Multiple shiny-servers returning same process ID?

[mmuurr]

I've set up a very basic load-balancing test with Nginx and multiple Shiny-Server 'server' instances, but it seems all traffic gets routed to only a single process, despite multiple servers.

First, the Shiny-Server setup (conf) is as follows:

run_as shiny; 

server {
  listen 9001;

  location /users {
    run_as :HOME_USER:;
    user_dirs;
    directory_index on;
  }

  location / {
    site_dir /srv/shiny-server;
    log_dir /var/log/shiny-server;
    directory_index on;
  }
}

server {
  listen 9002;

  location /users {
    run_as :HOME_USER:;
    user_dirs;
    directory_index on;
  }

  location / {
    site_dir /srv/shiny-server;
    log_dir /var/log/shiny-server;
    directory_index on;
  }
}

The (very) simple Shiny app has a server that looks like so:

library(shiny)
server_pid <- Sys.getpid()
server_id <- uuid::UUIDgenerate()

server <- function(input, output, session) {
  output$debug <- renderPrint({
    print(sprintf("pid: %s, id: %s", server_pid, server_id))
  })
}

And the UI simply displays those two values.

Now, regardless of hitting that test application on localhost:9001 or localhost:9002, I see the same IDs for both connections. To verify this, I indeed only see a single R process in the background, despite having two simultaneous connections to both ports.

My thought had been that each server block in the Shiny Server config file specifies a new 'source' for firing up R processes for each app, but apparently regardless of the server instance, all traffic for a specific app goes to the same R server process ... can anyone else verify this? And if so, is there a way to actually load balance such that multiple connections to the same app are spread across multiple R server processes?

Also (to answer the inevitable question), yes: I realize already that having distinct R server processes for the same app means no shared server state, and that's all right for my application. The goal is to handle many possible connections and with R being single-threaded I want to spread the load across multiple R server processes.