date - interval 10 month) union all
select last_day(current_date - interval 11 month) union all
select last_day(current_date - interval 12 month)
)
, data (date, amount, type) as
(
select last_day(a.date), a.amount, b.type
from a
join b on b.id = a.b_id
where b.type in (5, 7)
union all
select date, null, null from months
)
select
date,
sum(sum(case when type = 5 then amount end)) over (order by date) as t5,
sum(sum(case when type = 7 then amount end)) over (order by date) as t7
from data
group by date
order by date
limit 12;
Demo: https://dbfiddle.uk/?rdbms=mysql_8.0&fiddle=ddeb3ab3e086bfc182f0503615fba74b
I don't know whether this is faster than your own query or not. Just give it a try. (You'd get my query much faster by adding a generated column for last_day(date) to your table and use this. If you need this often, this may be an option.)
You are getting some complicated answers. I think it is easier. Start with knowing we can easily sum for each month:
SELECT SUM(amount) as monthtotal,
type,
MONTH(date) as month,
YEAR(date) as year
FROM A LEFT JOIN B on A.B_id=B.id
GROUP BY type,month,year
From that data, we can use a variable to get running total. Best to do by initializing the variable, but not necessary. We can get the data necessary like this
SET #running := 0;
SELECT (#running := #running + monthtotal) as running, type, LAST_DAY(CONCAT(year,'-',month,'-',1))
FROM
(SELECT SUM(amount) as monthtotal,type,MONTH(date) as month,YEAR(date) as year FROM A LEFT JOIN B on A.B_id=B.id GROUP BY type,month,year) AS totals
ORDER BY year,month
You really need to have a connector that supports multiple statements, or make multiple calls to initialize the variable. Although you can null check the variable and default to 0, you still have an issue if you run the query a second time.
Last thing, if you really want the types to be summed separately:
SET #running5 := 0;
SET #running7 := 0;
SELECT
LAST_DAY(CONCAT(year,'-',month,'-',1)),
(#running5 := #running5 + (CASE WHEN type=5 THEN monthtotal ELSE 0 END)) as running5,
(#running7 := #running7 + (CASE WHEN type=7 THEN monthtotal ELSE 0 END)) as running7
FROM
(SELECT SUM(amount) as monthtotal,type,MONTH(date) as month,YEAR(date) as year FROM A LEFT JOIN B on A.B_id=B.id GROUP BY type,month,year) AS totals
ORDER BY year,month
We still don't show months where there is no data. I'm not sure that is a requirement. But this should only need one pass of table A.
Also, make sure the id on table B is indexed.

How ot return 0 instead of null on mysql query?

The following query returns the visitors and pageviews of last 7 days. However, if there are no results (let's say it is a fresh account), nothing is returned.
How to edit this in order to return 0 in days that there are no entries?
SELECT Date(timestamp) AS day,
Count(DISTINCT hash) AS visitors,
Count(*) AS pageviews
FROM behaviour
WHERE company_id = 1
AND timestamp >= Subdate(Curdate(), 7)
GROUP BY day
Assuming that you always have at least one record in the table for each of the last 7 days (regardless of the company_id), then you can use conditional aggregation as follows:
select
date(timestamp) as day,
count(distinct case when company_id = 1 then hash end) as visitors,
sum(company_id = 1) as pageviews
from behaviour
where timestamp >= curdate() - interval 7 day
group by day
Note that I changed you query to use standard date arithmetics, which I find easier to understand that date functions.
Otherwise, you would need to move the condition on the date from the where clause to the aggregate functions:
select
date(timestamp) as day,
count(distinct case when timestamp >= curdate() - interval 7 day and company_id = 1 then hash end) as visitors,
sum(timestamp >= curdate() - interval 7 day and company_id = 1) as pageviews
from behaviour
group by day
If your table is big, this can be expensive so I would not recommend that.
Alternatively, you can generate a derived table of dates and left join it with your original query:
select
curdate - interval x.n day day,
count(distinct b.hash) visitors,
count(b.hash) page_views
from (
select 1 n union all select 2 union all select 3 union all select 4
union all select 5 union all select 6 union all select 7
) x
left join behavior b
on b.company_id = 1
and b.timestamp >= curdate() - interval x.n day
and b.timestamp < curdate() - interval (x.n - 1) day
group by x.n
Use a query that returns all the dates from today minus 7 days to today and left join the table behaviour:
SELECT t.timestamp AS day,
Count(DISTINCT b.hash) AS visitors,
Count(b.timestamp) AS pageviews
FROM (
SELECT Subdate(Curdate(), 7) timestamp UNION ALL SELECT Subdate(Curdate(), 6) UNION ALL
SELECT Subdate(Curdate(), 5) UNION ALL SELECT Subdate(Curdate(), 4) UNION ALL SELECT Subdate(Curdate(), 3) UNION ALL
SELECT Subdate(Curdate(), 2) UNION ALL SELECT Subdate(Curdate(), 1) UNION ALL SELECT Curdate()
) t LEFT JOIN behaviour b
ON Date(b.timestamp) = t.timestamp AND b.company_id = 1
GROUP BY day
Use IFNULL:
IFNULL(expr1, 0)
From the documentation:
If expr1 is not NULL, IFNULL() returns expr1; otherwise it returns expr2. IFNULL() returns >a numeric or string value, depending on the context in which it is used.
You can use next trick:
First, get query that return 1 dummy row: SELECT 1;
Next use LEFT JOIN to connect summary row(s) without condition. This join will return values in case data exists on NULL values in other case.
Last select from joined queries onle what we need and convert NULL's to ZERO's
using IFNULL dunction.
SELECT
IFNULL(b.day,0) AS DAY,
IFNULL(b.visitors,0) AS visitors,
IFNULL(b.pageviews,0) AS pageviews
FROM (
SELECT 1
) a
LEFT JOIN (
SELECT DATE(TIMESTAMP) AS DAY,
COUNT(DISTINCT HASH) AS visitors,
COUNT(*) AS pageviews
FROM behaviour
WHERE company_id = 1
AND TIMESTAMP >= SUBDATE(CURDATE(), 7)
GROUP BY DAY
) b ON 1 = 1;

