This vignette demonstrates how you might use pins and RStudio Connect together to automate ETL for a Shiny app. Pulling expensive data manipulation out of your Shiny app is a great way to improve performance, and pins makes it easy to schedule data updates in a way that automatically flows into your app.
Sharing tidied data
Imagine you’re interested in what’s going on in the news, so you’ve written a little code to scrape the BBC world news feed and turn it into a tidy dataset:
library(xml2)
if (interactive()) {
xml <- read_xml("http://feeds.bbci.co.uk/news/rss.xml")
} else {
# Read a saved version of the data to keep this vignette reproducible
xml <- read_xml("bbc-news.xml")
}
items <- xml %>% xml_find_all("//item")
bbc_news <- tibble::tibble(
title = items %>% xml_find_first("./title") %>% xml_text(),
date = items %>% xml_find_first("./pubDate") %>% xml_text(),
url = items %>% xml_find_first("./guid") %>% xml_text()
)
bbc_news
#> # A tibble: 71 × 3
#> title date url
#> <chr> <chr> <chr>
#> 1 Diana interview: PM concerned after inquiry into BBC deceit Fri,… http…
#> 2 Covid in Scotland: Glasgow to be only Scottish area in level thr… Fri,… http…
#> 3 Prince Harry says heavy drinking masked pain of mum Diana's death Fri,… http…
#> 4 Coronavirus: Spain to lift restrictions for UK and Japanese trav… Fri,… http…
#> 5 Diana interview: Whistleblower wants apology from BBC bosses Fri,… http…
#> 6 Mayhill: Swansea rioters face robust action, police vow Fri,… http…
#> 7 UK to offer Australia tariff-free trade deal despite farmers' fe… Fri,… http…
#> 8 Leonard Blavatnik named UK's richest person with £23bn fortune Fri,… http…
#> 9 Lady Gaga had a 'psychotic break' after sexual assault left her … Fri,… http…
#> 10 Big cats seized from park belonging to Tiger King couple Fri,… http…
#> # … with 61 more rows
Wouldn’t it be nice to spare your colleagues the work of reproducing this data, and provide them directly with the tidy tibble? That’s easy to do with pins:
board_rsc <- board_rsconnect()
board_rsc %>% pin_write(bbc_news)
If you open the pin on RSC (e.g. by running board_rsc %>% pin_browse("news_total")
), you’ll see some metadata about the pin, a preview of the data, and a control panel which allows you to control who can see the data. Assuming you’ve given your colleagues access, they can now easily pull down the data for their own analysis:
board_rsc <- board_rsconnect()
board_rsc %>% pin_read("your_name/bbc_news")
Automating
Here the underlying data will be changing regularly, but the pinned data will only change when you run pin_write()
. Wouldn’t it be great if we could automate that process? One way to do so is with RSC’s scheduled reports.
First, take your code and put it in an Rmd:
---
title: BBC news
---
```{r}
library(pins)
library(xml2)
xml <- read_xml("http://feeds.bbci.co.uk/news/rss.xml")
items <- xml %>% xml_find_all("//item")
bbc_news <- tibble::tibble(
title = items %>% xml_find_first("./title") %>% xml_text(),
date = items %>% xml_find_first("./pubDate") %>% xml_text(),
url = items %>% xml_find_first("./guid") %>% xml_text()
)
board_rsc <- board_rsconnect()
board_rsc %>% pin_write(bbc_news)
```
```{r}
bbc_news
```
Then publish it to RSC, and schedule it to run as often as you like. Assuming that you have RSC 1.9.0 or later you don’t need to provide any arguments to board_rsconnect()
; pins will automatically publish to the same Connect instance that’s running the report.
Shiny apps
If you’re using automatically updated data in a Shiny app, you can use pin_reactive_read()
to create a reactive dependency so that your app will automatically update shortly after the data changes:
library(shiny)
library(pins)
board <- board_rsconnect()
ui <- fluidPage(
titlePanel("News from the BBC"),
htmlOutput("news")
)
server <- function(input, output, session) {
news <- board %>% pin_reactive_read("hadley/bbc_news")
output$news <- renderUI({
title <- htmltools::htmlEscape(news()$title)
links <- paste0("<a href='", news()$url, "'>", title, "</a>")
bullets <- paste0(" <li>", links, "</li>", collapse = "\n")
HTML(paste0("<ul>", bullets, "</ul>"))
})
}
shinyApp(ui, server)
If you deploy this app and watch it while your scheduled reported runs, you’ll see the data update automatically.