[postgis-users] ST_ConcaveHull performance issue

Trevor Wiens tsw.web at gmail.com
Fri Aug 2 10:52:13 PDT 2019


Thanks to both of you. I tried running with perf now and found most of the
time was spent in libstdc++ doing dynamic casting and related functions for
80% of CPU time. Within geos itself, the top time spent is in the
SimplePoitnInAreaLocator and STRIntersectsOp functions.

I don't know how geos works well enough to say of this is a related issue.

TSW

On Fri, Aug 2, 2019 at 11:40 AM Regina Obe <lr at pcorp.us> wrote:

> There is was an issue with ST_Union as applied to invalid polygons
> introduced in 3.7.1 and 3.6.2 which would make ST_Union run indefinitely.
> It appears to still be an open bug.  This smells like that issue.
>
>
>
> https://trac.osgeo.org/geos/ticket/867
>
>
>
> https://lists.osgeo.org/pipermail/postgis-users/2018-April/042710.html
>
>
>
>
>
>
>
> *From:* postgis-users [mailto:postgis-users-bounces at lists.osgeo.org] *On
> Behalf Of *Trevor Wiens
> *Sent:* Friday, August 2, 2019 1:16 PM
> *To:* PostGIS Users Discussion <postgis-users at lists.osgeo.org>
> *Subject:* Re: [postgis-users] ST_ConcaveHull performance issue
>
>
>
> Thanks for your reply. It was mix of points, lines and polygons, but when
> I broke it down it was the lines causing the problem.
>
>
>
> I tried commenting out the line in the ST_ConcaveHull function you
> suggested but that doesn't make a difference. I suspect there is some
> underlying library that has changed.
>
>
>
> I will open a ticket as you suggest.
>
>
>
> Thanks
>
>
>
> TSW
>
>
>
> On Thu, Aug 1, 2019 at 11:55 PM Darafei "Komяpa" Praliaskouski <
> me at komzpa.net> wrote:
>
> Hi,
>
>
>
> There were changes ST_ConcaveHull that fixed robustness of it in 2.2 ->
> 2.5 chain.
> To point where the penalty comes from, try to run a query and on a side
> console server-side run `sudo perf top`. Function names will get you a
> rough idea where the execution process lives now.
>
>
>
> To further debug, go to blame view in github and try updating
> st_concavehull (it's in SQL) change by change. Most recent adds an union
> here, try commenting it out and hot reloading on your db.
> https://github.com/postgis/postgis/blame/svn-trunk/postgis/postgis.sql.in#L6155
>
> Can you share the data and ticket this on http://trac.osgeo.org/postgis/?
>
> What is the data structurally? Are these 6000 objects points, or polygons?
>
>
>
>
>
> On Fri, Aug 2, 2019 at 3:43 AM Trevor Wiens <tsw.web at gmail.com> wrote:
>
> I am having difficulty determining why I'm seeing significant differences
> in performance between two database configurations with the same data.
>
>
>
> One one machine (centos) I have the following software:
>
> geos 3.5.0
>
> sfcgal 1.3.1
>
> cgal 4.7.1
>
> postgis 2.2
>
> postgresql 9.4
>
>
>
> On a second machine (debian 10) I have the following:
>
> geos 3.7.1
>
> sfcgal 1.3.6
>
> cgal 4.13
>
> postgis 2.5.2
>
> postgresql 9.6
>
>
>
> In terms of hardware there is no significant difference, if anything the
> second machine is more capable, but that is not reflected in my performance
> results.
>
>
>
> On the first machine when I run a ST_ConcaveHull with about 6000 features,
> I get result a second or two. On the second machine, it won't finish within
> 30 minutes. Both are using geos as the postgis.backend. I don't understand
> why the one is so much faster than the other with the identical data and
> query.
>
>
>
> Any suggestions as to what to what the cause may be or how I might
> diagnose the cause?
>
>
>
> Any clues would be greatly appreciated.
>
>
>
> TSW
>
> _______________________________________________
> postgis-users mailing list
> postgis-users at lists.osgeo.org
> https://lists.osgeo.org/mailman/listinfo/postgis-users
>
>
>
>
> --
>
> Darafei Praliaskouski
>
> Support me: http://patreon.com/komzpa
>
> _______________________________________________
> postgis-users mailing list
> postgis-users at lists.osgeo.org
> https://lists.osgeo.org/mailman/listinfo/postgis-users
>
>
>
> --
>
> Trevor Wiens
> Apropos Information Systems
> aproposinfosystems.com
> Calgary, Alberta
>
> Ph. 403-973-5901
> _______________________________________________
> postgis-users mailing list
> postgis-users at lists.osgeo.org
> https://lists.osgeo.org/mailman/listinfo/postgis-users



-- 
Trevor Wiens
Apropos Information Systems
aproposinfosystems.com
Calgary, Alberta
Ph. 403-973-5901
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-users/attachments/20190802/594f4cc5/attachment.html>


More information about the postgis-users mailing list