How to query a hotel database to return the query for a single room available for three consecutive nights?

I'm trying to find an answer to the following query:
A customer wants a single room for three consecutive nights. Find the first available date in December 2016.
As per the question, this should be the right answer. But I don't know how to solve it.
+-----+------------+
| id | MIN(i) |
+-----+------------+
| 201 | 2016-12-11 |
+-----+------------+
The link is from question number 14 here.
This is the ER diagram of the database:
I apologize that I'm a bit rusty with this kind of query and I can't guarantee that I got all of the syntax correct, but I think that something like the following might work:
SELECT id, DATE_ADD(b.booking_date, INTERVAL (end_date + 1 DAY) as date
FROM (
SELECT r.id, STR_TO_DATE('2016-01-01', '%Y-%m-%d') as start_of_month, b.booking_date as start_date, DATE_ADD(b.booking_date, INTERVAL (nights - 1) DAY) as end_date
FROM room r
LEFT JOIN booking b ON r.id = b.room_no
ORDER BY r.id, b.booking_date
) as room_bookings
WHERE DATE_DIFF(room_bookings.start_of_month, room_bookings.start_date) >= 3
OR DATE_DIFF(room_bookings.end_date, (
SELECT b2.booking_date FROM booking b2
WHERE b2.room_no = room_bookings.id AND b2.booking_date > room_bookings.start_date
ORDER BY b2.booking_date LIMIT 1)
) >= 3
In fact, now that I type that all out, you might be able to tweak the WHERE of the main query so that you don't even need the room_bookings subselect. Hopefully this helps and isn't too far off the mark.
This seems very hard to do without a calendar table -- because an appropriate room might have no booking at all during the month. Without any booking, there is no record in the month to start with.
select r.id, dte
from rooms r cross join
(select date('2018-12-01') as dte union all
select date('2018-12-02') as dte union all
. . .
select date('2018-12-32') as dte
) d
where not exists (select 1 from bookings b where b.room_no = r.id and b.booking_date = d.dte) and
not exists (select 1 from bookings b where b.room_no = r.id and b.booking_date = d.dte + interval 1 day) and
not exists (select 1 from bookings b where b.room_no = r.id and b.booking_date = d.dte + interval 2 day)
order by d.dte
limit 1;
This assumes that booking_date is the start of the stay. You need to provide the logic for a "single room".
select distinct top 1 alll.i,alll.room_no,
case
when (select count(*) from booking where room_no = alll.room_no and booking_date between dateadd(day,1,alll.i) and dateadd(day,3,alll.i)) > 0 then 'Y'
else 'N'
end as av3
from
(select c.i,b.room_no,b.booking_date
from calendar c cross join booking b
where month(c.i) = 12 and year(c.i) = 2016 and b.room_type_requested = 'single'
) as alll
join
(
select distinct c.i, b.room_no
from calendar c join booking b
on c.i between b.booking_date and DATEADD(day,b.nights-1,b.booking_date)
where month(c.i) = 12 and year(c.i) = 2016 and b.room_type_requested = 'single'
) as booked
on alll.i = booked.i
and alll.room_no <> booked.room_no
order by 1
This works. It is a little complicated but basically first checks all the rooms that are booked and then does a comparison between rooms not booked on each day of the month till the next 3 days.
My solution is separate problem into 2 parts (in the end was 2 queries joined together). May not be the most efficient but the solution is correct.
1) Of the single rooms, look at the last check-out date, and see which one is vacant first (i.e. no more bookings for the rest of the month)
2) check in between current reservations - and see if there's a 3 day gap between them
3) join those together - grab the min
WITH subquery AS( -- existing single-bed bookings in Dec
SELECT room_no, booking_date,
DATE_ADD(booking_date, INTERVAL (nights-1) DAY) AS last_night
FROM booking
WHERE room_type_requested='single' AND
DATE_ADD(booking_date, INTERVAL (nights-1) DAY)>='2016-12-1' AND
booking_date <='2016-12-31'
ORDER BY room_no, last_night)
SELECT room_no, MIN(first_avail) AS first_avail --3) join the 2 together
FROM(
-- 1) check the last date the room is booked in December (available after)
SELECT room_no, MIN(first_avail) AS first_avail
FROM(
SELECT room_no, DATE_ADD(MAX(last_night), INTERVAL 1 DAY) AS first_avail
FROM subquery q3
GROUP BY 1
ORDER BY 2) AS t2
UNION
-- 2) check if any 3-day exist in between reservations
SELECT room_no, DATE_ADD(MIN(end2), INTERVAL 1 DAY) AS first_avail
FROM(
SELECT q1.booking_date AS beg1, q1.room_no, q1.last_night AS end1,
q2.booking_date AS beg2, q2.last_night AS end2
FROM subquery q1
JOIN subquery q2
ON q1.room_no = q2.room_no AND q2.booking_date > q1.last_night
GROUP BY 2,1
ORDER BY 2,1) AS t
WHERE beg2-end1 > 3) AS inner_t
This works conceptually as the first avaiable date should always be the end of the previous booking.
SELECT MIN(DATE_ADD(a.booking_date, INTERVAL nights DAY)) AS i
FROM booking AS a
WHERE DATE_ADD(a.booking_date, INTERVAL nights DAY)
>= '2016-12-01'
AND room_type_requested = 'single'
AND NOT EXISTS
(SELECT 1 FROM booking AS b
WHERE b.booking_date BETWEEN
DATE_ADD(a.booking_date, INTERVAL nights DAY)
AND DATE_ADD(a.booking_date, INTERVAL nights+2 DAY)
AND a.room_no = b.room_no)

