[postgis-users] ST_Intersection very slow.

[John Abraham]

So I've was running this query for 866000 s (10 days) before I decided to kill it:

create table taz_and_lancover_10_fast_2 as
SELECT p.lc_class, n.taz
 , CASE 
   WHEN ST_CoveredBy(p.geom, n.the_geom) 
   THEN p.geom 
   ELSE 
    ST_Multi(
      ST_Intersection(p.geom,n.geom)
      ) END AS geom 
 FROM lancover_polygons_snap AS p 
   INNER JOIN tazjan2 AS n 
    ON (ST_Intersects(p.geom, n.the_geom) 
      AND NOT ST_Touches(p.geom, n.the_geom) );

Explain shows this, it's using the spatial index:

"Nested Loop  (cost=0.00..310492.35 rows=483105 width=9973)"
"  Join Filter: (_st_intersects(p.geom, n.the_geom) AND ((NOT (p.geom && n.the_geom)) OR (NOT _st_touches(p.geom, n.the_geom))))"
"  ->  Seq Scan on tazjan2 n  (cost=0.00..843.09 rows=2709 width=9493)"
"  ->  Index Scan using lancover_polygons_snap_geom_idx on lancover_polygons_snap p  (cost=0.00..21.67 rows=5 width=480)"
"        Index Cond: (geom && n.the_geom)"

There are 2709 rows in tazjan2 and 998031 rows in lancover_polygons_snap, so I appreciate that it's a bit of a large problem.  But MapInfo was able to do it interactively in a few days and Geomedia was also able to do it in about a day.  

Both MapInfo and Geomedia ran out of memory (8GB machines) until the problem was broken into two regions (North and South), but postgresql seems to be chewing on the problem using only 400MB.  The interactive approach in MapInfo was to divide lancover_polygons_snap by lc_class, to further divide each region into about 10 subproblems.  Perhaps subprobleming this is the way to go?  Can't the query subproblem it based in the indices or would I have to do that manually?

One potential thing I've realized is that a few of the geometries in tazjan2 are multipolygons, not single polygons.  But it's only a few.  There are a few very large and complex polygons in lancover_polygons_snap, but again, most of the 998031 are simple small polygons, about half would be ST_CoveredBy the polygons in tazjan2 and most of the rest would only overlap two or three of the polygons in tazjan2.

I must be doing something wrong.  Any hints?

I have
 max_connections set to 100 (currently only 11 connections active)
 work_mem was defaulting to 1MB, I just bumped it to 256MB (machine has 32GB but it's Postgresql9.1 (x86), i.e. 32bit version
 shared_buffers was 32MB, I am trying 512MB, could go higher
 effective_cache_size was defaulting to 128MB, I am trying 20GB.
 random_page_cost was 4.0, this is VMWare virtual server.  I am trying 2.0 

I'm trying an "explain analyze" with just 5 rows of tazjan2, a 5 x 998031 intersection problem with indices shouldn't take that long, should it?  It's been running for a little over an hour.  I'm wondering if the few complex polygons in lancover_polygons_snap are causing the slowness?  Is there some fast way to divide complex polygons, e.g. apply a 1km grid over them?  

PostGISfullversion(): "POSTGIS="2.0.3 r11128" GEOS="3.3.8-CAPI-1.7.8" PROJ="Rel. 4.8.0, 6 March 2012" GDAL="GDAL 1.9.2, released 2012/10/08" LIBXML="2.7.8" LIBJSON="UNKNOWN" (core procs from "2.0.3 r11132" need upgrade) TOPOLOGY (topology procs from "2.0.3 r11132" need upgrade) RASTER (raster procs from "2.0.3 r11132" need upgrade)"

--
John Abraham


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-users/attachments/20150219/d037de4e/attachment.html>