Complex rank in SQL using Postgres - sql

Question

I'm in over my head with the SQL needed for a complex rank function. This is an app for a racing sport where I need to rank each Entry for a Timesheet based on the entry's :total_time.
The relevant models:
class Timesheet
has_many :entries
end
class Entry
belongs_to :timesheet
belongs_to :athlete
end
class Run
belongs_to :entry
end
An Entry's :total time isn't stored in the database. It's a calculated column of runs.sum(:finish). I use Postgres (9.3) rank() function to get Entries for a given Timesheet and rank them by this calculated column.
def ranked_entries
Entry.find_by_sql([
"SELECT *, rank() OVER (ORDER BY total_time asc)
FROM(
SELECT Entries.id, Entries.timesheet_id, Entries.athlete_id,
SUM(Runs.finish) AS total_time
FROM Entries
INNER JOIN Runs ON (Entries.id = Runs.entry_id)
GROUP BY Entries.id) AS FinalRanks
WHERE timesheet_id = ?", self.id])
end
So far so good. This returns my entry objects with a rank attribute which I can display on timesheet#show.
Now the tricky part. On a Timesheet, not every Entry will have the same number of runs. There is a cutoff (usually Top-20 but not always). This renders the rank() from Postgres inaccurate because some Entries will have a lower :total_time than the race winner because they didn't make the cutoff for the second heat.
My Question: Is it possible to do something like a rank() within a rank() to produce a table that looks like the one below? Or is there another preferred way? Thanks!
Note: I store times as integers, but I formatted them as the more familiar MM:SS in the simplified table below for clarity
| rank | entry_id | total_time |
|------|-----------|------------|
| 1 | 6 | 1:59.05 |
| 2 | 3 | 1:59.35 |
| 3 | 17 | 1:59.52 |
|......|...........|............|
| 20 | 13 | 56.56 | <- didn't make the top-20 cutoff, only has one run.

Answer 1

Let's create a table. (Get in the habit of including CREATE TABLE and INSERT statements in all your SQL questions.)
create table runs (
entry_id integer not null,
run_num integer not null
check (run_num between 1 and 3),
run_time interval not null
);
insert into runs values
(1, 1, '00:59.33'),
(2, 1, '00:59.93'),
(3, 1, '01:03.27'),
(1, 2, '00:59.88'),
(2, 2, '00:59.27');
This SQL statement will give you the totals in the order you want, but without ranking them.
with num_runs as (
select entry_id, count(*) as num_runs
from runs
group by entry_id
)
select r.entry_id, n.num_runs, sum(r.run_time) as total_time
from runs r
inner join num_runs n on n.entry_id = r.entry_id
group by r.entry_id, n.num_runs
order by num_runs desc, total_time asc
entry_id num_runs total_time
--
2 2 00:01:59.2
1 2 00:01:59.21
3 1 00:01:03.27
This statement adds a column for rank.
with num_runs as (
select entry_id, count(*) as num_runs
from runs
group by entry_id
)
select
rank() over (order by num_runs desc, sum(r.run_time) asc),
r.entry_id, n.num_runs, sum(r.run_time) as total_time
from runs r
inner join num_runs n on n.entry_id = r.entry_id
group by r.entry_id, n.num_runs
order by rank asc
rank entry_id num_runs total_time
--
1 2 2 00:01:59.2
2 1 2 00:01:59.21
3 3 1 00:01:03.27