Search code examples
rshinydrake-r-packagersconnect

How to deploy shiny app to shinyapps.io from drake plan

This is a follow-on question from closing the loop on passing the app and data to a Shiny deployment function:

How to use shiny app as a target in drake

I would like to deploy a Shiny app directly from a drake plan as below.

library(drake)
library(shiny)

plan <- drake_plan(
  cars_data = mtcars,
  deployment = custom_deployment_function(file_in("app.R"), cars_data)
)

custom_shiny_deployment <- function(file, data_input) {

    rsconnect::deployApp(
    appFiles = file,
    appName = "cars",
    forceUpdate = TRUE
  )
}

Saved as "app.R" in project working directory (the app works outside of drake if data_input <- mtcars) :

shinyApp(
  ui = fluidPage(
    selectInput("variable", "Variable:",
                c("Cylinders" = "cyl",
                  "Transmission" = "am")),
    tableOutput("data")
  ),
  server = function(input, output) {
    output$data <- renderTable({
      data_input[, c("mpg", input$variable), drop = FALSE]
    }, rownames = TRUE)
  }
)

Logs from shinyapps.io shows "cars_data" from drake is not being passed to the Shiny server "data_input" in the custom_deployment_function environment:

2020-07-14T20:12:57.418989+00:00 shinyapps[2569696]: htmltools version: 0.5.0
2020-07-14T20:12:57.419186+00:00 shinyapps[2569696]: Using pandoc: /opt/connect/ext/pandoc2
2020-07-14T20:12:57.589960+00:00 shinyapps[2569696]: Using jsonlite for JSON processing
2020-07-14T20:12:57.593223+00:00 shinyapps[2569696]: 
2020-07-14T20:12:57.593224+00:00 shinyapps[2569696]: Starting R with process ID: '25'
2020-07-14T20:12:57.628389+00:00 shinyapps[2569696]: 
2020-07-14T20:12:57.628390+00:00 shinyapps[2569696]: Listening on http://127.0.0.1:38608
2020-07-14T20:13:14.686424+00:00 shinyapps[2569696]: Running on host: e89f96d02b6e
2020-07-14T20:13:14.691423+00:00 shinyapps[2569696]: Server version: 1.8.4-11
2020-07-14T20:13:14.691436+00:00 shinyapps[2569696]: LANG: en_US.UTF-8
2020-07-14T20:13:14.691436+00:00 shinyapps[2569696]: R version: 4.0.0
2020-07-14T20:13:14.691437+00:00 shinyapps[2569696]: shiny version: 1.5.0
2020-07-14T20:13:14.691467+00:00 shinyapps[2569696]: rmarkdown version: (none)
2020-07-14T20:13:14.691648+00:00 shinyapps[2569696]: Using pandoc: /opt/connect/ext/pandoc2
2020-07-14T20:13:14.691475+00:00 shinyapps[2569696]: knitr version: (none)
2020-07-14T20:13:14.857243+00:00 shinyapps[2569696]: Using jsonlite for JSON processing
2020-07-14T20:13:14.691443+00:00 shinyapps[2569696]: httpuv version: 1.5.4
2020-07-14T20:13:14.895595+00:00 shinyapps[2569696]: 
2020-07-14T20:13:14.691480+00:00 shinyapps[2569696]: jsonlite version: 1.7.0
2020-07-14T20:13:14.860404+00:00 shinyapps[2569696]: 
2020-07-14T20:13:14.860404+00:00 shinyapps[2569696]: Starting R with process ID: '41'
2020-07-14T20:13:14.895596+00:00 shinyapps[2569696]: Listening on http://127.0.0.1:42022
2020-07-14T20:13:14.691496+00:00 shinyapps[2569696]: RJSONIO version: (none)
2020-07-14T20:13:14.691506+00:00 shinyapps[2569696]: htmltools version: 0.5.0
2020-07-14T20:15:57.441648+00:00 shinyapps[2569696]: Running on host: a2e664ad9837
2020-07-14T20:15:57.447236+00:00 shinyapps[2569696]: Server version: 1.8.4-11
2020-07-14T20:15:57.447326+00:00 shinyapps[2569696]: R version: 4.0.0
2020-07-14T20:15:57.447337+00:00 shinyapps[2569696]: knitr version: (none)
2020-07-14T20:15:57.447337+00:00 shinyapps[2569696]: rmarkdown version: (none)
2020-07-14T20:15:57.447327+00:00 shinyapps[2569696]: shiny version: 1.5.0
2020-07-14T20:15:57.447260+00:00 shinyapps[2569696]: LANG: en_US.UTF-8
2020-07-14T20:15:57.447337+00:00 shinyapps[2569696]: httpuv version: 1.5.4
2020-07-14T20:15:57.447338+00:00 shinyapps[2569696]: jsonlite version: 1.7.0
2020-07-14T20:15:57.447338+00:00 shinyapps[2569696]: RJSONIO version: (none)
2020-07-14T20:15:57.667089+00:00 shinyapps[2569696]: 
2020-07-14T20:15:57.447494+00:00 shinyapps[2569696]: Using pandoc: /opt/connect/ext/pandoc2
2020-07-14T20:15:57.630039+00:00 shinyapps[2569696]: Starting R with process ID: '24'
2020-07-14T20:15:57.626886+00:00 shinyapps[2569696]: Using jsonlite for JSON processing
2020-07-14T20:15:57.447347+00:00 shinyapps[2569696]: htmltools version: 0.5.0
2020-07-14T20:15:57.630037+00:00 shinyapps[2569696]: 
2020-07-14T20:15:57.667090+00:00 shinyapps[2569696]: Listening on http://127.0.0.1:40107
2020-07-14T20:16:14.758082+00:00 shinyapps[2569696]: Warning: Error in renderTable: object 'data_input' not found
2020-07-14T20:16:14.772348+00:00 shinyapps[2569696]:   108: renderTable [/srv/connect/apps/cars/app.R#13]
2020-07-14T20:16:14.772349+00:00 shinyapps[2569696]:   107: func
2020-07-14T20:16:14.772350+00:00 shinyapps[2569696]:    94: origRenderFunc
2020-07-14T20:16:14.772350+00:00 shinyapps[2569696]:    93: output$data
2020-07-14T20:16:14.772350+00:00 shinyapps[2569696]:    13: runApp
2020-07-14T20:16:14.772351+00:00 shinyapps[2569696]:    12: fn
2020-07-14T20:16:14.772351+00:00 shinyapps[2569696]:     7: connect$retry
2020-07-14T20:16:14.772352+00:00 shinyapps[2569696]:     6: eval
2020-07-14T20:16:14.772352+00:00 shinyapps[2569696]:     5: eval

