Query performace with "partition by" clause.

Question

Below is my query
>
select event_type, time, count(event_id) as no_of_events from (
select e.event_type, t.time , e.id as event_id from time t
left outer join events e partition by (event_type)
on t.time < e.end_time and (t.time + 1) > e.start_time
where t.time >= '2008-01-01' and t.time < '2008-02-01'
) events_by_event_type
group by event_type, time
order by event_type, time
The idea is to get a count of active "events" of each "event_type", for each day between 2 dates. The "time" table has one row each for each day. An event is said to be active for a day , when it's end_time - start_time overlaps the day's beginning and end.
The query works but always does a full table scan of the events table.
I tried creating following indexes on the events table , but none of them is ever used.
(event_type,start_time)
(event_type,end_time)
(event_type,start_time,end_time)
(start_time)
(end_time)
(start_time,end_time)
How can I avoid the full table scan of the "events" table in the above query ?
fyi the events table looks like
>
id number not null primary key,
event_type number not null,
start_date date not null,
end_date date not null

Answer 1

What I want is to avoid the full table scan on the
"events" table. I don't think adding an index on the
'time' table will help there.The conditions you have on events are
t.time < e.end_time and (t.time + 1) > e.start_timeSo you should have an index on the columns end_time and start_time to avoid a full table scan.
But anyway is that query slow?
Bye Alessandro
Message was edited by:
Alessandro Rossi
Plus I would add two more predicates to the query to enforce a range scan. If I did well they should always be true for the rows you want. They are there just to tell the CBO that the scan on end_time has to begin from '2008-02-01' and the scan on start_time has to finish on ('2008-01-01' - 1). Sometime this kind of additional conditions helped me.
select event_type, time, count(event_id) as no_of_events
from (
          select e.event_type, t.time , e.id as event_id from time t
          left outer join events e partition by (event_type) on (
               t.time between e.start_time - 1 and e.end_time
          where t.time >= '2008-01-01' and t.time < '2008-02-01'
               and e.end_time >= '2008-02-01' and e.start_time <= '2008-01-01' - 1
) events_by_event_type
group by event_type, time
order by event_type, time