Where is the syntax error in this WITH clause? - mysql

I'm trying out this problem on LeetCode. The interface is throwing a syntax error, but for the life of me I can't figure it out. Here's the question:
Table: Stadium
+---------------+---------+
| Column Name | Type |
+---------------+---------+
| id | int |
| visit_date | date |
| people | int |
+---------------+---------+
visit_date is the primary key for this table. Each row of this table
contains the visit date and visit id to the stadium with the number of
people during the visit. No two rows will have the same visit_date,
and as the id increases, the dates increase as well.
Write an SQL query to display the records with three or more rows with
consecutive id's, and the number of people is greater than or equal to
100 for each.
Return the result table ordered by visit_date in ascending order.
The query result format is in the following example.
Stadium table:
+------+------------+-----------+
| id | visit_date | people |
+------+------------+-----------+
| 1 | 2017-01-01 | 10 |
| 2 | 2017-01-02 | 109 |
| 3 | 2017-01-03 | 150 |
| 4 | 2017-01-04 | 99 |
| 5 | 2017-01-05 | 145 |
| 6 | 2017-01-06 | 1455 |
| 7 | 2017-01-07 | 199 |
| 8 | 2017-01-09 | 188 |
+------+------------+-----------+
Result table:
+------+------------+-----------+
| id | visit_date | people |
+------+------------+-----------+
| 5 | 2017-01-05 | 145 |
| 6 | 2017-01-06 | 1455 |
| 7 | 2017-01-07 | 199 |
| 8 | 2017-01-09 | 188 |
+------+------------+-----------+
The four rows with ids 5, 6, 7, and 8 have consecutive ids and each of
them has >= 100 people attended. Note that row 8 was included even
though the visit_date was not the next day after row 7. The rows with
ids 2 and 3 are not included because we need at least three
consecutive ids.
Here's a fiddle with the data (NOTE: I'm still trying to figure out how to insert dates, so I saved the dates as strings instead).
Can anyone spot the syntax error in the query below?
# Write your MySQL query statement below
SET #rowIndex := 0;
WITH s1 as (
SELECT #rowIndex := #rowIndex + 1 as rowIndex, s.*
FROM Stadium as s
WHERE s.people >= 100
GROUP BY s.id
)
SELECT s2.id, s2.visit_date, s2.people
FROM s1 as s2
GROUP BY s2.rowIndex - s2.id, s2.id, s2.visit_date, s2.people
ORDER BY s2.visit_date
Error message:
You have an error in your SQL syntax; check the manual that
corresponds to your MySQL server version for the right syntax to use
near 'WITH s1 (rowIndex, id, visit_date, people) as (
SELECT #rowIndex := #rowInde' at line 4
Also, the LeetCode interface uses MySQL v8.0, so I don't think that's the problem.
I was using the query below as a reference. (Original.)
SET #rowIndex := -1;
SELECT ROUND(AVG(t.LAT_N), 4) FROM
(
SELECT #rowIndex := #rowIndex+1 AS rowIndex, s.LAT_N FROM STATION AS s ORDER BY s.LAT_N
) AS t
WHERE t.rowIndex IN (FLOOR(#rowIndex / 2), CEIL(#rowIndex / 2));
Thanks.
Edit:
For future reference, here's the final query I came up with:
# Write your MySQL query statement below
WITH s1 as (
SELECT ROW_NUMBER() OVER (ORDER BY s.id) as rowIndex, s.*
FROM Stadium as s
WHERE s.people >= 100
GROUP BY s.id, s.visit_date, s.people
), s2 as (
SELECT COUNT(s.id) OVER (PARTITION BY s.id-s.rowIndex) as groupSize, s.*
FROM s1 as s
)
SELECT s3.id, s3.visit_date, s3.people
FROM s2 as s3
GROUP BY s3.groupSize, s3.id, s3.visit_date, s3.people
HAVING s3.groupSize >= 3
ORDER BY s3.visit_date

You confirmed that you are using MySQL 8.0.21 server so the only other suggestion I have is that you're trying to run two SQL statements in one call:
SET #rowIndex := 0;
WITH s1 as (
SELECT...
Most MySQL connectors do not support multi-query by default. In other words, you can only do one statement per call. As soon as MySQL sees any syntax following your first ; it treats this as a syntax error.
There's no reason you need to use multi-query. Just run the two statements separately. As long as you use the same session, your value of #rowIndex will be available to subsequent statements.
The former Director of Engineering for MySQL once told me, "there's no reason multi-query should exist."

