How do I get the sum of a column across multiple keys? - mysql

Question

I have data that looks like this:
id int (11) primary key auto_increment
key int (2)
type int (2)
data int (4)
timestamp datetime
There are 5 different keys - 1,2,3,4,5 and three types - 1,2,3
Data is put in continuously against a key and of a particular type.
What I need to extract is a sum of the data for a particular type (say, type 1) across all 5 keys (1,2,3,4,5) so it is a sum of exactly 5 records. I only want to sum the latest (max(timestamp) values (there are 5 of them) of data for each key, but they may all have different timestamps.
Something like this....
SELECT sum(data) FROM table WHERE type='1' AND timestamp=(SELECT max(timestamp FROM table WHERE type='1' GROUP BY key)
Or something like that. That isn't even close of course. I am completely lost on this one. it feels like I need to group by key but the syntax eludes me. Any suggestions are appreciated.
EDIT: additional info:
if: 'data' is temperature. 'key' is day of the week. 'type' is morning, noon or night
So the data might look like
morning mon 70 (timestamp)
noon tue 78 (timestamp)
morning wed 72 (timestamp)
night tue 74 (timestamp)
morning thu 76 (timestamp)
noon wed 77 (timestamp)
night fri 78 (timestamp)
noon tue 79 (timestamp)
If these are in timestamp order (desc) and I want the sum of most recent noon temps for all five days, the result would be: 155 in this case since the last noon was also tuesday and it was earlier and thus, not included. Make sense? I want sum of 'data' for any key, specific type, latest timestamp only. In this example, I would be summing at most 7 pieces of data.

Answer 1

If the timestamp column is guaranteed to be unique for each (key,type) (That is, there's a UNIQUE constraint ON (key,type,timestamp), then this query will return the specified resultset. (This isn't the only approach, but it is a familiar pattern):
SELECT SUM(t.data) AS latest_total
FROM mytable t
JOIN ( SELECT h.type
, h.key
, MAX(h.timestamp) AS max_ts
FROM mytable h
WHERE h.type='1'
GROUP
BY h.type
, h.key
) m
ON m.type = t.type
AND m.key = t.key
AND m.max_ts = t.timestamp
The inline view assigned an alias of m returns the "latest" timestamp for type=1 for all 5 key values (if at least one row exists)
That is joined to the original table, to retrieve the row that has that "latest" timestamp.
A suitable index with leading columns of type,key,timestamp will likely improve performance.
(That's based on my understanding of the specification; I may not be totally clear on the specification. What this query is doing is getting the latest timestamp for the type=1 rows. If there happen to be two (or more) rows with the same latest timestamp value for a given key and type, this query will retrieve both (or all) of those rows, and include them in the sum.
We could add a GROUP BY t.type on that query, and that wouldn't change the result, since we are guaranteed that the t.type will be equal to the constant 1 (specified in the predicate in the WHERE clause of the inline view query.)
But we would need to add the GROUP BY if we wanted to get totals for all three type in the same query:
SELECT t.key
, SUM(t.data) AS latest_total
FROM mytable t
JOIN ( SELECT h.type
, h.key
, MAX(h.timestamp) AS max_ts
FROM mytable h
WHERE h.type IN ('1','2','3')
GROUP
BY h.type
, h.key
) m
ON m.type = t.type
AND m.key = t.key
AND m.max_ts = t.timestamp
GROUP
BY t.key
NOTE:
Using reserved words as identifiers (e.g. TIMESTAMP and KEY isn't illegal, but those identifiers (usually) need to be enclosed in backticks. But changing the names of these columns so that they aren't reserved words is best practice.

Answer 2

SELECT SUM(data)
FROM ( SELECT CONCAT(MAX(timestamp), '_', type) AS customId
FROM table
WHERE type = '1'
GROUP BY key ) a
JOIN table b ON a.customId = CONCAT(b.timestamp, '_', type)
GROUP BY type;
This would probably do the trick...
SQL-Fiddle

Answer 3

I would for simplicity and maintainability use a temp-table and fill it with several statements. The solution with "union-subselect" looks a bit long for me.
So
drop tamporary table if exists tmp_data;
create temporary table tmp_data (type int, value int);
insert into tmp_data select 1, value from data_table where type=1 order by timestamp desc limit 5;
insert into tmp_data select 2, value from data_table where type=2 order by timestamp desc limit 5;
insert into tmp_data select 3, value from data_table where type=3 order by timestamp desc limit 5;
select type, sum(value) as total from tmp_data group by type;
EDIT:
The subselect-solution would be similar, and since there are only 3 types not too bad
select type, sum(value) as total from
(select 1 as type, value from data_table where type=1 order by timestamp desc limit 5
union
select 2 as type, value from data_table where type=2 order by timestamp desc limit 5
union
select 3 as type, value from data_table where type=3 order by timestamp desc limit 5) as subtab group by type;
Hope that helps.