Efficiently select the most specific result from a table - postgresql

Question

I have a table roughly as follows:
CREATE TABLE t_table (
f_userid BIGINT NOT NULL
,f_groupaid BIGINT
,f_groupbid BIGINT
,f_groupcid BIGINT
,f_itemid BIGINT
,f_value TEXT
);
The groups are orthogonal, so no hierarchy can be implied beyond the fact that every entry in the table will have a user ID. There is no uniqueness in any of the columns.
So for example a simple setup might be:
INSERT INTO t_table VALUES (1, NULL, NULL, NULL, NULL, 'Value for anything by user 1');
INSERT INTO t_table VALUES (1, 5, 2, NULL, NULL, 'Value for anything by user 1 in groupA 5 groupB 2');
INSERT INTO t_table VALUES (1, 4, NULL, 1, NULL, 'Value for anything by user 1 in groupA 5 and groupC 1');
INSERT INTO t_table VALUES (2, NULL, NULL, NULL, NULL, 'Value for anything by user 2');
INSERT INTO t_table VALUES (2, 1, NULL, NULL, NULL, 'Value for anything by user 2 in groupA 1');
INSERT INTO t_table VALUES (2, 1, 3, 4, 5, 'Value for item 5 by user 2 in groupA 1 and groupB 3 and groupC 4');
For any given set of user/groupA/groupB/groupC/item I want to be able to obtain the most specific item in the table that applies. If any of the given set are NULL then it can only match relevant columns in the table which contain NULL. For example:
// Exact match
SELECT MostSpecific(1, NULL, NULL, NULL, NULL) => "Value for anything by user 1"
// Match the second entry because groupC and item were not specified in the table and the other items matched
SELECT MostSpecific(1, 5, 2, 3, NULL) => "Value for anything by user 1 in groupA 5 groupB 2"
// Does not match the second entry because groupA is NULL in the query and set in the table
SELECT MostSpecific(1, NULL, 2, 3, 4) => "Value for anything by user 1"
The obvious approach here is for the stored procedure to work through the parameters and find out which are NULL and not, and then call the appropriate SELECT statement. But this seems very inefficient. IS there a better way of doing this?

Answer 1

This should do it, just filter out any non matching rows using a WHERE, then rank the remaining rows by how well they match. If any column doesn't match, the whole bop expression will result in NULL, so we filter that out in an outer query where we also order by match and limit the result to only the single best match.
CREATE FUNCTION MostSpecific(BIGINT, BIGINT, BIGINT, BIGINT, BIGINT)
RETURNS TABLE(f_userid BIGINT, f_groupaid BIGINT, f_groupbid BIGINT, f_groupcid BIGINT, f_itemid BIGINT, f_value TEXT) AS
'WITH cte AS (
SELECT *,
CASE WHEN f_groupaid IS NULL THEN 0 WHEN f_groupaid = $2 THEN 1 END +
CASE WHEN f_groupbid IS NULL THEN 0 WHEN f_groupbid = $3 THEN 1 END +
CASE WHEN f_groupcid IS NULL THEN 0 WHEN f_groupcid = $4 THEN 1 END +
CASE WHEN f_itemid IS NULL THEN 0 WHEN f_itemid = $5 THEN 1 END bop
FROM t_table
WHERE f_userid = $1
AND (f_groupaid IS NULL OR f_groupaid = $2)
AND (f_groupbid IS NULL OR f_groupbid = $3)
AND (f_groupcid IS NULL OR f_groupcid = $4)
AND (f_itemid IS NULL OR f_itemid = $5)
)
SELECT f_userid, f_groupaid, f_groupbid, f_groupcid, f_itemid, f_value FROM cte
WHERE bop IS NOT NULL
ORDER BY bop DESC
LIMIT 1'
LANGUAGE SQL
//
An SQLfiddle to test with.

Answer 2

Try something like:
select *
from t_table t
where f_userid = $p_userid
and (t.f_groupaid is not distinct from $p_groupaid or t.f_groupaid is null) --null in f_groupaid matches both null and not null values
and (t.f_groupbid is not distinct from $p_groupbid or t.f_groupbid is null)
and (t.f_groupcid is not distinct from $p_groupcid or t.f_groupcid is null)
order by (t.f_groupaid is not distinct from $p_groupaid)::int -- order by count of matches
+(t.f_groupbid is not distinct from $p_groupbid)::int
+(t.f_groupcid is not distinct from $p_groupcid)::int desc
limit 1;
It will give you the best match on groups.
A is not distinct from B fill return true if A and B are equal or both null.
::int means cast ( as int). Casting boolean true to int will give 1 (You can not add boolean values directly).

Answer 3

SQL Fiddle
create or replace function mostSpecific(
p_userid bigint,
p_groupaid bigint,
p_groupbid bigint,
p_groupcid bigint,
p_itemid bigint
) returns t_table as $body$
select *
from t_table
order by
(p_userid is not distinct from f_userid or f_userid is null)::integer
+
(p_groupaid is not distinct from f_groupaid or f_userid is null)::integer
+
(p_groupbid is not distinct from f_groupbid or f_userid is null)::integer
+
(p_groupcid is not distinct from f_groupcid or f_userid is null)::integer
+
(p_itemid is not distinct from f_itemid or f_userid is null)::integer
desc
limit 1
;
$body$ language sql;