Oracle group qualifying rows by time period including zero counts

Question

Lets say I have an Oracle 10g table that logs all response times on my system. How can I get a count slow transactions by date, including zero counts for dates where no transactions were slow?

As an example

create table response_times (system_datetime timestamp, server_name varchar2(3), response_time number(9));

insert into response_times (system_datetime, server_name, response_time)
values (to_date('01/05/2014 12:30', 'DD/MM/YYYY HH24:MI'), 'D01', 500);

insert into response_times (system_datetime, server_name, response_time)
values (to_date('01/05/2014 13:45', 'DD/MM/YYYY HH24:MI'), 'D02', 700);

insert into response_times (system_datetime, server_name, response_time)
values (to_date('01/05/2014 14:01', 'DD/MM/YYYY HH24:MI'), 'D01', 2500);

insert into response_times (system_datetime, server_name, response_time)
values (to_date('02/05/2014 07:45', 'DD/MM/YYYY HH24:MI'), 'D02', 2500);

insert into response_times (system_datetime, server_name, response_time)
values (to_date('02/05/2014 08:30', 'DD/MM/YYYY HH24:MI'), 'D02', 500);

I want to see the number of transactions where the response time was more than 2000 milliseconds, like this:

TRUNC(system_datetime)  D01    D02
=====================   ===    ===
2014/05/01              1      0
2014/05/02              0      1
2014/05/03              0      0

Is there a way to retrieve this in a single sql statement? In practice there will be millions of rows, with (hopefully) only a few slow transactions on each day, and on some days there will be no slow transactions at all.

Now I know I can write a short PL/SQL procedure to obtain this using a temporary table of candidate dates, and a loop to issue a select count(*) for each time period in order to get the count values, but I am hoping there is a more elegant way...

Answer 1

The key to this is generating the dates. Here is one method:

with n as (
      select level as lvl from dual
      connect by level < 5
     ),
     dates as (
      select (to_date('2014-05-01', 'YYYY-MM-DD') + lvl - 1) as d
      from n
     )
select dates.d, coalesce(r.D01, 0) as D01, coalesce(r.D02, 0) as D02
from dates left outer join
     (select TRUNC(system_datetime) as dt,
             sum(case when server_name = 'D01' then 1 else 0 end) as D01,
             sum(case when server_name = 'D02' then 1 else 0 end) as D02
      from response_times
      group by TRUNC(system_datetime) 
     ) r
     on dates.d = r.dt;

Once you have the dates, the rest is just a left outer join.