Performance(?) issues with ST_Reskew() (on rasters)

[Paul Ramsey]

Replicated. I had to go up to 430 to make mine flip behaviour, but it is similar.

P.

> On Feb 24, 2025, at 9:46 AM, Frank Gevaerts <frank.gevaerts at fks.be> wrote:
> 
> Ok, I managed to simplify this to basically the example in the documentation:
> 
> This works fine, in a reasonable (typically 100ms to 150ms) time:
> SELECT ST_Rotation(ST_Reskew(ST_AddBand(ST_MakeEmptyRaster(100, 359, 0, 0, 0.001, -0.001, 0, 0, 4269), '8BUI'::text, 1, 0), 0.0015));
> 
> This does not work fine, I stopped it after more than five minutes:
> SELECT ST_Rotation(ST_Reskew(ST_AddBand(ST_MakeEmptyRaster(100, 360, 0, 0, 0.001, -0.001, 0, 0, 4269), '8BUI'::text, 1, 0), 0.0015));
> 
> The only difference is the size of the raster, the broken one having just one more row. The postgres process was using 100% CPU during that time, I don't think it was doing any IO (so no tempfiles as far as I could see).
> 
> I reproduced it on two systems, and things break for the same numbers, both on debian with postgresql 16, one with postgis 3.5.2/gdal 3.10.2, and one with postgis 3.4.2/gdal 3.2.2
> 
> Frank
> 
> On Mon, Feb 24, 2025 at 07:49:07AM -0800, Paul Ramsey wrote:
>> If you provide a dump of a “good” raster, and a “bad” raster, and the exact SQL you are running, the odds of someone duplicating your issue and then fixing your issue go up 10-fold. I like fixing problems, not trying to replicate problems. 
>> ATB,
>> P
>> 
>>> On Feb 24, 2025, at 7:38 AM, Frank Gevaerts <frank.gevaerts at fks.be> wrote:
>>> 
>>> Hello,
>>> 
>>> I'm having issues with ST_Reskew(). The source raster (32BSI) contains 150000 points, and ST_Reskew() takes forever. If I rescale the raster to anywhere under ~80000 points, it suddenly only takes around 300ms. This is not linear, ~85000 points is bad, and then suddenly 80000 points is good.
>>> 
>>> I'm not actually sure if it's "just" a performance problem or an infinite loop, I have not had the patience yet to have it run for more than 20 minutes.
>>> 
>>> This is using postgis 3.5.2 on debian installed from apt.postgresql.org, and libgdal36 3.10.2
>>> 
>>> Is there anything I'm overlooking, or is there a memory usage parameter I can set to help with this?
>>> 
>>> Frank
>> 
> 
> -- 
> Frank Gevaerts                                 frank.gevaerts at fks.be
> fks bvba - Formal and Knowledge Systems        http://www.fks.be/
> Grote Baan 79                                  Tel:  ++32-(0)11-21 49 11
> B-3511 KURINGEN-HASSELT                        Fax:  ++32-(0)11-22 04 19