Not sure what problem you might be facing, yet this would pass on LeetCode:
Not sure if that's a right way to do it:
SELECT DISTINCT S1.id,
S1.visit_date,
S1.people
FROM stadium AS S1,
stadium AS S2,
stadium AS S3
WHERE S1.people > 99
AND S2.people > 99
AND S3.people > 99
AND ( (S2.id = S1.id + 1
AND S3.id = S1.id + 2)
OR (S2.id = S1.id - 1
AND S3.id = S1.id + 1)
OR (S2.id = S1.id - 1
AND S3.id = S1.id - 2) )
ORDER BY id ASC;

Related

combine 3 queries to one, SELECT / COUNT / INSERT

I need help to optimize my 3 queries into one.
I have 2 tables, the first has a list of image processing servers I use, so different servers can handle different simultaneous job loads at a time, so I have a field called quota as seen below.
First table name, "img_processing_servers"
| id | server_url | server_key | server_quota |
| 1 | examp.uu.co | X0X1X2XX3X | 5 |
| 2 | examp2.uu.co| X0X1X2YX3X | 3 |
The second table registers if there is a job being performed at this moment on the server
Second table, "img_servers_lock"
| id | lock_server | timestamp |
| 1 | 1 | 2020-04-30 12:08:09 |
| 2 | 1 | 2020-04-30 12:08:09 |
| 3 | 1 | 2020-04-30 12:08:09 |
| 4 | 2 | 2020-04-30 12:08:09 |
| 5 | 2 | 2020-04-30 12:08:09 |
| 6 | 2 | 2020-04-30 12:08:09 |
Basically what I want to achieve is that my image servers don't go past the max quota and crash, so the 3 queries I would like to combine are:
Select at least one server available that hasn't reached it's quota and then insert a lock record for it.
SELECT * FROM `img_processing_servers` WHERE
SELECT COUNT(timestamp) FROM `img_servers_lock` WHERE `lock_server` = id
! if the count is < than quota, go ahead and register use
INSERT INTO `img_servers_lock`(`lock_server`, `timestamp`) VALUES (id_of_available_server, now())
How would I go about creating this single query?
My goal is to keep my image servers safe from overload.
Join the two tables and put that into an INSERT query.
INSERT INTO img_servers_lock(lock_server, timestamp)
SELECT s.id, NOW()
FROM img_processing_servers s
LEFT JOIN img_servers_lock l ON l.lock_server = s.id
GROUP BY s.id
HAVING IFNULL(COUNT(l.id), 0) < s.server_quota
ORDER BY s.server_quota - IFNULL(COUNT(l.id), 0) DESC
LIMIT 1
The ORDER BY clause makes it select the server with the most available quota.
OK, so I encountered just a small addition that was giving me a bug and it was that the s.server_quota had to be added to GROUP BY for it to work in the HAVING
INSERT INTO img_servers_lock(lock_server, timestamp)
SELECT s.id, NOW()
FROM alpr_servers s
LEFT JOIN img_servers_lock l ON l.lock_server = s.id
GROUP BY s.id, s.server_quota
HAVING IFNULL(COUNT(l.id), 0) < s.server_quota
ORDER BY s.server_quota - IFNULL(COUNT(l.id), 0) DESC
LIMIT 1
Thanks again Barmar!

