Script using GPU throws error when running app on Shiny server - r

Question

I have written some code in R using the Shiny package. The app runs alright when running shiny::runApp() from RStudio.
I've tried to host the app on a Shiny server and the GUI starts up correctly. However, a plot should appear on the right when the button "Enviar informacion" is clicked. (You can see how the app looks here, but it won't work though because that link is not on Shiny server but on Shinyapps.io).
The relevant part of the logfile is the following:
Listening on http://127.0.0.1:38327
Loading required package: Rcpp
Warning: Error in [: subscript out of bounds
211: FUN [/srv/shiny-server/spike_sorting/server.R#82]
210: apply
209: cluster_som_h [/srv/shiny-server/spike_sorting/server.R#82]
...
So the error is inside the function cluster_som_h. After trying some stuff, I found out what is causing the error, but don't know how to fix it. Inside this function, I make use of the function Rsomoclu.train() from the package Rsomoclu, with kernelType = 1. This means that the function is run on the GPU. If I set kernelType = 0, then the app works okay with both runApp() and from the Shiny server. So that's where the problem is: for some reason, running the function on the GPU is not working if the app is run from the Shiny server, but it does work if it is run from RStudio using runApp().
Maybe there is something going on because I have CUDA installed on my computer but when I try to run it from the server something happens. Is there a way to fix this? I believe running the code from this specific computer no matter where on the LAN the app is run would solve this. Is this possible?

Answer 1

After doing some (basic) research, I found out what was happening. I have CUDA installed only in my user in Ubuntu, so I had to change the first line of the file /etc/shiny-server/shiny-server.conf so that the user was the right one:
# Instruct Shiny Server to run applications as the user "my_name"
run_as my_name;