Memory leak with reactiveTimer and invalidateLater

[jcheng5]

(Reported by Aaron Berg on behalf of a customer)

If a session ends while a timer (reactiveTimer or invalidateLater) is pending, the memory used by the session will be retained until the timer's next timeout. For apps that have very long timeouts (such as refreshing data on a dashboard app every 24 hours), lots of data, and frequent browser refreshes (or lots of visitors coming and going), this can lead to significant memory leakage.

The examples below demonstrate the problem, one for each type of timer. Reload the app many times in succession and you'll see that the RAM usage climbs with each refresh. If you wait until the 100 second timeouts start expiring, you'll see the RAM usage start to decline.

library("shiny")

ui <- function(req) {
  basicPage(
    "Memory usage is ",
    capture.output(print(pryr::mem_used()))
  )
}

server <- function(input, output, session) {
  session$userData$data <- runif(1000000)
  isolate(invalidateLater(100*1000))
}

shinyApp(ui,server)

library("shiny")

ui <- function(req) {
  basicPage(
    "Memory usage is ",
    capture.output(print(pryr::mem_used()))
  )
}

server <- function(input, output, session) {
  session$userData$data <- runif(1000000)
  reactiveTimer(100*1000)
}

shinyApp(ui,server)