Currently I have a shiny web app that can do some calculations on a 3GB data.frame loaded in memory.
Now instead of implementing this feature on shiny web app, I need to make it a restful service that pipelines its calculation to another app in JSON format, so that people can use it by sending http request with url like http://my-app.com/function
I'm trying opencpu right now, but I don't quite understand how can I load and keep the large data in memory so that I can use api of opencpu to call functions in package just to do calculations, not to load the large data from disk every time I send http request.
One workaround might be to use hbase as in memory database and use rhbase to load the data. But before I invest time in learning it I want to know if it is a reasonable choice for 3GB data.frame since it might add more overhead in serialization and other stuff that offset its speed benefit.
What would be a better way to implement this functionality? Solutions using packages other than opencpu are also welcome, and it's better to be free.
You may want to look at Plumber. You decorate your R functions with comment code (that can include you loading data) and it makes it available via a REST API.