A simple workflow for async `{shiny}` with `{mirai}`

[This article was first published on R | Discindo, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

In my previous post, I developed a shiny module to encapsulate the logic of sending and monitoring background async tasks. The main advantage of this approach was to simplify making repeated async calls in larger applications. In the first version of this module, the async process was created with callr:r_bg, an approach that my self and others have used before.

However, there is one, potentially significant, drawback of using callr in such a way. Take this hypotetical scenario as an example. You have a shiny app with five async tasks triggered in response to a user changing a dataset. You test it locally, and everything works great. Then you deploy and share with the world. Ten of your followers click on the link more-or-less at the same time and visit the application, each choosing one of three datasets available in your data science app. The app’s server, featuring async functions gets to work, and initializes 5 (tasks) * 10 (users) = 50 callr::r_bg calls, each running in a separate child R process. Some of these copy nothing the child enviroment, some only a few small objects, but others a large data object needed for the async function to transform or run a model. It should be no surprise if the app is no longer that fast. The hosting server, even with a fast, multi-thread processor, still hast to contend with many R processes and the shiny session is also getting a bit bogged down, as it has potentially dozens of observers monitoring background processes. Clearly, we need to rethink our approach.

Wouldn’t it be great if we had a way to limit the total number of concurrent child R processes that our shiny session would spawn, and have a queue system that would start another background job as soon as one completes? Enter mirai. mirai lets us initialize a set number of R daemons (persistent background processes) that are ready to receive mirai requests and ensures FIFO (first in, first out) scheduling. Using mirai, we can handle a large number of async background jobs elegantly without overburdening the system. If the number of jobs requested by the shiny app exceeds the number of available daemons, mirai would hold the jobs until one of the daemons (threads) frees up and submit on a first-come, first-serve basis. Just great!

So how does it work?

For example setups for shiny, check out the documentation, where you can read about mirai-only solutions, as well as approaches combining mirai with promises.

For my application, I’ll adapt the callr approach I described in my previous post to work with mirai. In fact, there very little to change to make the callr example work with mirai:

  1. Change the async version of our function to use mirai
head_six <- function(x, sleep) {
head_six_async_mirai <- function(x, sleep) {
args <- list(head_six = head_six, x = x, sleep = sleep)
bg_process <- mirai::mirai(.expr = head_six(x, sleep), .args = args)
  1. Change the polling logic in the module’s server to use mirai::unresolved, rather than the is_alive method of the callr process object.
mod_async_srv_mirai <- function(id, fun_async, fun_args, wait_for_event = FALSE, verbose = FALSE) {
moduleServer( id, function(input, output, session){
res_rct <- shiny::reactiveVal(NULL)
poll_rct <- shiny::reactiveVal(TRUE)
if (isTRUE(wait_for_event)) {
bg_job <- reactive({
do.call(fun_async, fun_args)
}) |> bindEvent(poll_rct())
if (verbose) {
message(sprintf("checking: %s", id))
alive <- mirai::unresolved(bg_job())
if (isFALSE(alive)) {
if (verbose) {
message(sprintf("done: %s", id))
start_job = function() poll_rct(TRUE),
get_result = reactive(res_rct())
  1. In the app’s server, or better yet global.R or equivalents, we need to initialize the daemons:
onStop(function() mirai::daemons(0L))

In this setup, our shiny can run up to two parallel async jobs handled by the mirai queue. These daemons are shared across all users of our application, irrespective of the shiny session. This is because mirai’s daemons apply to the entire R session, not individual shiny sessions.


For a running example of mirai async with the module, visit this gist:


In this post I went over an approach to organize mirai background async jobs using a shiny module, in order to make the async code faster to write, less error prone and overall cleaner.

To leave a comment for the author, please follow the link and comment on their blog: R | Discindo.

R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)