Postgres dynamic filter conditions - postgresql

Question

I want to dynamically filter through data based on condition, which is stored in specific column. This condition can change for every row.
For example I have a table my_table with couple of columns, one of them is called foo, where there are couple of filter conditions such as AND bar > 1 or in the next row AND bar > 2 or in the next row AND bar = 33.
I have a query which looks like:
SELECT something from somewhere
LEFT JOIN otherthing on some_condition
WHERE first_condition AND second_condition AND
here_i_want_dynamically_load_condition_from_my_table.foo
What is the correct way to do it? I have read some articles about dynamic queries, but I am not able to find a correct way.

Answer 1

This is impossible in pure SQL: at query time, the planner has to know your exact logic. Now, you can hide it away in a function (in pseudo-sql):
CREATE FUNCTION do_I_filter_or_not(some_id) RETURNS boolean AS '
BEGIN
value = select some_value from table where id = some_id
condition_type = SELECT ... --Query a condition type for this row
condition_value = SELECT ... --Query a condition value for this row
if condition_type = 'equals' and condition_value = value
return true
if condition_type = 'greater_than' and condition_value < value
return true
if condition_type = 'lower_than' and condition_value > value
return true
return false;
END;
'
LANGUAGE 'plpgsql';
And query it like this:
SELECT something
FROM somewhere
LEFT JOIN otherthing on some_condition
WHERE first_condition
AND second_condition
AND do_I_filter_or_not(somewhere.id)
Now the performance will be bad: you have to invoke that function potentially on every row in the query; triggering lots of subqueries.
Thinking about it, if you just want <, >, =, and you have a table (filter_criteria) describing for each id what the criteria is you can do it:
CREATE TABLE filter_criteria AS(
some_id integer,
equals_threshold integer,
greater_than_threshold integer,
lower_than_threshold integer
-- plus a check that two thresholds must be null, and one not null
)
INSERT INTO filter_criteria (1, null, 5, null); -- for > 5
And query like this:
SELECT something
FROM somewhere
LEFT JOIN otherthing on some_condition
LEFT JOIN filter_criteria USING (some_id)
WHERE first_condition
AND second_condition
AND COALESCE(bar = equals_threshold, true)
AND COALESCE(bar > greater_than_threshold, true)
AND COALESCE(bar < lower_than_threshold, true)
The COALESCEs are here to default to not filtering (AND true) if the threshold is missing (bar = equals_threshold will yield null instead of a boolean).
The planner has to know your exact logic at query time: now you're just doing 3 passes of filtering, with a =, <, > check each time. That'd still be more performant than idea #1 with all the subquerying.