Is it possible to stop executing of R code inside shiny (without stopping the shiny process)?

Question

Let's say I have a shiny app that has a function that can take a long time to run. Is it possible to have a "stop" button that tells R to stop the long-running call, without having to stop the app?

Example of what I mean:

analyze <- function() {
  lapply(1:5, function(x) { cat(x); Sys.sleep(1) })
}

runApp(shinyApp(
  ui = fluidPage(
    actionButton("analyze", "Analyze", class = "btn-primary"),
    actionButton("stop", "Stop")
  ),
  server = function(input, output, session) {
    observeEvent(input$analyze, {
      analyze()
    })
    observeEvent(input$stop, {
      # stop the slow analyze() function
    })
  }
))

edit: x-post from shiny-discuss

Solution 1

So another answer, outside a loop : use a child process.

library(shiny)
library(parallel)

#
# reactive variables
# 
rVal <- reactiveValues()
rVal$process <- NULL
rVal$msg <- NULL
rVal$obs <- NULL
counter <- 0
results <- list()
dfEmpty <- data.frame(results = numeric(0))


#
# Long computation
#
analyze <- function() {
  out <- lapply(1:5, function(x) {
    Sys.sleep(1)
    rnorm(1)
})
  data.frame(results = unlist(out))
}

#
# Shiny app
#
shinyApp(
  ui = fluidPage(
    column(6,
      wellPanel(
        tags$label("Press start and wait 5 seconds for the process to finish"),
        actionButton("start", "Start", class = "btn-primary"),
        actionButton("stop", "Stop", class = "btn-danger"),
        textOutput('msg'),
        tableOutput('result')
        )
      ),
    column(6,
      wellPanel(
        sliderInput(
          "inputTest",
          "Shiny is responsive during computation",
          min = 10,
          max = 100,
          value = 40
          ),
        plotOutput("testPlot")
        ))),
  server = function(input, output, session)
  {
    #
    # Add something to play with during waiting
    #
    output$testPlot <- renderPlot({
      plot(rnorm(input$inputTest))
    })

    #
    # Render messages
    #
    output$msg <- renderText({
      rVal$msg
    })

    #
    # Render results
    #
    output$result <- renderTable({
      print(rVal$result)
      rVal$result
    })

    #
    # Start the process
    #
    observeEvent(input$start, {
      if (!is.null(rVal$process))
        return()
      rVal$result <- dfEmpty
      rVal$process <- mcparallel({
        analyze()
      })

      rVal$msg <- sprintf("%1$s started", rVal$process$pid)

    })


    #
    # Stop the process
    #
    observeEvent(input$stop, {
      rVal$result <- dfEmpty
      if (!is.null(rVal$process)) {
        tools::pskill(rVal$process$pid)
        rVal$msg <- sprintf("%1$s killed", rVal$process$pid)
        rVal$process <- NULL

        if (!is.null(rVal$obs)) {
          rVal$obs$destroy()
        }
      }
    })

    #
    # Handle process event
    #
    observeEvent(rVal$process, {
      rVal$obs <- observe({
        invalidateLater(500, session)
        isolate({
        result <- mccollect(rVal$process, wait = FALSE)
        if (!is.null(result)) {
          rVal$result <- result
          rVal$obs$destroy()
          rVal$process <- NULL
        }
      })
      })
    })
  }
  )

edit

See also :

• shiny-discuss : child process
• asynchronous-command-dispatch-in-interactive-r

Solution 2

Provided you can split the heavy duty calculations into several parts, or have access to the part of the code that is involved in the computation, you can insert a breaker part. I implemented this in a Shiny app that listens for a button press before continuing with the rest of the calculation. You can run the app from R by

library(shiny)
runGitHub("romunov/shinyapps", subdir = "breaker")

or copy/paste the code into a server.R and ui.R and run it using runApp().