Return a zero for a day with no results

I have a query which returns the total of users who registered for each day. Problem is if a day had no one register it doesn't return any value, it just skips it. I would rather it returned zero
this is my query so far
SELECT count(*) total FROM users WHERE created_at < NOW() AND created_at >
DATE_SUB(NOW(), INTERVAL 7 DAY) AND owner_id = ? GROUP BY DAY(created_at)
ORDER BY created_at DESC
Edit
i grouped the data so i would get a count for each day- As for the date range, i wanted the total users registered for the previous seven days
A variation on the theme "build your on 7 day calendar inline":
SELECT D, count(created_at) AS total FROM
(SELECT DATE_SUB(NOW(), INTERVAL D DAY) AS D
FROM
(SELECT 0 as D
UNION SELECT 1
UNION SELECT 2
UNION SELECT 3
UNION SELECT 4
UNION SELECT 5
UNION SELECT 6
) AS D
) AS D
LEFT JOIN users ON date(created_at) = date(D)
WHERE owner_id = ? or owner_id is null
GROUP BY D
ORDER BY D DESC
I don't have your table structure at hand, so that would need adjustment probably. In the same order of idea, you will see I use NOW() as a reference date. But that's easily adjustable. Anyway that's the spirit...
See for a live demo http://sqlfiddle.com/#!2/ab5cf/11
If you had a table that held all of your days you could do a left join from there to your users table.
SELECT SUM(CASE WHEN U.Id IS NOT NULL THEN 1 ELSE 0 END)
FROM DimDate D
LEFT JOIN Users U ON CONVERT(DATE,U.Created_at) = D.DateValue
WHERE YourCriteria
GROUP BY YourGroupBy
The tricky bit is that you group by the date field in your data, which might have 'holes' in it, and thus miss records for that date.
A way to solve it is by filling a table with all dates for the past 10 and next 100 years or so, and to (outer)join that to your data. Then you will have one record for each day (or week or whatever) for sure.
I had to do this only for MS SqlServer, so how to fill a date table (or perhaps you can do it dynamically) is for someone else to answer.
A bit long winded, but I think this will work...
SELECT count(users.created_at) total FROM
(SELECT DATE_SUB(CURDATE(),INTERVAL 6 DAY) as cdate UNION ALL
SELECT DATE_SUB(CURDATE(),INTERVAL 5 DAY) UNION ALL
SELECT DATE_SUB(CURDATE(),INTERVAL 4 DAY) UNION ALL
SELECT DATE_SUB(CURDATE(),INTERVAL 3 DAY) UNION ALL
SELECT DATE_SUB(CURDATE(),INTERVAL 2 DAY) UNION ALL
SELECT DATE_SUB(CURDATE(),INTERVAL 1 DAY) UNION ALL
SELECT CURDATE()) t1 left join users
ON date(created_at)=t1.cdate
WHERE owner_id = ? or owner_id is null
GROUP BY t1.cdate
ORDER BY t1.cdate DESC
It differs from your query slightly in that it works on dates rather than date times which your query is doing. From your description I have assumed you mean to use whole days and therefore have used dates.