---
title: "mirai - Shiny Integration"
vignette: >
  %\VignetteIndexEntry{mirai - Shiny Integration}
  %\VignetteEngine{litedown::vignette}
  %\VignetteEncoding{UTF-8}
---

### Shiny Integration

`mirai` may be used as an asynchronous backend to scale [Shiny](https://shiny.posit.co/) applications.

Depending on the options suppled to `daemons()`, mirai tasks may be distributed across local background processes or multiple networked servers in an efficient and performant manner.

For use with Shiny, `mirai` implements truly event-driven promises, developed in collaboration with Joe Cheng (creator of Shiny).

- Each promise is not polled in a loop for completion (as for other types of promise).
- Instead, promise actions are triggered for execution as soon as each 'mirai' completes (asynchronously).
- Allows for much higher responsiveness (lower latency) and massive scalability (situations with thousand of promises or more).

mirai natively supports Shiny ExtendedTask to create scalable Shiny apps, which remain responsive intra-session for each user, as well as inter-session for multiple concurrent users.

'mirai' may be used anywhere a 'promise' or 'future_promise' would be accepted (with `promises` >= 1.3.0).

#### Shiny ExtendedTask Example: Plot with Clock

In the example below, the app remains responsive, with the clock continuing to tick whilst the simulated expensive computation is running asynchronously in a parallel process. Also the button is disabled and the plot greyed out until the computation is complete.

By wrapping the `runApp()` call in `with(daemons(...), ...)` the daemons are set up for the duration of the app, exiting automatically when the app is stopped.


``` r
library(shiny)
library(bslib)
library(mirai)

ui <- page_fluid(
  p("The time is ", textOutput("current_time", inline = TRUE)),
  hr(),
  numericInput("n", "Sample size (n)", 100),
  numericInput("delay", "Seconds to take for plot", 5),
  input_task_button("btn", "Plot uniform distribution"),
  plotOutput("plot")
)

server <- function(input, output, session) {
  output$current_time <- renderText({
    invalidateLater(1000)
    format(Sys.time(), "%H:%M:%S %p")
  })

  extended_task <- ExtendedTask$new(
    function(...) mirai({Sys.sleep(y); runif(x)}, ...)
  ) |> bind_task_button("btn")

  observeEvent(input$btn, extended_task$invoke(x = input$n, y = input$delay))

  output$plot <- renderPlot(hist(extended_task$result()))

}

app <- shinyApp(ui = ui, server = server)

# run app using 2 local daemons
with(daemons(2), runApp(app))
```
*Thanks to Joe Cheng for providing examples on which the above is based.*

The key components to using ExtendedTask are:

1. In the UI, use `bslib::input_task_button()`. This is a button which is disabled during computation to prevent additional clicks.


``` r
input_task_button("btn", "Plot uniform distribution")
```

2. In the server, create an ExtendedTask object by calling `ExtendedTask$new()` on an anonymous function passing `...` arguments to `mirai()`, and bind it to the button created in (1).


``` r
extended_task <- ExtendedTask$new(
    function(...) mirai({Sys.sleep(y); runif(x)}, ...)
  ) |> bind_task_button("btn")
```

3. In the server, create an observer on the input button, which invokes the ExtendedTask, passing in named arguments to the anonymous function (and hence the mirai) above.


``` r
observeEvent(input$btn, extended_task$invoke(x = input$n, y = input$delay))
```

4. In the server, create a render function for the output, which consumes the result of the ExtendedTask.


``` r
output$plot <- renderPlot(hist(extended_task$result()))
```


#### Shiny ExtendedTask Example: Generative Art

The following app produces pretty spiral patterns.

The user can add multiple plots, making use of Shiny modules, each having a different calculation time.