Apologies if this is trivial, but between drake and shiny, it is escaping me.

Solution

  • Now that I see how you are deploying the app, I can say that this is expected behavior. Yes, your custom_shiny_deployment() has access to the data, but the deployed app does not because rsconnect::deployApp() does not ship objects from the calling environment. If you want the data to be available to the app, I recommend saving it (and tracking it with file_in() and file_out()) then passing it to the appFiles argument of deployApp() via custom_shiny_deployment().

    EDIT

    Your app.R can stay like it is.

    app.R is the same as what you wrote.

    library(shiny)

cars_data <- readRDS("cars_data.RDS")

shinyApp(
  ui = fluidPage(
    selectInput("variable", "Variable:",
                c("Cylinders" = "cyl",
                  "Transmission" = "am")),
    tableOutput("data")
  ),
  server = function(input, output) {
    output$data <- renderTable({
      cars_data[, c("mpg", input$variable), drop = FALSE]
    }, rownames = TRUE)
  } 
)

    But from drake's side, you need something like file_in() and file_out() to ensure the correct targets run in the correct order and in response to changes to cars_data.RDS and app.R.

    library(drake)

plan <- drake_plan(
  cars_data = mtcars,
  save_cars = saveRDS(cars_data, file_out("cars_data.RDS")),
  deployment = rsconnect::deployApp(
    appFiles = file_in("app.R", "cars_data.RDS"),
    appName = "cars",
    forceUpdate = TRUE
  )
)

make(plan)

    If you really want appFiles to be NULL, drake still needs file_in() so it knows to depend on physical files.

    plan <- drake_plan(
  cars_data = mtcars,
  save_cars = saveRDS(cars_data, file_out("cars_data.RDS")),
  deployment = deploy_app(file_in("cars_data.RDS", "app.R"))
)

deploy_app <- function(...) {
  rsconnect::deployApp(
    appFiles = NULL,
    appName = "cars",
    forceUpdate = TRUE
  )
}

    Either way, I recommend checking the plan with plot() or vis_drake_graph() before you run it with make(). Both plans above show deployment downstream of the data file (with incoming arrows) which is exactly what we want.

    plot(plan)

    enter image description here

    Without these file markers, there is a chance that drake will deploy the app before the data is ready to go.

    disconnected_plan <- drake_plan(
  cars_data = mtcars,
  saveRDS(cars_data, "cars_data.RDS"),
  deployment =  rsconnect::deployApp(
    appFiles = NULL,
    appName = "cars",
    forceUpdate = TRUE
  )
)

plot(disconnected_plan)

    enter image description here