Unstable query behavior in windowing function in PostgreSQL 9.6

Question

I downloaded OSM data for my country from geofabrik.de, successfully imported it to PostgreSQL 9.6 installed on Ubuntu 16.04 and used it for several times. I also created web application, which works correctly. So I decided to add another functionality which returns top nearest special points (eg restaurants) from some points. For one nearest point it works, but when I want return array of them, it doesn't work. So I decomposed my problem and found strange behavior. When I executed following query:

SELECT t.osm_id
      FROM (
        SELECT DISTINCT ON (a.points) a.points, v.osm_id AS osm_id, MIN(ST_DISTANCE(v.the_geom, a.points)) OVER (PARTITION BY a.points ORDER BY ST_DISTANCE(v.the_geom, a.points))
        FROM (SELECT ST_GEOMFROMEWKT('SRID=4326;POINT(17.104854583740238 48.15099866770469)') AS points) a
        CROSS JOIN ways_vertices_pgr v
      ) AS t

it returns:

| osm_id            |
| ----------------- |
| 2338524511        |

When I displayed this point on map, it is placed far away from original point and after I changed point in subquery, the result remains same. Also I know there are many points between displayed and original point, which should be returned by query. Then I tried run following query:

SELECT t.*, t.osm_id
      FROM (
        SELECT DISTINCT ON (a.points) a.points, v.osm_id AS osm_id, MIN(ST_DISTANCE(v.the_geom, a.points)) OVER (PARTITION BY a.points ORDER BY ST_DISTANCE(v.the_geom, a.points))
        FROM (SELECT ST_GEOMFROMEWKT('SRID=4326;POINT(17.104854583740238 48.15099866770469)') AS points) a
        CROSS JOIN ways_vertices_pgr v
      ) AS t

and it returns:

| points                                             | osm_id   | min                  | osm_id     |
| -------------------------------------------------- | -------- | -------------------- | --------   |
| 0101000020E6100000010000C0D71A3140FFC3A1EC53134840 | 33169309 | 0.000124886435658481 | 33169309   |

Whole query except SELECT part remains same, but result is different and now it is correct. Can anyone suggest me how to change query to works properly?

Answer 1

When you use distinct on , you need an order by . I think this is the logic you want for the first query:

    SELECT DISTINCT ON (a.points) a.points, v.osm_id AS osm_id,ST_DISTANCE(v.the_geom, a.points) as dist
    FROM (SELECT ST_GEOMFROMEWKT('SRID=4326;POINT(17.104854583740238 48.15099866770469)') AS points) a CROSS JOIN
         ways_vertices_pgr v
    ORDER BY a.points, dist;

Answer 2

Check the output of EXPLAIN ANALYZE with your query to see exactly why the results are changing when you add the columns. Likely it's using a slightly different execution plan which affects the ordering of rows.

DISTINCT ON is by definition non-deterministic, meaning the results can change between executions. From the PostgreSQL 9.6 manual :

SELECT DISTINCT ON ... Note that the "first row" of a set is unpredictable unless the query is sorted on enough columns to guarantee a unique ordering of the rows arriving at the DISTINCT filter. (DISTINCT ON processing occurs after ORDER BY sorting.)

Adding an ORDER BY as Gordon suggested should give you repeatable results.