#ui.R
library(shiny)

shinyUI(fluidPage(

  titlePanel("Interrupting calculation"),

  sidebarLayout(
    sidebarPanel(
      sliderInput(inputId = "num.rows", 
                  label = "Generate number of rows",
                  min = 1e1,
                  max = 1e7,
                  value = 3e3),
      actionButton(inputId = "ok", label = "Stop computation")
    ),
    mainPanel(
      verbatimTextOutput("result")
    )
  )
))

#server.R
library(shiny)

shinyServer(function(input, output) {
  initial.ok <- 0

  part1 <- reactive({
    nr.f <- floor(input$num.rows/2)
    out1 <- data.frame(col = sample(letters[1:5], size = nr.f, 
                                    replace = TRUE), 
                       val = runif(nr.f))
    out1
  })

  part2 <- reactive({

    nr.c <- ceiling(input$num.rows/2)
    out2 <- data.frame(col = sample(letters[1:5], size = nr.c, 
                                    replace = TRUE),
                       val = runif(nr.c))
    out2
  })

  output$result <- renderPrint({

    out1 <- part1()

    if (initial.ok < input$ok) {
      initial.ok <<- initial.ok + 1
      stop("Interrupted")
    }

    out2 <- part2()
    out <- rbind(out1, out2)

    print("Successful calculation")
    print(str(out))
  })
})

Solution 3

What about httpuv::service() ?

library(shiny)
analyze <- function(session=shiny::getDefaultReactiveDomain()){
  continue = TRUE
  lapply(1:100, function(x) {
    if(continue){
      print(x)
      Sys.sleep(1)
      # reload inputs
      httpuv:::service()
      continue <<- !isTRUE(session$input$stopThis)
    }
  }
  )
}

shinyApp(
  ui = fluidPage(
    actionButton("start","Start",class="btn-primary", onclick="Shiny.onInputChange('stopThis',false)"),
    actionButton("stop","Stop",class="btn-danger", onclick="Shiny.onInputChange('stopThis',true)")
  ),
  server = function(input, output, session) {
    observeEvent(input$start, {
      analyze()
    })
  }
)

Solution 4

maybe also not exactly what you are looking for, but could do the trick (at least on mighty Linux). For me it works the way I want since I use bash scripts that are triggered by R shiny and I want to be able to abort them. So how about putting your R code in a script and trigger the script by the system command?

In the example below I just use a simple dummy bash script that runs a sleep command, while the first CL argument is the amount of sleep. Everything below 10 secs is not accepted and puts the exit status to 1. In addition, I get some output in a logfile that I can monitor, and thus the progress in realtime.

Hope you find this helpful.

library(shiny)

ui <- fluidPage(

# we need this to send costumized messages
tags$head(tags$script(HTML('Shiny.addCustomMessageHandler("jsCode",function(message) {eval(message.value);});'))),

# Sidebar with a slider input for number of bins 
sidebarLayout(
sidebarPanel(

    textInput("duration", "How long you want to wait?"),hr(),
    p("Are you experienced?"),
    actionButton("processbtn", "Yes"),hr(),
    p("Show me what's going on"),
    actionButton("logbtn", "Show me by clicking here."),hr(),
    p("Tired of being experienced?"),
    actionButton("abortbtn", "Yes")

    ), # close sidebar panel 

  # Show a plot of the generated distribution
  mainPanel(
     textOutput("outText"),hr(),
     verbatimTextOutput("outLog")
  ) # close mainpanel
 ) # close sidebar
) # close fluidpage

#------SERVER------------

# Define server logic required to draw a histogram
server <- function(input, output, session) {

# our reactive values that change on button click by the observe functions below
values <- reactiveValues(process = 0, abort = 0, log = 0)

observeEvent(input$processbtn, {
  values$process = 1
  values$abort = 0
  values$log = 0
})

observeEvent(input$abortbtn, {
  values$process = 0
  values$abort = 1
})

observeEvent(input$logbtn, {
   values$log = 1
})

current_state = function(exitfile) {
# get the pid
pid = as.integer(system2("ps", args = "-ef | grep \"bash ~/dummy_script.sh\" | grep -v grep | awk '{print $2}'", stdout = TRUE))
print(pid)

if (length(pid) > 0)
 return("RUNNING")

if (file.exists(exitfile))
 return("TERMINATED")

return("NOT_STARTED")
} 

start_function = function(exitfile) {
 if(input$duration == "") {
  end_message="The text input field is empty!"
  js_string <- 'alert("SUCCESS");'
  js_string <- sub("SUCCESS",end_message,js_string)
  session$sendCustomMessage(type='jsCode', list(value = js_string)) 
  values$process = 0
  return("NOT_STARTED")

