Get IDs from multiple columns in multiple tables as one set or array

I have multiple tables with each two rows of interest: connection_node_start_id and connection_node_end_id. My goal is to get a collection of all those IDs, either as a flat ARRAY or as a new TABLE consisting of one row.

Example output ARRAY:

result = {1,4,7,9,2,5}

Example output TABLE:

IDS
-------
1
4
7
9
2 
5

My fist attempt is somewhat clumsy and does not work properly as the SELECT statement just returns one row. It seems there must be a simple way to do this, can someone point me into the right direction?

CREATE OR REPLACE FUNCTION get_connection_nodes(anyarray)
  RETURNS anyarray AS
$$
DECLARE
  table_name varchar;
  result integer[];
  sel integer[];
BEGIN
  FOREACH table_name IN ARRAY $1
  LOOP
     RAISE NOTICE 'table_name(%)',table_name;
     EXECUTE 'SELECT ARRAY[connection_node_end_id, 
                           connection_node_start_id] FROM ' || table_name INTO sel;
    RAISE NOTICE 'sel(%)',sel;
    result  := array_cat(result, sel);  
  END LOOP;
  RETURN result;            
END
$$
  LANGUAGE 'plpgsql';

Test table:

connection_node_start_id | connection_node_end_id
--------------------------------------------------
 1                       | 4 
 7                       | 9 

Call:

SELECT get_connection_nodes(ARRAY['test_table']);

Result:

{1,4}  -- only 1st row, rest is missing

CREATE OR REPLACE FUNCTION get_connection_nodes(text[])
  RETURNS TABLE (ids int)
  LANGUAGE plpgsql AS
$func$
DECLARE
   _tbl text;
BEGIN
   FOREACH _tbl IN ARRAY $1
   LOOP
      RETURN QUERY EXECUTE format('
         SELECT t.id
         FROM   %I, LATERAL (VALUES (connection_node_start_id)
                                  , (connection_node_end_id)) t(id)'
       , _tbl);
   END LOOP;
END
$func$;

Related answer on dba.SE:

• SELECT DISTINCT on multiple columns

Or drop the loop and concatenate a single query. Probably fastest:

CREATE OR REPLACE FUNCTION get_connection_nodes2(text[])
  RETURNS TABLE (ids int)
  LANGUAGE plpgsql AS
$func$
BEGIN
RETURN QUERY EXECUTE (    
   SELECT string_agg(format(
             'SELECT t.id FROM %I, LATERAL (VALUES (connection_node_start_id)
                                                 , (connection_node_end_id)) t(id)'
           , tbl), ' UNION ALL ')
   FROM   unnest($1) tbl
   );
END
$func$;

Related:

• Loop through like tables in a schema

LATERAL was introduced with Postgres 9.3.

For very old Postgres version

You can use the set-returning function unnest() in the SELECT list, too:

CREATE OR REPLACE FUNCTION get_connection_nodes2(text[])
  RETURNS TABLE (ids int)
  LANGUAGE plpgsql AS
$func$
BEGIN
   RETURN QUERY EXECUTE (
   SELECT string_agg(
            'SELECT unnest(ARRAY[connection_node_start_id
                               , connection_node_end_id]) FROM ' || tbl
          , ' UNION ALL '
          )
   FROM (SELECT quote_ident(tbl) AS tbl FROM unnest($1) tbl) t
   );
END
$func$;

Should work with pg 8.4+ (or maybe even older). Works with current Postgres (9.4) as well, but LATERAL is much cleaner.

Or make it very simple:

CREATE OR REPLACE FUNCTION get_connection_nodes3(text[])
  RETURNS TABLE (ids int)
  LANGUAGE plpgsql AS
$func$
BEGIN
   RETURN QUERY EXECUTE (
   SELECT string_agg(format(
             'SELECT connection_node_start_id FROM %1$I
              UNION ALL
              SELECT connection_node_end_id FROM %1$I'
           , tbl), ' UNION ALL ')
   FROM   unnest($1) tbl
   );
END
$func$;

format() was introduced with pg 9.1.

Might be a bit slower with big tables because each table is scanned once for every column (so 2 times here). Sort order in the result is different, too - but that does not seem to matter for you.

Be sure to sanitize escape identifiers to defend against SQL injection and other illegal syntax. Details:

• Table name as a PostgreSQL function parameter