[mapserver-dev] Proposal/request to implement FlatGeobuf as built-in format

[Even Rouault]

I'm 0 on this. While this is always nice to see a perf improvement, I'm 
somewhat concerned by the duplication of code between MapServer and GDAL.

And from a purely technical point of view, one of my worry is that 
symbol clashes might occur between GDAL and MapServer on the common code 
they share in the FlatGeobuf and flatbuffers C++ namespaces. If the 
implementations are not kept in sync (and that will necessarily happen 
at some point because we cannot control which GDAL version is used by 
MapServer), weird things (crashes) could possibly happen at runtime (or 
at build time if doing static builds). That worry could be solved by 
possibly having a mapserver:: prefix in front of those namespace names 
(can be done with some #define trickery. See 
https://github.com/OSGeo/PROJ/blob/9f89c7288e44c64ba234a299b1d5528a54d9d61e/include/proj/util.hpp#L104 
for potential inspiration)

The interesting take-away from my perspective is that the OGR overhead 
for FlatGeobuf is only 5 ms, but 13 ms for Shapefile. So one 
interpretation would be that there's some inefficiency in the OGR 
Shapefile driver vs the native MapServer provider.

Even

Le 17/05/2022 à 15:19, Jeff McKenna a écrit :
> Update: Björn has completed the effort (which now includes several 
> msautotests) and we've created an RFC (137) for the native FlatGeobuf 
> support: https://mapserver.org/development/rfc/ms-rfc-137.html
>
> I therefore motion to include FlatGeobuf as a native MapServer driver, 
> per RFC 137, and maintained by Björn Harrtell.
>
> starting with my +1
>
> Some other interesting updates:
>
> - great to see changes here made through testing benefiting so many 
> other projects
>
> - here is an updated benchmark result (that is mentioned in the RFC) :
>
>
>   FlatGeobuf (native) 0.008s
>   Shapefile (native)  0.010s
>   FlatGeobuf (OGR)    0.013s
>   Shapefile (OGR)     0.023s
>   GeoPackage (OGR)    0.042s
>   SpatiaLite (OGR)    0.045s
>   PostGIS (native)    0.053s
>   GeoJSON (OGR)       0.089s
>
> (that was with yesterday's main, GDAL 3.4.3, with MS4W on Windows)
>
> - with all this testing done, and the existing FlatGeobuf 
> documentation effort, it will be very easy to add this to the main 
> documentation now...
>
> Thanks,
>
>
> -Björn & Jeff
>
>
>
>
>
> On 2022-04-27 3:07 p.m., Jeff McKenna wrote:
>> Great point Even, here is an updated result:
>>
>>    Shapefile       0.011s
>>    FlatGeobuf      0.014s
>>    Shapefile (OGR) 0.024s
>>    GeoPackage      0.042s
>>    SpatiaLite      0.045s
>>    PostGIS         0.053s
>>    GeoJSON         0.089s
>>
>>
>> -jeff
>>
>>
>>
>>
>> On 2022-04-27 2:09 p.m., Even Rouault wrote:
>>> Jeff,
>>>
>>> as a data point, perhaps you could enhance your bench with 
>>> shapefiles through OGR ? It would be interesting to have a sense of 
>>> the OGR overhead (that will be the same for any OGR supported 
>>> datasource) to have an idea if it is really worth the effort to have 
>>> a native FlatGeobuf provider for MapServer w.r.t going through OGR 
>>> (it is hard to beat shapefile, and the current difference between 
>>> shapefile and flatgeobuf is small)
>>>
>>> Even
>>>
>>> Le 26/04/2022 à 12:57, Jeff McKenna a écrit :
>>>> (to followup from our chat on IRC yesterday)
>>>>
>>>> To clarify: this would be a new native format in MapServer, such as 
>>>> Shapefile or PostGIS, and the goal would be to not connect to 
>>>> FlatGeobuf through OGR, but instead directly, such as:
>>>>
>>>>   DATA countries.fgb
>>>>
>>>> instead of:
>>>>
>>>>   CONNECTIONTYPE OGR
>>>>   CONNECTION "countries.fgb"
>>>>   DATA "countries"
>>>>
>>>> This should improve the performance even more.  And it will be easy 
>>>> to update the documentation, as the existing files just need to be 
>>>> updated (see the new page at 
>>>> https://mapserver.org/input/vector/flatgeobuf.html )
>>>>
>>>> Great plan Björn!  I look forward to your Pull Request, ha, and 
>>>> will tackle the documentation changes for you.
>>>>
>>>> This is also great news for MapServer, and will give us some nice 
>>>> publicity.
>>>>
>>>> Thanks so much!
>>>>
>>>> -jeff
>>>>
>>>>
>>>>
>>>> On 2022-04-25 5:57 p.m., Björn Harrtell wrote:
>>>>> Hi mapserver devs!
>>>>>
>>>>> I got interested in the subject because of the tests made recently 
>>>>> by Jeff that shows the potential of the format. I believe it 
>>>>> should be possible to get significant additional performance out 
>>>>> of FlatGeobuf in MapServer if it was built in just like Shapefile 
>>>>> support is.
>>>>>
>>>>> FlatGeobuf C++ implementation is rather small, doesn't require any 
>>>>> dependencies, and I don't expect it to require any invasive 
>>>>> changes to existing code.
>>>>>
>>>>> Would love to get started on this ASAP, if there are no 
>>>>> objections. :)
>>>>>
>>>>> Regards,
>>>>> Björn Harrtell
>>>>>
>> ______
> _______________________________________________
> MapServer-dev mailing list
> MapServer-dev at lists.osgeo.org
> https://lists.osgeo.org/mailman/listinfo/mapserver-dev

-- 
http://www.spatialys.com
My software is free, but my time generally not.