Question

I should be able to do this but the brain appears to have melted down

Database is postgres

Table structure is simple, just four columns that matter:

Location, User, Activity, DateTime

I need to sort by location and output just the latest occurrence of each activity that has taken place. I don’t know in advance what the activities are

Something like

Location    User    Activity   DateTime
London      Fred    A          08-29-2012
London      Fred    B          08-27-2012
Paris       John    A          08-29-2012
Tokyo       Fred    A          08-17-2012
Tokyo       Jane    D          08-29-2012

Thanks Folks

Answer 1

Fastest way in PostgreSQL is probably with DISTINCT ON:

SELECT DISTINCT ON (location, activity)
       location, activity, datetime, usr
FROM   tbl
ORDER  BY location, activity, datetime DESC, usr; -- usr only to break ties

Short form with positional parameters:

SELECT DISTINCT ON (1 ,2)
       location, activity, datetime, usr
FROM   tbl
ORDER  BY 1, 2, 3 DESC, 4;

This combines sorting and reducing to distinct rows in one operation.
More details, explanation and benchmark in this related answer.

user is a reserved word. Don’t actually use it as column name. I substituted with usr.

If performance should be crucial an index like the following will make the difference:

CREATE INDEX tbl_multi_idx ON tbl (location, activity, datetime DESC, usr);