How expensive is ST_GeomFromText

Question

In postgis, is the ST_GeomFromText call very expensive? I ask mostly because I have a frequently called query that attempts to find the point that is nearest another point that matches some criteria, and which is also within a certain distance of that other point, and the way I currently wrote it, it's doing the same ST_GeomFromText twice:

 $findNearIDMatchStmt = $postconn->prepare(
    "SELECT     internalid " .
    "FROM       waypoint " .
    "WHERE      id = ? AND " .
    "           category = ? AND ".
    "           (b.category in (1, 3) OR type like ?) AND ".
    "           ST_DWithin(point, ST_GeomFromText(?," . SRID .
    "           ),".  SMALL_EPSILON . ") " .
    "           ORDER BY ST_Distance(point, ST_GeomFromText(?,", SRID .
    "           )) " .
    "           LIMIT 1");

Is there a better way to re-write this?

Slightly OT: In the preview screen, all my underscores are being rendered as & # 9 5 ; - I hope that's not going to show up that way in the post.

Answer 1

I don't believe ST_GeomFromText() is particularly expensive, although in the past I've optimized PostGIS queries by creating a function, declaring a variable and then assigning the result of ST_GeomFromText to the variable.

Have you tried checking the execution plan for you query with a variety of different parameters because that should give you a definite idea of which bits of the query are taking the time?

I'm guessing most of the execution time will be in the calls to ST_DWithin() and ST_Distance(), although if the id and category columns aren't indexed then it might be doing some interesting table scanning.

Answer 2

@Ubiguch It appears that ST_DWithin uses the spatial index, so that seems to cut down on the number of points to be queried pretty quickly.

 navaid=> explain select internalid from waypoint where id != 'KROC' AND ST_DWithin(point,                                                                  ST_GeomFromText('POINT(-77.6723888888889 43.1188611111111)',4326), 0.05) order by st_distance(point, st_geomfromtext('POINT(-77.6723888888889 43.1188611111111)',4326)) limit 1;
                                                                                                                                                                                                                                                      QUERY PLAN                                                                                                                                                                                                                                                       
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 Limit  (cost=8.37..8.38 rows=1 width=104)
   ->  Sort  (cost=8.37..8.38 rows=1 width=104)
         Sort Key: (st_distance(point, '0101000020E61000002FFE676B086B53C0847E44D7368F4540'::geometry))
         ->  Index Scan using waypoint_point_idx on waypoint  (cost=0.00..8.36 rows=1 width=104)
               Index Cond: (point && '0103000020E61000000100000005000000000000C03B6E53C000000060D0884540000000C03B6E53C0000000409D95454000000020D56753C0000000409D95454000000020D56753C000000060D0884540000000C03B6E53C000000060D0884540'::geometry)
               Filter: (((id)::text <> 'KROC'::text) AND (point && '0103000020E61000000100000005000000000000C03B6E53C000000060D0884540000000C03B6E53C0000000409D95454000000020D56753C0000000409D95454000000020D56753C000000060D0884540000000C03B6E53C000000060D0884540'::geometry) AND ('0101000020E61000002FFE676B086B53C0847E44D7368F4540'::geometry && st_expand(point, 0.05::double precision)) AND (st_distance(point, '0101000020E61000002FFE676B086B53C0847E44D7368F4540'::geometry) < 0.05::double precision))
(6 rows)

Without the order by and the limit, it looks like a typical query is only returning 5-10 waypoints max. So I probably shouldn't worry about the additional cost of the filter that's applied to the points returned.