 } else { # all checks are fine. send a message and start processing
    end_message="We start waiting, yeah!!!"
   js_string <- 'alert("SUCCESS");'
   js_string <- sub("SUCCESS",end_message,js_string)
   session$sendCustomMessage(type='jsCode', list(value = js_string))  

 # here we execute the outsourced script and
 # write the exit status to a file, so we can check for that and give an error message
 system(paste("( bash ~/dummy_script.sh", input$duration,"; echo $? >", exitfile, ")"), wait = FALSE)
 return("RUNNING")
 }  
}

on_terminated = function(exitfile) {
  # get the exit state of the script
  status = readLines(exitfile)
  print(status)
  # we want to remove the exit file for the next run
  unlink(exitfile, force = TRUE)

  # message when we finished
  if ( status != 0 ){
    end_message="Duration is too short."
    js_string <- 'alert("SUCCESS");'
    js_string <- sub("SUCCESS",end_message,js_string)
    session$sendCustomMessage(type='jsCode', list(value = js_string))
  }
  else {
    end_message="Success"
    js_string <- 'alert("SUCCESS");'
    js_string <- sub("SUCCESS",end_message,js_string)
    session$sendCustomMessage(type='jsCode', list(value = js_string))
  }
  values$process = 0
}

# our main processing fucntion
output$outText = renderText({
   # trigger processing when action button clicked
   if(values$process) {

    # get the homefolder
     homedir=Sys.getenv("HOME")

     # create the path for an exit file (we'll need to evaluate the end of the script)
     exitfile=file.path(homedir, "dummy_exit")
     print(exitfile)

     state = current_state(exitfile) # Can be NOT_STARTED, RUNNING, COMPLETED
     print(state)
     if (state == "NOT_STARTED")
        state = start_function(exitfile)

     if (state == "RUNNING")
        invalidateLater(2000, session = getDefaultReactiveDomain())

     if (state == "TERMINATED")
        on_terminated(exitfile)



   # Abort processing
   } else
   if(values$abort) {
      pid = as.integer(system2("ps", args = "-ef | grep \"bash ~/dummy_script.sh\" | grep -v grep | awk '{print $2}'", stdout = TRUE))
    print(pid)
    system(paste("kill", pid), wait = FALSE)
   }

 }) # close renderText function 

 output$outLog = renderText({

 if(values$log) {

   homedir=Sys.getenv("HOME")
   logfile=file.path(homedir, "/dummy_log")

 if(file.exists(logfile)){
   invalidateLater(2000)
   paste(readLines(logfile), collapse = "\n")
 }
 else {
   print("Nothing going on here")
 }
}

})


} # close server

# Run the application 
shinyApp(ui = ui, server = server)

View more solutions

Comments

[DeanAttali]

Sure, but I mean if you're calling a function where a single expression takes minutes, I'm not sure how you'd make it stop without terminating the session. And if you're calling some function from a different package that takes a while, you could copy-n-paste that function's code and add these checks within the code to achieve this, but there's no really native R way to do this

[ismirsehregal]

Here a related post using library(callr) can be found.

[DeanAttali]

Thank you, but the problem with this solution is that it can only stop between iterations of something. I want to be able to call a function that takes a long time, which I don't have access to so I can't enter "breakpoints" inside it, and be able to just stop "ok, nevermind, stop that function call!"

[fxi]

Yeah. I see. I just realised that I have exactly the same problem.

[DeanAttali]

I can't run that code as-is because mcparallel is not defined (maybe I need a newer version of the parallel package? or is it from a different package?). But I do see what you're doing, and yes I think that would work. It's not the prettiest solution but it's good you posted this here so that if someone needs to do this, they'll know of a way. Thanks!

[fxi]

With R being single threaded, there is no other way, for now. I think. Are you on Windows ? This will not run on that platform: see parallel doc. You could ask Shiny team for a reactiveChildProcess(). Haha.

[DeanAttali]

Yep, on Windows. As are many (most?) of Shiny users. This isn't a big deal, I don't think it'll get priority, I'm not pressed to file an issue about it... but it's good this solution is now out there

[Martin Smith]

This stopped working in 2018: see github.com/rstudio/httpuv/issues/148 and the alternative solution at gist.github.com/jcheng5/1ff1efbc539542ecedde92f25458a872