Optimizing SQL Query for max value with various conditions from a single MySQL table

I have the following SQL query
SELECT *
FROM `sensor_data` AS `sd1`
WHERE (sd1.timestamp BETWEEN '2017-05-13 00:00:00'
AND '2017-05-14 00:00:00')
AND (`id` =
(
SELECT `id`
FROM `sensor_data` AS `sd2`
WHERE sd1.mid = sd2.mid
AND sd1.sid = sd2.sid
ORDER BY `value` DESC, `id` DESC
LIMIT 1)
)
Background:
I've checked the validity of the query by changing LIMIT 1 to LIMIT 0, and the query works without any problem. However with LIMIT 1 the query doesn't complete, it just states loading until I shutdown and restart.
Breaking the Query down:
I have broken down the query with the date boundary as follows:
SELECT *
FROM `sensor_data` AS `sd1`
WHERE (sd1.timestamp BETWEEN '2017-05-13 00:00:00'
AND '2017-05-14 00:00:00')
This takes about 0.24 seconds to return the query with 8200 rows each having 5 columns.
Question:
I suspect the second half of my Query, is not correct or well optimized.
The tables are as follows:
Current Table:
+------+-------+-------+-----+-----------------------+
| id | mid | sid | v | timestamp |
+------+-------+-------+-----+-----------------------+
| 51 | 10 | 1 | 40 | 2015-05-13 11:56:01 |
| 52 | 10 | 2 | 39 | 2015-05-13 11:56:25 |
| 53 | 10 | 2 | 40 | 2015-05-13 11:56:42 |
| 54 | 10 | 2 | 40 | 2015-05-13 11:56:45 |
| 55 | 10 | 2 | 40 | 2015-05-13 11:57:01 |
| 56 | 11 | 1 | 50 | 2015-05-13 11:57:52 |
| 57 | 11 | 2 | 18 | 2015-05-13 11:58:41 |
| 58 | 11 | 2 | 19 | 2015-05-13 11:58:59 |
| 59 | 11 | 3 | 58 | 2015-05-13 11:59:01 |
| 60 | 11 | 3 | 65 | 2015-05-13 11:59:29 |
+------+-------+-------+-----+-----------------------+
Q: How would I get the MAX(v)for each sid for each mid?
NB#1: In the example above ROW 53, 54, 55 have all the same value (40), but I would like to retrieve the row with the most recent timestamp, which is ROW 55.
Expected Output:
+------+-------+-------+-----+-----------------------+
| id | mid | sid | v | timestamp |
+------+-------+-------+-----+-----------------------+
| 51 | 10 | 1 | 40 | 2015-05-13 11:56:01 |
| 55 | 10 | 2 | 40 | 2015-05-13 11:57:01 |
| 56 | 11 | 1 | 50 | 2015-05-13 11:57:52 |
| 58 | 11 | 2 | 19 | 2015-05-13 11:58:59 |
| 60 | 11 | 3 | 65 | 2015-05-13 11:59:29 |
+------+-------+-------+-----+-----------------------+
Structure of the table:
NB#2:
Since this table has over 110 million entries, it is critical to have have date boundaries, which limits to ~8000 entries over a 24 hour period.
The query can be written as follows:
SELECT t1.id, t1.mid, t1.sid, t1.v, t1.ts
FROM yourtable t1
INNER JOIN (
SELECT mid, sid, MAX(v) as v
FROM yourtable
WHERE ts BETWEEN '2015-05-13 00:00:00' AND '2015-05-14 00:00:00'
GROUP BY mid, sid
) t2
ON t1.mid = t2.mid
AND t1.sid = t2.sid
AND t1.v = t2.v
INNER JOIN (
SELECT mid, sid, v, MAX(ts) as ts
FROM yourtable
WHERE ts BETWEEN '2015-05-13 00:00:00' AND '2015-05-14 00:00:00'
GROUP BY mid, sid, v
) t3
