Oracle SQL - Counting entities per day from entity start date and end date

Question

I'm starting with a table like so:

ENTITY_ID	META_ATTRIB_1	META_ATTRIB_2	START_DATE	END_DATE
1	FOO	BAR	2020-01-01	2020-12-01

I'm would like to end up with a count of entities per day that fall within given sets of meta-attributes:

DAY	META_ATTRIB_1	META_ATTRIB_2	COUNT
2020-01-01	FOO	BAR	1
2020-01-02	FOO	BAR	1
2020-01-03	FOO	BAR	1

Right now I'm doing this by generating a sequence of dates from DUAL and joining the target table in via DAY BETWEEN START_DATE AND END_DATE and grouping by DAY, META_ATTRIB_1, META_ATTRIB_2 .

This method is running into performance problems. Is there a better method for splitting out each of these entity rows across the desired sequence of days and then aggregating it back for a by day count?

Answer 1

A typical approach uses a recursive to generate one row per day in each range, then aggregation:

with cte (meta_attrib_1, meta_attrib_2, dt, end_date) as (
    select meta_attrib_1, meta_attrib_2, start_date, end_date from mytable
    union all
    select meta_attrib_1, meta_attrib_2, dt + 1, end_date from cte where dt < end_date
)
select dt, meta_attrib_1, meta_attrib_2, count(*) as cnt
from cte
group by dt, meta_attrib_1, meta_attrib_2

This is pretty close to the logic that you described. You did not show your actual query so it is hard to tell whether this is a better solution than what you are doing currently.

Answer 2

You might find that a recursive CTE is faster:

with cte (day, meta_attrib1, meta_attrib2, end_date)
      select start_date, meta_attrib1, meta_attrib2, end_date
      from t
      union all
      select start_date + interval '1' day, meta_attrib1, meta_attrib2, end_date
      from cte
      where day < end_date
     )
select day, meta_attrib1, meta_attrib2, count(*)
from cte
group by day, meta_attrib1, meta_attrib2;

The advantage to a recursive CTE is that it "localizes" the expansion of the dates. Instead of relying on a non-equijoin, it simply churns out the additional dates.

This is still producing a separate row for each day for each of the original rows. That means that the aggregation could be the bottleneck.