The plots are generated asynchronously, and it is easy to see the practical limitations of the number of daemons set. For example, if updating 4 plots, and there are only 3 daemons, the 4th plot will not start to be generated until one of the other plots has finished.


``` r
library(shiny)
library(mirai)
library(bslib)
library(ggplot2)
library(aRtsy)

# function definitions

run_task <- function(calc_time) {
  Sys.sleep(calc_time)
  list(
    colors = aRtsy::colorPalette(name = "random", n = 3),
    angle = runif(n = 1, min = - 2 * pi, max = 2 * pi),
    size = 1,
    p = 1
  )
}

plot_result <- function(result) {
  do.call(what = canvas_phyllotaxis, args = result)
}

# modules for individual plots

plotUI <- function(id, calc_time) {
  ns <- NS(id)
  card(
    strong(paste0("Plot (calc time = ", calc_time, " secs)")),
    input_task_button(ns("resample"), "Resample"),
    plotOutput(ns("plot"), height="400px", width="400px")
  )
}

plotServer <- function(id, calc_time) {
  force(id)
  force(calc_time)
  moduleServer(
    id,
    function(input, output, session) {
      extended_task <- ExtendedTask$new(
        function(time, run) mirai(run(time), environment())
      ) |> bind_task_button("resample")

      observeEvent(input$resample,
                   extended_task$invoke(calc_time, run_task))

      output$plot <- renderPlot(plot_result(extended_task$result()))

    }
  )
}

# ui and server

ui <- page_sidebar(fillable = FALSE,
  sidebar = sidebar(
    numericInput("calc_time", "Calculation time (secs)", 5),
    actionButton("add", "Add", class="btn-primary"),
  ),
  layout_column_wrap(id = "results", width = "400px", fillable = FALSE)
)

server <- function(input, output, session) {

  observeEvent(input$add, {
    id <- nanonext::random(4)
    insertUI("#results", where = "beforeEnd", ui = plotUI(id, input$calc_time))
    plotServer(id, input$calc_time)
  })
}

app <- shinyApp(ui, server)

# run app using 3 local daemons
with(daemons(3), runApp(app))
```
*The above example builds on original code by Joe Cheng, Daniel Woodie and William Landau.*

The above uses `environment()` instead of `...` as an alternative and equivalent way of passing variables present in the calling environment to the mirai.

The key components to using this ExtendedTask example are:

1. In the UI, use `bslib::input_task_button()`. This is a button which is disabled during computation to prevent additional clicks.


``` r
input_task_button(ns("resample"), "Resample")
```

2. In the server, create an ExtendedTask object by calling `ExtendedTask$new()` on an anonymous function passing _named_ arguments to `mirai()`, and bind it to the button created in (1). These are passed through to the mirai by the use of `environment()`.


``` r
extended_task <- ExtendedTask$new(
  function(time, run) mirai(run(time), environment())
) |> bind_task_button("resample")
```

3. In the server, create an observer on the input button, which invokes the ExtendedTask, supplying the arguments to the anonymous function above.


``` r
observeEvent(input$resample, extended_task$invoke(calc_time, run_task))
```

4. In the server, create a render function for the output, which consumes the result of the ExtendedTask.


``` r
output$plot <- renderPlot(plot_result(extended_task$result()))
```

#### Advanced Promises Example: Coin Flips

The below example demonstrates how to integrate a `mirai_map()` operation into a Shiny app.

By specifying the '.promise' argument, this registers a promise action against each mapped operation. These can then be used to update reactive values or otherwise interact with the Shiny app.


``` r
library(shiny)
library(mirai)

flip_coin <- function(...) {
  Sys.sleep(0.1)
  rbinom(n = 1, size = 1, prob = 0.501)
}

ui <- fluidPage(
  div("Is the coin fair?"),
  actionButton("task", "Flip 1000 coins"),
  textOutput("status"),
  textOutput("outcomes")
)

server <- function(input, output, session) {

  # Keep running totals of heads, tails, and task errors
  flips <- reactiveValues(heads = 0, tails = 0, flips = 0)

  # Button to submit a batch of coin flips
  observeEvent(input$task, {
    flips$flips <- flips$flips + 1000
    m <- mirai_map(1:1000, flip_coin, .promise = \(x)
      if (x) flips$heads <- flips$heads + 1 else flips$tails <- flips$tails + 1)
  })

  # Print time and task status
  output$status <- renderText({
    input$task
    invalidateLater(millis = 1000)
    time <- format(Sys.time(), "%H:%M:%S")
    sprintf("%s %s flips submitted", time, flips$flips)
  })

  # Print number of heads and tails
  output$outcomes <- renderText(
    sprintf("%s heads %s tails", flips$heads, flips$tails)
  )

}

app <- shinyApp(ui = ui, server = server)

# run app using 8 local non-dispatcher daemons (tasks are the same length)
with(daemons(8, dispatcher = "none"), {
  # pre-load flip_coin function on all daemons for efficiency
  everywhere({}, flip_coin = flip_coin)
  runApp(app)
})
```

*This is an adaptation of an original example provided by Will Landau for use of `crew` with Shiny. Please see <https://wlandau.github.io/crew/articles/shiny.html>.*