ON t1.mid = t3.mid
AND t1.sid = t3.sid
AND t1.v = t3.v
AND t1.ts = t3.ts;
Edit and Explanation:
The first sub-query (first INNER JOIN) fetches MAX(v) per (mid, sid) combination. The second sub-query is to identify MAX(ts) for every (mid, sid, v). At this point, the two queries do not influence each others' results. It is also important to note that ts date range selection is done in the two sub-queries independently such that the final query has fewer rows to examine and no additional WHERE filters to apply.
Effectively, this translates into getting MAX(v) per (mid, sid) combination initially (first sub-query); and if there is more than one record with the same value MAX(v) for a given (mid, sid) combo, then the excess records get eliminated by the selection of MAX(ts) for every (mid, sid, v) combination obtained by the second sub-query. We then simply associate the output of the two queries by the two INNER JOIN conditions to get to the id of the desired records.
Demo
select * from sensor_data s1 where s1.v in (select max(v) from sensor_data s2 group by s2.mid)
union
select * from sensor_data s1 where s1.v in (select max(v) from sensor_data s2 group by s2.sid);
IN ( SELECT ... ) does not optimize well. It is even worse because of being correlated.
What you are looking for is a groupwise-max .
Please provide SHOW CREATE TABLE; we need to know at least what the PRIMARY KEY is.
Suggested code
You will need:
With the WHERE: INDEX(timestamp, mid, sid, v, id)
Without the WHERE: INDEX(mid, sid, v, timestamp, id)
Code:
SELECT id, mid, sid, v, timestamp
FROM ( SELECT #prev_mid := 99999, -- some value not in table
#prev_sid := 99999,
#n := 0 ) AS init
JOIN (
SELECT #n := if(mid != #prev_mid OR
sid != #prev_sid,
1, #n + 1) AS n,
#prev_mid := mid,
#prev_sid := sid,
id, mid, sid, v, timestamp
FROM sensor_data
WHERE timestamp >= '2017-05-13'
timestamp < '2017-05-13' + INTERVAL 1 DAY
ORDER BY mid DESC, sid DESC, v DESC, timestamp DESC
) AS x
WHERE n = 1
ORDER BY mid, sid; -- optional
Notes:
The index is 'composite' and 'covering'.
This should make one pass over the index, thereby providing 'good' performance.
The final ORDER BY is optional; the results may be in reverse order.
All the DESC in the inner ORDER BY must be in place to work correctly (unless you are using MySQL 8.0).
Note how the WHERE avoids including both midnights? And avoids manually computing leap-days, year-ends, etc?
With the WHERE (and associated INDEX), there will be filtering, but a 'sort'.
Without the WHERE (and the other INDEX), sort will not be needed.
You can test the performance of any competing formulations via this trick, even if you do not have enough rows (yet) to get reliable timings:
FLUSH STATUS;
SELECT ...
SHOW SESSION STATUS LIKE 'Handler%';
This can also be used to compare different versions of MySQL and MariaDB -- I have seen 3 significantly different performance characteristics in a related groupwise-max test.

