23-performance.Rmd

# Performance

## Restaurant Shiny

- Server =~ Restaurant
- User =~ Customer
- Request =~ Order
- R process =~ Chef

To make each chef more efficient

- Identify any slow steps
- Brainstorm to make them faster

To make the restaurant more efficient

- Hire more chefs
- Buy more restaurants

## Benchmarking

How does the app handle multiple users?

### Tools

- [shinyloadtest](https://rstudio.github.io/shinyloadtest/)
  - Run an app
  - Record how you interact with the app
  - [`shinycannon`](https://rstudio.github.io/shinyloadtest/#shinycannon): rerun the session
    multiple times
  - analyse the
    [log](https://rstudio.github.io/shinyloadtest/articles/analyzing-load-test-logs.html)
    of the reran sessions

```{r, eval = FALSE}
runApp("some_app.R")
shinyloadtest::record_session("http://127.0.0.1:9876")
# See "./recording.log"
```

```bash
shinycannon ./recording.log http://127.0.0.1:9876 \
  --workers 10 \
  --loaded-duration-minutes 5 \
  --output-dir run1
```

```{r, eval = FALSE}
library(shinyloadtest)
df <- load_runs("./run1")
shinyloadtest_report(df, "report.html")
```

## Profiling

Where is the app slow / memory-hungry?

### Tools

- profvis

The call stack and the "Flame graph"

```{r, eval = FALSE}
library(profvis)

f <- function() {
    pause(0.2)
  g()
    h()
    10
}
g <- function() {
    pause(0.1)
  h()
}
h <- function() {
    pause(0.3)
}

profvis::profvis(f())
```

Similar for {shiny}, but need explicit call to start the app (and close it with [Ctrl-C])

```{r, eval = FALSE}
ui <- fluidPage(
    actionButton("x", "Push me"),
      textOutput("y")
    )
server <- function(input, output, session) {
    output$y <- eventReactive(input$x, f())
}

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

```{r, eval = FALSE}
profvis::profvis(
  shiny::runApp(
    app()
  )
)
```

Note that data downloads may not be tracked.

## Optimisation

### Cacheing

- `bindCache`
  - cacheing of any reactive value or render function
  - uses 'cache keys'
    - keep them as simple as possible
  - date/time-dependent results
  - cache has fixed size
  - scope (per session, or per process; in-memory or on-disk)

```{r, eval = FALSE}
r <- reactive(slow_function(input$x, input$y)) %>%
  bindCache(input$x, input$y)

output$text <- renderText(slow_function2(input$z)) %>%
  bindCache(input$z)
```

### Other Optimisations

- Do as little as possible
- Do your prep outside of the server
- Do your prep outside of shiny
- Use the fastest import method available
- Only do the slow stuff if requested
- Learn async programming

## Resources

- [https://rstudio.com/resources/rstudioconf-2019/shiny-in-production-principles-practices-and-tools/]()
- [https://rstudio.github.io/shinyloadtest/articles/analyzing-load-test-logs.html]()
- [https://queue.acm.org/detail.cfm?id=2927301]()
- [https://adv-r.hadley.nz/perf-improve.html]()
- [https://csgillespie.github.io/efficientR/]()
- [https://rstudio.com/resources/rstudioconf-2018/make-shiny-fast-by-doing-as-little-work-as-possible/]()
- [https://shiny.rstudio.com/articles/caching.html]()
- [https://shiny.rstudio.com/app-stories/weather-lookup-caching.html]()


## Meeting Videos

### Cohort 1

`r knitr::include_url("https://www.youtube.com/embed/LLO6vDLqhWw")`

<details>
  <summary> Meeting chat log </summary>
  
```
00:07:09	Russ Hyde:	Hi
00:45:54	Robert Overman:	Are we having a session on shiny server?
01:01:58	docksbox@pm.me:	Thanks!
01:04:51	docksbox@pm.me:	Yes, sure
```
</details>

### Cohort 2

`r knitr::include_url("https://www.youtube.com/embed/UwrBxS39E3g")`


### Cohort 3

`r knitr::include_url("https://www.youtube.com/embed/GFbRR--E8Ow")`

<details>
  <summary>Meeting chat log</summary>
```
00:06:14	Brendan Lam:	What function records interactions with your Shiny app? How do you simulate users interacting with your app?
What is a flame graph and how do you interpret it?
What are the cache key and scope?
What are two ways of optimizing your app
00:09:25	Brendan Lam:	https://rstudio.github.io/promises/
00:14:39	Brendan Lam:	https://github.com/rstudio/shinycannon/issues/69#issuecomment-1322390215
00:22:46	Brendan Lam:	We start with f().
Then f() calls g(),
Then g() calls h().
Then f() calls h().
00:28:03	Brendan Lam:	https://memoise.r-lib.org/
00:38:16	Brendan Lam:	arrow::write_feather()
00:38:23	Brendan Lam:	arrow::read_feather()
00:38:30	Brendan Lam:	qs::qread()/qs::qsave() instead of readRDS()/saveRDS()
00:41:15	Brendan Lam:	https://rstudio.com/resources/rstudioconf-2019/shiny-in-production-principles-practices-and-tools/
https://rstudio.github.io/shinyloadtest/articles/analyzing-load-test-logs.html
https://queue.acm.org/detail.cfm?id=2927301
https://adv-r.hadley.nz/perf-improve.html
https://csgillespie.github.io/efficientR/
https://rstudio.com/resources/rstudioconf-2018/make-shiny-fast-by-doing-as-little-work-as-possible/
https://shiny.rstudio.com/articles/caching.html
https://shiny.rstudio.com/app-stories/weather-lookup-caching.html
00:45:15	Brendan Lam:	https://twitter.com/brendannathanl1
```
</details>


### Cohort 4

`r knitr::include_url("https://www.youtube.com/embed/bxSj2VdA8XI")`

<details>
  <summary>Meeting chat log</summary>
```
00:09:46	Trevin Flickinger:	start
00:15:04	Trevin Flickinger:	https://rstudio.github.io/shinyloadtest/articles/case-study-scaling.html
00:15:54	Trevin Flickinger:	https://posit.co/resources/videos/shiny-in-production-principles-practices-and-tools/
00:22:35	Trevin Flickinger:	https://rstudio.github.io/promises/
00:47:58	Matthew Efoli:	https://www.youtube.com/watch?v=hltOgAC2mC4&list=PLexAKolMzPcriOdeLwoMxQOyHRnMguEv4&index=28
00:50:44	Trevin Flickinger:	stop
```
</details>


### Cohort 5

`r knitr::include_url("https://www.youtube.com/embed/URL")`

<details>
  <summary>Meeting chat log</summary>
```
LOG
```
</details>