Fetch max value from a sort-of incomplete dataset - sql

Question

A number of devices return a value. Only upon change, this value gets stored in a table:
Device Value Date
B 5 2017-07-01
C 2 2017-07-01
A 3 2017-07-02
C 1 2017-07-04
A 6 2017-07-04
Values may enter the table at any date (i.e. date doesn't increment continiously). Several devices may store their value on the same date.
Note that, even though there are usually only a few devices for each date in the table, all devices actually have a value at that date: it's the latest one stored until then. For example, on 2017-07-02 only device A stored a value. The values for B and C on that date are the ones stored on 2017-07-01; these are still valid on -02, they just did not change.
To retrieve the values for all devices on a given date, e.g. 2017-07-04, I'm using this:
select device, value from data inner join (select device, max(date) as date from data where date <= "2017-07-04" group by device) latestdate on data.device = latestdate.device and data.date = latestdate.date
Device Value
A 6
B 5
C 1
Question: I'd like to read the max value of all devices on all dates in a given range. The result set would be like this:
Date max(value)
2017-07-01 5
2017-07-02 5
2017-07-04 6
.. and I have no clue if that's possible using only SQL. Until now all I got was lost in an exceptional bunch of joins and groupings.
(Database is sqlite3. Generic SQL would be nice, but I'd still be happy to hear about solutions specific to other databases, especially PostgreSQL or MariaDB.)
Extra bonus: Include the missing date -03, to be exact: returning values at given dates, not necessarily the ones appearing in the table.
Date max(value)
2017-07-01 5
2017-07-02 5
2017-07-03 5
2017-07-04 6

Answer 1

I think the most generic way to approach this is using a separate query for each date. There are definitely simpler methods, depending on the database. But getting one that works for SQLite, MariaDB, and Postgres is not going to use any sophisticated functionality:
select '2017-07-01' as date, max(data.value)
from data inner join
(select device, max(date) as date
from data
where date <= '2017-07-01' group by device
) latestdate
on data.device = latestdate.device and data.date = latestdate.date
union all
select '2017-07-02' as date, max(data.value)
from data inner join
(select device, max(date) as date
from data
where date <= '2017-07-02' group by device
) latestdate
on data.device = latestdate.device and data.date = latestdate.date
select '2017-07-03' as date, max(data.value)
from data inner join
(select device, max(date) as date
from data
where date <= '2017-07-03' group by device
) latestdate
on data.device = latestdate.device and data.date = latestdate.date
select '2017-07-04' as date, max(data.value)
from data inner join
(select device, max(date) as date
from data
where date <= '2017-07-04' group by device
) latestdate
on data.device = latestdate.device and data.date = latestdate.date;

Answer 2

This should be a solution for your problem.
It should be cross-database, since OVER clause is supported by the most of the databases.
You should create a table with all the dates("ALL_DATE" in the query), otherwise every database has a specific way to do it without a table.
WITH GROUPED_BY_DATE_DEVICE AS (
SELECT DATE, DEVICE, SUM(VALUE) AS VALUE FROM DEVICE_INFO
GROUP BY DATE, DEVICE
), GROUPED_BY_DATE AS (
SELECT A.DATE, MAX(VALUE) AS VALUE
FROM ALL_DATE A
LEFT JOIN GROUPED_BY_DATE_DEVICE B
ON A.DATE = B.DATE
GROUP BY A.DATE
)
SELECT DATE, MAX(VALUE) OVER (ORDER BY DATE) AS MAX_VALUE
FROM GROUPED_BY_DATE
ORDER BY DATE;