How to group same table differently in outer query and subquery

Table looks like this:
id | number | provider| datetime | keyword|country|
1 | 1 | Mobitel |2012-11-05| JAM | RS |
2 | 2 | Telekom |2013-04-25| ASTRO| RS |
3 | 1 | Si.Mobil|2013-04-27| DOMACE| BA |
4 | 4 | Telenor |2013-04-21| BIP | HR |
5 | 7 | VIP |2013-04-18| WIN | CZ |
6 | 13 | VIP |2014-05-21| DOMACE| RS |
7 | 5 | VIP |2014-06-04| WIN | HU |
I need to sum all numbers grouped by keyword and country and to sum all numbers again grouped by keyword, country and provider all in one query.
Here is how I tried to do it:
SELECT (SELECT SUM(number),country, keyword
FROM daily_subscriptions
WHERE datetime >= '2016-02-01 23:59:59'
GROUP BY country, keyword )
num_of_all_subs,
SUM(number) as num_of_subs,
country,
keyword,
provider
FROM daily_subscriptions
WHERE datetime >= '2016-02-01 23:59:59'
GROUP BY country, keyword, provider
But this query throws an error:
#1241 - Operand should contain 1 column(s)
Here is what I expect to get:
id | num_of_all_subs|num_of_subs | provider| datetime | keyword|country|
1 | 19 | 4 | Mobitel |2012-11-05| JAM | RS |
2 | 12 | 5 |Telekom |2013-04-25| ASTRO| RS |
3 | 18 | 1 |Si.Mobil |2013-04-27| DOMACE| BA |
4 | 42 | 21 |Telenor |2013-04-21| BIP | HR |
5 | 76 | 23 |VIP |2013-04-18| WIN | CZ |
6 | 13 | 3 |VIP |2014-05-21| DOMACE| RS |
7 | 53 | 11 |VIP |2014-06-04| WIN | HU |
Field num_of_all_subs meaning that sum of all numbers for lets say JAM(keyword) and RS(country) is 19 , but per Mobitel(provider) is num_of_subs 4 from all 19, since there are other providers for that country and keyword(even though they are not displayed in table schema).
Please help me to extract this data, since I'm stuck.
Your subquery for num_of_all_subs (which is a single number) must only return one column and, next problem, one row. Also, this subquery will be evaluated before you group, while you actually want to first group and get the columns num_of_subs, country, keyword and provider, and, afterwards, add another column num_of_all_subs to that first resultset.
You can do this exactly as just described: first get the grouped subquery (here called details), then use a dependent subquery to get, for each row in that subquery, the value for num_of_all_subs by looking at the table (again) and sum over all rows that have the same provider and country:
SELECT
(SELECT SUM(number)
FROM daily_subscriptions ds
WHERE datetime >= '2016-02-01 23:59:59'
and ds.country = details.country
and ds.keyword = details.keyword
) as num_of_all_subs,
details.*
from
(select
SUM(number) as num_of_subs,
country,
keyword,
provider
FROM daily_subscriptions
WHERE datetime >= '2016-02-01 23:59:59'
GROUP BY country, keyword, provider
) as details;
An alternative would be to do calculate both groups seperately, one including provider (details), and one without (all_subs). One will contain num_of_subs, one will contain num_of_all_subs. The you can combine (join) these two queries when they have the same country and keyword:
SELECT
all_subs.num_of_all_subs,
details.*
from
(select
SUM(number) as num_of_subs,
country,
keyword,
provider
FROM daily_subscriptions
WHERE datetime >= '2016-02-01 23:59:59'
GROUP BY country, keyword, provider
) as details
left join
(SELECT
SUM(number) as num_of_all_subs,
country,
keyword
FROM daily_subscriptions
WHERE datetime >= '2016-02-01 23:59:59'
GROUP BY country, keyword
) as all_subs
on all_subs.keyword = details.keyword and all_subs.country = details.country;
In your case, you can use a join instead of a left join, because every row in the first subquery will have a row in the second subquery, although it's usually the safer way do keep it.
While in theory, MySQL could execute these queries identically (and for less complicated queries, it will actually optimize and treat, whenever possible and useful, dependent subqueries like joins), in current MySQL versions this will most likely not be the case and the 2nd option is probably faster. Anyway, for both versions, a composite index on (country, keyword, provider) will do wonders.

What is SQL to select a property and the max number of occurrences of a related property?

