Question

I have a table that looks like something like this:

timestamp                value           person
===============================================
2010-01-12 00:00:00       33              emp1
2010-01-12 11:00:00       22              emp1
2010-01-12 09:00:00       16              emp2
2010-01-12 08:00:00       16              emp2
2010-01-12 12:12:00       45              emp3
2010-01-12 13:44:00       64              emp4
2010-01-12 06:00:00       33              emp1
2010-01-12 15:00:00       12              emp5

I wanted to find the maximum value associated with each person. The obvious query was:

select person,max(value) from table group by person

Now I wanted to include the timestamp associated with each max(value). I could not use timestamp column in the above query because as everyone knows, it won’t appear in the group by clause. So I wrote this instead:

select x.timestamp,x.value,x.person from table as x,
(select person,max(value) as maxvalue from table group by person order by maxvalue 
 desc) as y
where x.person = y.person
and x.value = y.maxvalue

This works — to an extent. I now see:

timestamp                value           person
===============================================
2010-01-12 13:44:00       64              emp4
2010-01-12 12:12:00       45              emp3
2010-01-12 06:00:00       33              emp1
2010-01-12 00:00:00       33              emp1
2010-01-12 08:00:00       16              emp2
2010-01-12 09:00:00       16              emp2
2010-01-12 15:00:00       12              emp5

The problem is now I get all the entries for emp1 and emp2 that ends up with the same max(value).

Suppose among emp1 and emp2, I only want to see the entry with the latest timestamp. IOW, I want this:

timestamp                value           person
===============================================
2010-01-12 13:44:00       64              emp4
2010-01-12 12:12:00       45              emp3
2010-01-12 06:00:00       33              emp1
2010-01-12 09:00:00       16              emp2
2010-01-12 15:00:00       12              emp5

What kind of query would I have to write? Is it possible to extend the nested query I wrote to achieve what I want or does one have to rewrite everything from the scratch?

If its important, because I am using Sqlite, timestamps are actually stored as julian days. I use the datetime() function to convert them back to a string representation in every query.

Answer 1

You were almost there:

SELECT max(x.timestamp) AS timestamp, x.value, x.person
     , y.max_value, y.ct_value, y.avg_value
FROM   table AS x
JOIN  (
    SELECT person
         , max(value) as max_value
         , count(value) as ct_value
         , avg(value) as avg_value
    FROM   table
    GROUP  BY person
    ) AS y ON (x.person, x.value) = (y.person, y.max_value)
GROUP BY x.person, x.value, y.max_value, y.ct_value, y.avg_value
-- ORDER  BY x.person, x.value

You cannot compute max(x.timestamp) in the same nested query, because you don’t want the absolute maximum per person, but the one accompanying the maximum value. So you have to aggregate another time on the next query level.

Compute max(x.timestamp) before you convert it to its string representation – though your format would sort correctly, too. But that that should perform better.

Note how I transformed your cross join with where conditions to an [inner] join with a (simplified) join condition. Does the same, just more like the canonical way of the SQL standard and more readable.

All of this could be done in one query level with window functions (max() and first_value()), which are implemented in all the bigger RDBMS (except MYSQL), but not in SQLite.

---

Edit

Included additional aggregates after request in comment.