Recursive query challenge - simple parent/child example

Question

Note: with help from RhodiumToad on #postgresql, I\'ve arrived at a solution, which I posted as answer. If anyone can improve on this, please chime in!

I have not be

Answer 1

An alternative approach would be to traverse the graph in reversed order:

WITH RECURSIVE cte AS (
   SELECT array[r.ancestor_node_id, r.descendant_node_id] AS path
   FROM   node_relations r
   LEFT   JOIN node_relations r0 ON r0.ancestor_node_id = r.descendant_node_id
   WHERE  r0.ancestor_node_id IS NULL  -- start at the end

   UNION ALL 
   SELECT r.ancestor_node_id || c.path
   FROM   cte c
   JOIN   node_relations r ON r.descendant_node_id = c.path[1]
   ) 
SELECT path
FROM   cte
ORDER  BY path;

This produces a subset with every path from each root node to its ultimate descendant. For deep trees that also spread out a lot this would entail much fewer join operations. To additionally add every sub-path, you could append a LATERAL join to the outer SELECT:

WITH RECURSIVE cte AS (
   SELECT array[r.ancestor_node_id, r.descendant_node_id] AS path
   FROM   node_relations r
   LEFT   JOIN node_relations r0 ON r0.ancestor_node_id = r.descendant_node_id
   WHERE  r0.ancestor_node_id IS NULL  -- start at the end

   UNION ALL 
   SELECT r.ancestor_node_id || c.path
   FROM   cte c
   JOIN   node_relations r ON r.descendant_node_id = c.path[1]
   ) 
SELECT l.path
FROM   cte, LATERAL (
   SELECT path[1:g] AS path
   FROM   generate_series(2, array_length(path,1)) g
   ) l
ORDER  BY l.path;

I ran a quick test, but it didn't run faster than RhodiumToad's solution. It might still be faster for big or wide tables. Try with your data.