Hi all, I'm wondering what the major advantages are to running a stack like:
nginx -> shiny-server -> shiny
vs something like:
nginx -> shiny where the R/Shiny processes are managed by something like SystemD or PM2.
In particular, if those stacks are placed inside a logical compute unit (whether a Docker container or AWS AMI or whatever) and scaling is handled outside that unit (e.g. via Docker Compose or AWS ELB & Auto-Scaling Groups). It seems many of the Shiny Server configuration elements are geared towards managing a single machine serving a suite of applications, but in the 'one machine/container for one application' paradigm, have people had success/failure with bypassing the Shiny Server layer? My major sensitivity is memory consumption (i.e. trying to get closer to the 'atomic' unit required to serve an application) and communication latency (i.e. removing unneeded networking packet hand-offs between layers).