PostgreSQL 9.4 - select dynamic subset of JSONB fields - postgresql-9.4

Question

Suppose I have table with some JSONB column. JSON is rather huge, and in most of the cases I need to retrieve just 5% of its content by field names (just to improve performance). Set of the needed fields varies from case to case, but it is still around 5% of whole JSON data.
I know how to do this for hardcoded set of fields. The question is - is it possible to make for externally provided set of fields. This needs to be implemented as stored function.
Here is sample with 'hardcoded set of fields':
CREATE TABLE test_data (
json_data jsonb NOT NULL
);
INSERT INTO test_data(json_data)
VALUES ('{"row_id": 1, "f1": 1, "f2": 2, "f3": 3, "f4": 4}'),
('{"row_id": 2, "f1": 1, "f2": 2, "f3": 3, "f4": 4}');
CREATE OR REPLACE FUNCTION get_important_data(IN data_set varchar = 'row_id,f1,f2')
RETURNS table (data json)
LANGUAGE plpgsql AS $$
BEGIN
RETURN QUERY
SELECT (SELECT row_to_json(data_row) FROM (
-- this needs to be dynamic
select json_data->>'row_id' as "row_id",
json_data->>'f1' as "f1",
json_data->>'f2' as "f2"
) data_row ) as data
FROM test_data;
END $$;
SELECT get_important_data(/* use default data set*/);
SELECT get_important_data('row_id,f2,f4');

Answer 1

While I think (I'm not sure though) you could get away without any dynamic SQL and some cumbersome joins and pivot-like queries, I think the most straightforward and possibly performing way here is a little dynamic query in your function.
This would be my quick take on your solution. Note the explicit conversion to text, to recreate the effect of your ->> operator:
CREATE OR REPLACE FUNCTION get_important_data(data_set text DEFAULT 'row_id,f1,f2') RETURNS TABLE (
data json
) AS $$
BEGIN
RETURN QUERY EXECUTE format('SELECT row_to_json(j.*) FROM test_data, LATERAL jsonb_to_record(json_data) AS j(%s);',
(SELECT array_to_string(array_agg(trim(field)), ' text, ') || ' text'
FROM unnest(string_to_array(data_set, ',')) AS field)
);
END;
$$ LANGUAGE plpgsql;
But in my opinion, since you are ultimately returning a subset of a json filed still in json format, you'd better leave it untouched. I like the following version more, and it should run faster too:
CREATE OR REPLACE FUNCTION get_important_data(data_set text DEFAULT 'row_id,f1,f2') RETURNS TABLE (
data json
) AS $$
DECLARE
rec_fields record;
arr_fields text[];
BEGIN
FOR rec_fields IN
SELECT trim(unnest(string_to_array(data_set, ','))) AS field
LOOP
arr_fields := arr_fields || format('json_data->%1$L AS %1$I', rec_fields.field);
END LOOP;
RETURN QUERY EXECUTE format('SELECT row_to_json(r) FROM (SELECT %s FROM test_data) AS r', array_to_string(arr_fields, ','));
END;
$$ LANGUAGE plpgsql;
The final touch would be dropping that text input parameter in favour of a json one ('["row_id", "f1", "f2"]'), but I'll leave that for you.