Re: Memory leak (possibly connected to postgis) leading to server crash

Roman Cervenak <[email protected]> Tue, 17 Dec 2019 14:20:41 +0100
Newsgroups gmane.comp.db.postgresql.bugs
Message-ID <CAGjExY3qtdQnA3XiOok=8yr6tXSAz86nCNcE92u6K-z8FrHZyA@mail.gmail.com>
.NET core is cross platform, you can run it under windows or linux, just
install .net core runtime.
I did not try to run loop with fixed params, I tried to mimic my production
workload (which is not random of course, but highly variable by location).
You can certainly try.

On Tue, Dec 17, 2019, 13:50 Tomas Vondra <[email protected]>
wrote:

> On Tue, Dec 17, 2019 at 12:51:27PM +0100, Roman Cervenak wrote:
> >Hey guys,
> >I have reproducer for you.
> >
> >Luckily, the issue is not specific for our data, and can be demonstrated
> >also on OpenStreepMap data. I have imported OSM planet with Imposm, which
> >creates tables with PostGIS geometry columns.
> >I have used roads for linestring testing and landuse for polygon testing,
> >with queries combining several geometric operations (similar to my
> >production workload).
> >
> >The linesting test with 7 concurrent workers increases memory consumption
> >by about 6 GB per hour on my environment.
> >
> >If you don't have OSM database with geometries, I have dumped the roads
> >table (about 25GB download):
> >
> https://sygicpublic.blob.core.windows.net/postgres-bug-data/osm_road.sql.gzip
> >Here is the tool running queries in parallel workers, written in .NET Core
> >(just fill in database connection info in Program.cs, build and run):
> >https://github.com/rouen-sk/PostgisMemoryLeakDemoCore
> >
> >Let me know if I can help further.
> >
>
> I'll try reproducing it, but I don't have any environment to run c#
> stuff. Is it possible to reproduce a single query reproducing the issue,
> with fixed parameters? Or is the randomization necessary?
>
> regards
>
> --
> Tomas Vondra                  http://www.2ndQuadrant.com
> PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
>