I have a table like this:
Table: p
+----------------+
| id | w_id |
+---------+------+
| 5 | 8 |
| 5 | 10 |
| 5 | 8 |
| 5 | 10 |
| 5 | 8 |
| 6 | 5 |
| 6 | 8 |
| 6 | 10 |
| 6 | 10 |
| 7 | 8 |
| 7 | 10 |
+----------------+
What is the best SQL to get the following result? :
+-----------------------------+
| id | most_used_w_id |
+---------+-------------------+
| 5 | 8 |
| 6 | 10 |
| 7 | 8 |
+-----------------------------+
In other words, to get, per id, the most frequent related w_id.
Note that on the example above, id 7 is related to 8 once and to 10 once.
So, either (7, 8) or (7, 10) will do as result. If it is not possible to
pick up one, then both (7, 8) and (7, 10) on result set will be ok.
I have come up with something like:
select counters2.p_id as id, counters2.w_id as most_used_w_id
from (
select p.id as p_id,
w_id,
count(w_id) as count_of_w_ids
from p
group by id, w_id
) as counters2
join (
select p_id, max(count_of_w_ids) as max_counter_for_w_ids
from (
select p.id as p_id,
w_id,
count(w_id) as count_of_w_ids
from p
group by id, w_id
) as counters
group by p_id
) as p_max
on p_max.p_id = counters2.p_id
and p_max.max_counter_for_w_ids = counters2.count_of_w_ids
;
but I am not sure at all whether this is the best way to do it. And I had to repeat the same sub-query two times.
Any better solution?
Try to use User defined variables
select id,w_id
FROM
( select T.*,
if(#id<>id,1,0) as row,
#id:=id FROM
(
select id,W_id, Count(*) as cnt FROM p Group by ID,W_id
) as T,(SELECT #id:=0) as T1
ORDER BY id,cnt DESC
) as T2
WHERE Row=1
SQLFiddle demo
Formal SQL
In fact - your solution is correct in terms of normal SQL. Why? Because you have to stick with joining values from original data to grouped data. Thus, your query can not be simplified. MySQL allows to mix non-group columns and group function, but that's totally unreliable, so I will not recommend you to rely on that effect.
MySQL
Since you're using MySQL, you can use variables. I'm not a big fan of them, but for your case they may be used to simplify things:
SELECT
c.*,
IF(#id!=id, #i:=1, #i:=#i+1) AS num,
#id:=id AS gid
FROM
(SELECT id, w_id, COUNT(w_id) AS w_count
FROM t
GROUP BY id, w_id
ORDER BY id DESC, w_count DESC) AS c
CROSS JOIN (SELECT #i:=-1, #id:=-1) AS init
HAVING
num=1;
So for your data result will look like:
+------+------+---------+------+------+
| id | w_id | w_count | num | gid |
+------+------+---------+------+------+
| 7 | 8 | 1 | 1 | 7 |
| 6 | 10 | 2 | 1 | 6 |
| 5 | 8 | 3 | 1 | 5 |
+------+------+---------+------+------+
Thus, you've found your id and corresponding w_id. The idea is - to count rows and enumerate them, paying attention to the fact, that we're ordering them in subquery. So we need only first row (because it will represent data with highest count).
This may be replaced with single GROUP BY id - but, again, server is free to choose any row in that case (it will work because it will take first row, but documentation says nothing about that for common case).
One little nice thing about this is - you can select, for example, 2-nd by frequency or 3-rd, it's very flexible.
Performance
To increase performance, you can create index on (id, w_id) - obviously, it will be used for ordering and grouping records. But variables and HAVING, however, will produce line-by-line scan for set, derived by internal GROUP BY. It isn't such bad as it was with full scan of original data, but still it isn't good thing about doing this with variables. On the other hand, doing that with JOIN & subquery like in your query won't be much different, because of creating temporery table for subquery result set too.
But to be certain, you'll have to test. And keep in mind - you already have valid solution, which, by the way, isn't bound to DBMS-specific stuff and is good in terms of common SQL.
Try this query
select p_id, ccc , w_id from
(
select p.id as p_id,
w_id, count(w_id) ccc
from p
group by id,w_id order by id,ccc desc) xxx
group by p_id having max(ccc)
here is the sqlfidddle link
You can also use this code if you do not want to rely on the first record of non-grouping columns
select p_id, ccc , w_id from
(
select p.id as p_id,
w_id, count(w_id) ccc
from p
group by id,w_id order by id,ccc desc) xxx
group by p_id having ccc=max(ccc);

