[QGIS-Developer] Processing plugin algorithm slows down in graphical modeler

[Nyall Dawson]

On Sun, 25 Aug 2019 at 00:39, Clemens Raffler <clemens.raffler at gmail.com> wrote:
>
> Dear QGIS Devs,
>
> Recently, I received two messages from users concerning the speed of
> processing algorithms which are implemented in the QNEAT3 plugin. More
> specifically, they report that the execution time of algorithms slow
> down up to 10 times when the algorithm is used in the graphical modeler
> compared to manual execution from the processing toolbox (used
> algorithm: OD Matrix from Layers as Lines (m:n)). I link the ticket from
> the QNEAT3 repository here for reference:
> https://github.com/root676/QNEAT3/issues/21
>
> I would like to identify the cause of this problem because on the first
> look I can't relate it to any implementation flaws on the plugin side.
> Has anyone else experienced this problem with other processing plugins
> or core algorithms? I didn't have time yet to check if core algorithms
> are affected too. Could it be some kind of bug related to the graphical
> modeler? Do you have any hints what could cause this problem on the
> plugin side so I can fix it? Tanks in advance!

The only difference I can think of between running them in the toolbox
vs models is that when you run a model it's always run in the main
thread, including all the child algorithms inside that model. When you
run direct from the toolbox then **some** algorithms are flagged as
safe to run in background threads and won't be run on the main thread.

So potentially they are seeing a slowdown due to this, and other stuff
clogging up the main thread and slowing down the execution...

Nyall