Fetch Unit consumption date-wise

I am struggling in to get result from mysql in the following way. I have 10 records in mysql db table having date and unit fields. I need to get used units on every date.
Table structure as follows, adding today unit with past previous unit in every record:
Date Units
---------- ---------
10/10/2012 101
11/10/2012 111
12/10/2012 121
13/10/2012 140
14/10/2012 150
15/10/2012 155
16/10/2012 170
17/10/2012 180
18/10/2012 185
19/10/2012 200
Desired output will be :
Date Units
---------- ---------
10/10/2012 101
11/10/2012 10
12/10/2012 10
13/10/2012 19
14/10/2012 10
15/10/2012 5
16/10/2012 15
17/10/2012 10
18/10/2012 5
19/10/2012 15
Any help will be appreciated. Thanks
There's a couple of ways to get the resultset. If you can live with an extra column in the resultset, and the order of the columns, then something like this is a workable approach.
using user variables
SELECT d.Date
, IF(#prev_units IS NULL
,#diff := 0
,#diff := d.units - #prev_units
) AS `Units_used`
, #prev_units := d.units AS `Units`
FROM ( SELECT #prev_units := NULL ) i
JOIN (
SELECT t.Date, t.Units
FROM mytable t
ORDER BY t.Date, t.Units
) d
This returns the specified resultset, but it includes the Units column as well. It's possible to have that column filtered out, but it's more expensive, because of the way MySQL processes an inline view (MySQL calls it a "derived table")
To remove that extra column, you can wrap that in another query...
SELECT f.Date
, f.Units_used
FROM (
query from above goes here
) f
ORDER BY f.Date
but again, removing that column comes with the extra cost of materializing that result set a second time.
using a semi-join
If you are guaranteed to have a single row for each Date value, either stored as a DATE, or as a DATETIME with the timecomponent set to a constant, such as midnight, and no gaps in the Date value, and Date is defined as DATE or DATETIME datatype, then another query that will return the specifid result set:
SELECT t.Date
, t.Units - s.Units AS Units_Used
FROM mytable t
LEFT
JOIN mytable s
ON s.Date = t.Date + INTERVAL -1 DAY
ORDER BY t.Date
If there's a missing Date value (a gap) such that there is no matching previous row, then Units_used will have a NULL value.
using a correlated subquery
If you don't have a guarantee of no "missing dates", but you have a guarantee that there is no more than one row for a particular Date, then another approach (usually more expensive in terms of performance) is to use a correlated subquery:
SELECT t.Date
, ( t.Units - (SELECT s.Units
FROM mytable s
WHERE s.Date < t.Date
ORDER BY s.Date DESC
LIMIT 1)
) AS Units_used
FROM mytable t
ORDER BY t.Date, t.Units
spencer7593's solution will be faster, but you can also do something like this...
SELECT * FROM rolling;
+----+-------+
| id | units |
+----+-------+
| 1 | 101 |
| 2 | 111 |
| 3 | 121 |
| 4 | 140 |
| 5 | 150 |
| 6 | 155 |
| 7 | 170 |
| 8 | 180 |
| 9 | 185 |
| 10 | 200 |
+----+-------+
SELECT a.id,COALESCE(a.units - b.units,a.units) units
FROM
( SELECT x.*
, COUNT(*) rank
FROM rolling x
JOIN rolling y
ON y.id <= x.id
GROUP
BY x.id
) a
LEFT
JOIN
( SELECT x.*
, COUNT(*) rank
FROM rolling x
JOIN rolling y
ON y.id <= x.id
GROUP
BY x.id
) b
ON b.rank= a.rank -1;
+----+-------+
| id | units |
+----+-------+
| 1 | 101 |
| 2 | 10 |
| 3 | 10 |
| 4 | 19 |
| 5 | 10 |
| 6 | 5 |
| 7 | 15 |
| 8 | 10 |
| 9 | 5 |
| 10 | 15 |
+----+-------+
This should give the desired result. I don't know how your table is called so I named it "tbltest".
Naming a table date is generally a bad idea as it also refers to other things (functions, data types,...) so I renamed it "fdate". Using uppercase characters in field names or tablenames is also a bad idea as it makes your statements less database independent (some databases are case sensitive and some are not).
SELECT
A.fdate,
A.units - coalesce(B.units, 0) AS units
FROM
tbltest A left join tbltest B ON A.fdate = B.fdate + INTERVAL 1 DAY